Design of a Six-Component External Wind Tunnel

Balance

Manuel Amaro Matoso Aguiar Ferreira

Thesis to obtain the Master of Science Degree in

Mechanical Engineering

Supervisors: Prof. Luis Rego da Cunha de Eça

Prof. Luis Alberto Gonçalves de Sousa

Examination Committee
Chairperson: Prof. Viriato Sérgio de Almeida Semião
Supervisor: Prof. Luis Rego da Cunha de Eça
Members of the Committee: Prof. Joao Manuel Melo de Sousa
Prof. Moisés Simões Piedade

July, 2015
II
Agradecimentos

Este projecto contou com o envolvimento de um grupo de pessoas sem as quais certamente não teria sido
possível a sua realização.
Em primeiro lugar, devo uma nota de agradecimento aos meus orientadores, os professores Luís Eça e
Luís Sousa, pela sua exemplar orientação e por acreditarem na minha capacidade de levar avante este projecto
(ou por assim o fazerem entender).
Ao professor Moisés Piedade, expresso a minha gratidão pelas várias sessões de intrução em electrónica,
pelas luminosas sugestões técnicas e por facilitar o fabrico tanto dos protótipos das PCB’s como das versões
finais. Enfim, pelo o seu apoio incondicional.
Ao professor Agostinho Fonseca, que também foi um dos porpulsores deste projecto e que contribuiu
activamente durante a fase conceptual. Gostaria ainda de lhe agradecer pela sua disponibilidade e conversas,
que tantas vezes suscitaram questões relevantes ao desenvolvimento deste protótipo.
Dirigo os meus sinceros agradecimentos ao Projecto FST Novabase. Em especial ao Bruno Santos,
Daniel Pinho e Pedro Costa, pelo seu acompanhamento constante desde o início deste projecto, particularmente
durante a fase de devenvolvimento do sistema de aquisição de dados. A sua contribuição foi inestimável.
Não posso deixar também de agradecer ao Diogo Henriques, pelo equipamento cedido, sugestões
téncnicas e formação em instalação de extensómetros. Bem como ao Rishi Vadher, pelo apoio prestado na
operação do equipamento do FabLab, e aos técnicos do Laboratório de Técnicas Oficinais do DEM-IST, o Srs.
Pedro e João.
Finalmente, porque a gratidão tem memória curta, o meu profundo e sentido agradecimento à com-
panhia dos lanches das 17h00 e a todos os meus restantes amigos, por isto e por aquilo.

III
IV
Resumo

Hoje em dia, mais de 70% dos ensaios de corpos fuselados e não fuselados em túnel vento exigem a mediação
das cargas aerodinâmicas que se fazem sentir. As balanças utlizadas são geralmente instrumentos de precisão
altamente especializados, e portanto, caros. Neste contexto, o objectivo deste projecto é desenvolver uma
balança externa de seis componentes, apresentando-se como uma alternativa simples e económica às soluções
comerciais existentes. Sendo um protótipo, foi desenvolvido visando a aplicação num túnel de vento genérico
de baixa velocidade. Assim, descreve-se todo o processo de desenvolvimento do produto, nomeadamente as
suas fases: projecto da mecânica e electrónica inerente, fabrico e testes.
Este equipamento é composto por um sistema mecânico que facilita o desacoplamento da carga aero-
dinâmica e que permite um rápido posicionamento do modelo. Consiste numa estrutura octaédrica do tipo
Plataforma de Stewart e numa viga que supporta o modelo. Cada tirante da estrutura está instrumentado
com quatro extensómetros que formam uma ponte completa de Wheatstone. Desta forma, os esforços axiais
a que estão sujeitos são monitorizados através de um sistema de aquisição de dados, também desenvolvido no
âmbito deste projecto tendo como base soluções digitais integradas. É composto por um módulo mestre que
comunica, através de um barramento Serial Peripheral Interface, com seis módulos escravos (um por tirante).
Os dados adquiridos são depois tratados utilizando um programa implementado em MATLAB.
Para obter uma relação numérica directa entre as cargas aerodinâmicas e a resposta dos sensores, foi
efectuada uma calibração estática, antes de qualquer teste, aplicando pesos padrão na balança. Foi estimado
um polinómio linear multi-variável, utilizando o método dos mínimos quadrados, para interpolar os dados
da calibração. As incertezas das cargas aplicadas e das leituras dos sensores foram consideradas no cálculo
dessa regressão. Assim, é possível complementar, aos dados obtidos dos testes, informação acerca das suas
respectivas incertezas e qualidade da aproximação polinomial. Por último, é realizada uma série de ensaios em
túnel de vento utilizando como modelos de teste uma placa plana e um disco, normais ao escoamento, um
cilindro atravessado e outro orientado com o escoamento. No ultimo caso de estudo, um cone de pequena
dimensão é colocado a montante do modelo para avaliar a consequente redução da força de resistência
aerodinâmica.

Palavras-Chave: Tunel de vento, Balança, Cargas aerodinâmicas, Sistema de aquisição de dados,

Calibração

V
Abstract

Over 70% of the wind tunnel tests to scaled models demand the measurement of the aerodynamic load acting
there. Wind tunnel balances are highly customized and accurate instruments, hence expensive. In this context,
the object of the present master thesis is the development of a 6-component external force-moment balance,
to stand as a simple and economical alternative to the existing commercial solutions. It has been designed
aiming its application to a generic low velocity flow wind tunnel. The process of the product development is
described as follows: mechanical and electronic design, manufacture and test.
The device includes a mechanical system consisting of a spatial 6-6 in-parallel mechanism, which allows
an easy decoupling of the aerodynamic load, and a single vertical strut fixing the model. Each leg connector
has four strain gages attached to its surface in a full Wheatstone bridge configuration. The applied load is
monitored via a Data Aquisition System (DAQ), developed in-house based on fully digital integrated solutions.
It comprises one master module, which communicates through Serial Peripheral Interface (SPI) with six slaves
- each per leg. The data acquired is then reduced using a software, integrated in a MatLab environment.
In order to obtain a numerical relationship between the aerodynamic load and the output signal of the
force transducers, a static calibration is performed prior to testing by applying known weights on the balance.
A calibration rig has been developed to calibrate the balance without the aid of reference points provided
by the test section. A multi-variable polynomial fitting using the least squares method is used to interpolate
the calibration data points. The uncertainties of both the applied loads and the reading of the sensors are
considered in the regression. The data reduction includes the estimation of the calibration coefficients, the
fitted values of the aerodynamic load components and their respective uncertainties, as well as the goodness
of fit. Finally, a series of wind tunnel tests are carried out: a flat rectangular plate facing the oncoming flow
is tested to access the uncertainty of the measurement in the drag direction, a disk, a cylinder transverse to
the airflow, and another oriented with the axial direction.

Keywords: Wind tunnel, Six-component balance, Aerodynamic loads, Data acquisition system,
Calibration

VI
Contents

Agradecimentos III

Resumo V

Abstract VI

List of Figures XIII

List of Tables XV

Acronym XVII

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Wind-axis Reference Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Product Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 Required Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 Load Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.3 Dynamic Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.4 Data Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 State of the Art 7

2.1 Mechanical Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Internal Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 External Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Model Motion Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Rotary-balance Test Rig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Forced Oscillation Test Rig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.3 Parallel Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Stewart Platform Based Force-Moment Sensor . . . . . . . . . . . . . . . . . . . . . . . . . 11

VII
 2.4 Force Transducers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.1 Metal-foil Strain Gage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Bragg Grating Based Strain Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.3 Piezoelectric Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.4 Semiconductor Strain Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Project Scope 15
3.1 Mechanical System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.1 Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.2 Design Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Signal Processing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2.1 DAQ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.2 Data Reduction System (DRS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Static Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4 Mechanical Design & Manufacture 25

4.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Mapping Matrix Expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Singularity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3.1 Singular Geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3.2 Prototype Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4 Isotropy Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.5 Structural Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.5.1 Sensing Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.5.2 Finite Element Analysis (FEA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.6 Manufacture Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5 Data Acquisition Design & Manufacture 41

5.1 Strain Gage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Slave Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2.1 Bridge Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2.2 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2.3 Driving Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.4 Bridge Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.3 Master Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3.1 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

6 Static Calibration 53

VIII
 6.1 Calibration Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.1.1 Formulation of the Weighted Least Squares (LS) Method . . . . . . . . . . . . . . . . 54
6.1.2 Weighting Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.1.3 Uncertainty Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.2 Calibration Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.2.1 Loading Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2.2 Parameters Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.2.3 Quality-of-fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

7 Experimental Methods 63
7.1 Facility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.2 Peripheral Systems and Test Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7.3 Description of the Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
7.4 Results - Rectangular Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.4.1 Drag Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.4.2 Pressure Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.4.3 Experimental Uncertainty Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

8 Conclusion and Future Work 77

8.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
8.2 Features of the Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
8.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

Appendices 81

A Budget 82

B Structural 85

C DAQ and DRS 88

D Calibration 97

E Experimental Methods 102

IX
X
List of Figures

1.1 FST05e running the endurance event, Formula Student UK (FS-UK) 2014 . . . . . . . . . . . 2
1.2 European axis system, schematic of the test section . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 The Strouhal-Reynolds number relationship for circular cylinders. Data from Lienhard (1966) . 5

2.1 Schematic of aircraft model mounted on conning mechanisms at Defence Science and Tech-
nology Organisation (DSTO), in Erm (2006) . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 The Model Positioning Mechanism (MPM) in the open test section of the German-Dutch Wind
Tunnels (DNW)-Braunschweig Low-speed Wind Tunnel (NWB), in Bergmann et al. (2008) . 11

3.1 Illustration of the Stewart platform-based sensors . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2 Ball-and-Socket joint on the left, Point Contact joint (unilateral constraint) on the right . . . 16
3.3 Schematic of a Stewart platform-based sensor with ring sensing elements (Dwarakanath et al.
, 2001) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Signal Processing Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 Summary of the Strain Sensor Technologies Features . . . . . . . . . . . . . . . . . . . . . . 19
3.6 Wheatstone Bridge Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.7 Complete force transducer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.1 Octahedron in the central configuration, plan view . . . . . . . . . . . . . . . . . . . . . . . . 26

4.2 Spatial 6-6 in-parallel mechanism, plan view . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Configurations with singular matrix of line coordinates, from Lee (2000) . . . . . . . . . . . . 29
4.4 Classical Stewart platform notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.5 Modified Stewart platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.6 Balance Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.7 Bar fitted with longitudinal and transverse Strain-Gage (SG)s . . . . . . . . . . . . . . . . . . 33
4.8 FEA model - boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.9 FEA results - static analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.10 FEA results - static analysis detail, rotation vector sum . . . . . . . . . . . . . . . . . . . . . 36
4.11 Failure criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.12 Detail of the platform-strut assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.13 FEA results - modal analysis without test model, displacement vector sum . . . . . . . . . . . 38

XI
4.14 FEA results - modal analysis with test model, displacement vector sum . . . . . . . . . . . . 38

4.15 FEA results - mode 2 detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.16 Uniform cantilever beam with lumped mass, in Harris & Piersol (2002) . . . . . . . . . . . . 39

4.17 Manufacture of the bottom and top brackets . . . . . . . . . . . . . . . . . . . . . . . . . . 40

5.1 The strain gage system, L Hannah & E Reed (1994) . . . . . . . . . . . . . . . . . . . . . . 41

5.2 Considered configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

5.3 Integrated solution schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

5.4 Ratiometric operation functional diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5.5 AC excitation configuration schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.6 MIC4427 test circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.7 Data Acquisition Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5.8 dsPIC FST Module, v1.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5.9 Independent Slave configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5.10 Data acquisition box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

5.11 Flowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

6.1 Calibration Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

6.2 Calibration algorithm flowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

6.3 Eccentric loading setup, FZ MX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

6.4 Frame of reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

7.1 Open-return wind tunnel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

7.2 Test section dimension analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

7.3 Wind tunnel test apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

7.4 Pitot-static tube schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

7.5 Expected major flow disturbances of the flow past a model. In Tropea et al. (2007) . . . . . 66

7.6 Pressure sensor performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

7.7 Flat plate mounting setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

7.8 Rectangular plate - Up approach scatter plot . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

7.9 Rectangular plate - Down approach scatter plot . . . . . . . . . . . . . . . . . . . . . . . . . 69

7.10 Rectangular plate - drag coefficient results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

7.11 Influence of the test articles on the airflow velocity . . . . . . . . . . . . . . . . . . . . . . . 71

7.12 Rectangular plate - corrected drag coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . 71

7.13 Rectangular plate - Pressure center -Z coordinate . . . . . . . . . . . . . . . . . . . . . . . . 72

7.14 Experiment Uncertainty Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

7.15 Rectangular plate - offset at the end of runs . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

XII
B.1 Manufacture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

C.1 WTB_Amp v1.0 production masks (not to scale) . . . . . . . . . . . . . . . . . . . . . . . . 88

C.2 WTB_Ampv0.2 mounted on test module v0.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 88
C.3 WTB_PDmodule v1.0 production masks (not to scale) . . . . . . . . . . . . . . . . . . . . . 89
C.4 Wind tunnel testing GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
C.5 Wind tunnel static calibration GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
C.6 Procedure to glue the SG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
C.7 Force tranducers quality control test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

D.1 Uniform uncertainty distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

E.1 Rectangular plate wind tunnel test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

E.2 DRS test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
E.3 80I transverse cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
E.4 120I disk model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
E.5 120I cylinder, axial to the oncomign flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
E.6 80I transverse cylinder - Up approach scatter plot . . . . . . . . . . . . . . . . . . . . . . . 106
E.7 120I disk - Up approach scatter plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
E.8 Transverse cylinder and disk results - 95% confidence interval . . . . . . . . . . . . . . . . . . 106
E.9 Pressure sensor calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

XIII
List of Tables

1.1 Definition of the positive axis direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 Wind tunnels specs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Predicted load cases relative to the airflow axis system . . . . . . . . . . . . . . . . . . . . . 4

4.1 Geometry of the Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4.2 Drawn tubes - price comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

4.3 Vibration modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 HBM strain gages features, HBM (n.d.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

5.2 HBM strain gages Price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

5.3 Bridge output voltage estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

5.4 Force transducers resolution requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.5 Effective Resolution (Peak-to-Peak Resolution) vs. Gain and Output Data Rate . . . . . . . . 46

5.6 Peak-to-Peak Noise (nV) vs. Gain and Output Data Rate . . . . . . . . . . . . . . . . . . . . 46

6.1 Applied calibration loads and associated uncertainties . . . . . . . . . . . . . . . . . . . . . . 60

6.2 Force transducers output readings rmV s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

6.3 Calibration coefficients and uncertainty estimations of the load components FX and MX . . . 61

6.4 Estimated calibration loading conditions and associated uncertainty . . . . . . . . . . . . . . 61

6.5 Goodness-of-fit χ2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

7.1 Cd uncertainty as a 95% confidence range - drag counts . . . . . . . . . . . . . . . . . . . . 70

7.2 CP uncertainty as a 95% confidence range [mm] . . . . . . . . . . . . . . . . . . . . . . . . 72

7.3 Uncertainty of measured variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

A.1 WTB_Ampv1.0 list of components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

A.2 WTB_PDModulev1.0 list of components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

A.3 WTB Force transducers list of components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

A.4 WTB Mechanical structure list of parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

A.5 Cost of the balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

XIV
B.1 Material properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

D.1 Weight of the calibration weights and their supports . . . . . . . . . . . . . . . . . . . . . . . 97

D.2 Calibration loading plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
D.3 Uncertainty of the calibration loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
D.4 R matrix - average of all calibration runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
D.5 Relative error of the estimated calibration loading Ŷ [%] . . . . . . . . . . . . . . . . . . . . 101

E.1 Output files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

E.2 Rectangular plate - CD data points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
E.3 Rectangular plate - drag coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
E.4 Rectangular plate - Up approach, 1st run values . . . . . . . . . . . . . . . . . . . . . . . . . 107
E.5 Rectangular plate - CD partial derivatives and uncertainties associated with its variables . . . 107
E.6 Rectangular plate - variance of the CD variables . . . . . . . . . . . . . . . . . . . . . . . . . 107
E.7 Pressure sensor calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

XV
Acronyms

ADC Analog-to-Digital Converter.

BFSL Best Fit Straight Line.

CFD Computational Fluid Dynamic.

CMRR Common-mode Rejection Ratio.

CS Chip Select.

DAQ Data Aquisition System.

DC Direct Current.

DERA Defence Evaluation Research Agency.

DNW German-Dutch Wind Tunnels.

DOF Degrees of Freedom.

DRS Data Reduction System.

DSTO Defence Science and Technology Organisation.

EMI Electromagnetic Interference.

FBG Fiber Bragg Grating.

FEA Finite Element Analysis.

FS Formula Student.

FS-UK Formula Student UK.

FSAE Series Formula Society of Automotive Engineers Series.

GF Gage Factor.

IMU Inertial Measurement Unit.

IST Instituto Superior Técnico.

LNEC Laboratório Nacional de Engenharia Civil.

LS Least Squares.

XVI
MPM Model Positioning Mechanism.

MSP Mixed Signal Processing.

NACA National Advisory Committee for Aeronautics.

NWB Braunschweig Low-speed Wind Tunnel.

OMS Oscillatory Model Support.

PCB Printed Circuit Board.

PGA Programmable Gain Amplifier.

PKM Parallel Kinematic Mechanism.

Projecto FST Novabase Projecto Formula Student Técnico Novabase.

QI Quality Index.

Re Reynolds number.

SG Strain-Gage.

SPI Serial Peripheral Interface.

St Strouhal number.

UART Universal Asynchronous Reciever/Transmitter.

UL University of Lisbon.

VIP Vacuum Infusion Process.

XVII
Chapter 1

Introduction

Wind tunnel testing is a powerful tool aiding engineers to expedite the design process, allowing to understand
the true nature of aerodynamic loads. In many cases, however, without a proper instrumentation and data
acquisition system, it becomes useless. One of the key measuring systems in wind tunnel testing is the multi-
component force and moment measure instrumentation - more than 70% of the tests in a wind tunnel require
some kind of load measuring system, in Tropea et al. (2007). Early on, these devices were purely mechanical
and their mechanism resembles balances for weighting, hence its name. Today, these balances are often based
on force transducers and may assume many different types of configurations.

1.1 Motivation

Projecto Formula Student Técnico Novabase (Projecto FST Novabase) is the Formula Student (FS) team of
the University of Lisbon (UL) and, over its 13 years of existence, it has always aimed to set the standards of
the Portuguese engineering excellence in motorsports (FST (May 2015)). In 2012 the team decided to design
a prototype, the FST05e (figure 1.1), which comprises a full aerodynamic package that, despite the rather
low velocities at which FS cars run, allows taking advantage of the downforce effect by increasing the limits
of acceleration of the prototype.
One of the major setbacks, to the design phase of the aerodynamics of this prototype, was the lack
of computational power required to perform the several Computational Fluid Dynamic (CFD) simulations.
Moreover, the short lead time and the limited human and financial resources were decisive factors taken into
account on the decisions of the early design phase. The outcome is an aero-package with a very clean design
and simple solutions that, despite its drawbacks, offers a good platform for future developments.
Projecto FST Novabase looked to validate the CFD theoretical models by running a set of wind tunnel
tests at the facilities of Laboratório Nacional de Engenharia Civil (LNEC). Since there were only Strain-Gage
(SG) drag balances available, the axial and normal components of the aerodynamic load had to be measured
in turn. The former was monitored through the use of their force transducers, and to estimate the latter, a
new device had to be designed - the model was supported by a set of load cells, which in turn were fixed
to a structure. Despite the several attempts and time invested, the team was not able to gather useful data
mainly due to the lack of accuracy and, in the case of the normal force component test, it was found later
on a major flaw in the concept of the equipment developed. In addition, it has been noticed while testing a
slightly movement of the model in relation to its original position, compromising before hand any data that
would be retrieved. Nevertheless, this period of trial and error allowed to deepen the understanding of wind
tunnel testing. The team addressed all the problems it encountered, and started working on a project of a
six-component external wind tunnel balance. A primarily prototype was built to access its structural integrity

1
and feasibility but soon after this initiative had to be left in standby.
There are several types of multiple axis weighting devices available on the market but these products are
usually very expensive since they are highly customized and there are very few generic solutions, a direct effect
from its short demand. It has a very specific application, being employed mainly by research and development
organizations, fluid dynamics consultancy companies and at universities for post graduate research work or
undergraduate teaching. It would be then a major asset, if an affordable and flexible wind tunnel balance
was developed, that could still achieve an accurate estimation of the aerodynamic loads. This way it would
be possible to spread its use allowing smaller institutions or organizations to take advantage of this kind of
technology.
It is therefore object of the present thesis to continue the work that has been done so far by Projecto
FST Novabase and to provide a solution that overcomes the above described deficiencies in the prior art.
In other words, to design, build, and test a six-component dynamic force-moment external balance for both
steady or unsteady load measurements, aiming its application to a generic low velocity flow wind tunnel. This
work is within the scope of the rehabilitation project of the fluid mechanics laboratory of Instituto Superior
Técnico (IST). Based on past experience, the quality of the data acquisition, the reliability of the equipment,
and ease of use and maintenance will be privileged.

Figure 1.1: FST05e running the endurance event, Formula Student UK (FS-UK) 2014

1.2 Wind-axis Reference Frame

Given the end-use of this equipment, and for clarity sake, it is important to assign a standard reference frame
to the tunnel. This work, uses an European inertial coordinate system fixed to the wind tunnel and aligned
with the main flow direction. It is consistent with the right-hand system and the definition is based on the
standard ISO 1151, as shown in figure 1.2.
This reference frame corresponds to the standard definition of a wind-axis system in which the x-axis
points towards the oncoming free stream velocity. This way, a balance that stays fixed in the tunnel, and
relative to the airflow axis system, always gives the pure aerodynamic loads acting on the model.

2
 Figure 1.2: European axis system, schematic of the test section

Table 1.1: Definition of the positive axis direction

Axis Symbol Component Name Positive Direction, ISO 1151
XX Axial Force, Drag Against the flow direction
YY Side Force To starboard
ZZ Normal Force, Lift Down
MX Rolling Moment Roll to starboad
MY Pitching Moment Turn up
MZ Yawing Moment Turn to starboard

1.3 Product Scope

The device herein presented is a 6-Degrees of Freedom (DOF) dynamic force-moment external balance, to be
placed below the wind tunnel test section, which resolves the aerodynamic load acting on the test model into
three orthogonal forces and their associated moments. The available space for its mechanical frame is not a
limiting factor.

1.3.1 Required Features

For utility sake, the equipment should include the following features:

• Mounting of the model with a single strut support, i.e. a single mounting point must be able to securely
fix the test articles. It provides more flexibility regarding their size and shape, thereby allowing several
different arrangements. Also, it adds simplicity, enabling to quickly change between setups.

• The balance must allow a quick and easy setup of the pitch, yaw and roll attitudes of the model.

• It must be easy to handle, such that it can be easily operated and moved around by a single person.

• A standalone Data Aquisition System (DAQ) based on a set of force transducers. This system should
be designed so it can monitor both steady and fluctuating aerodynamic loads.

• A Data Reduction System (DRS) to reduce the raw data to coefficient form and apply corrections
to remove unwanted effects due to test conditions and techniques. It should be able to perform strut
interference and weight corrections, as well as, customized filtering depending on the test being performed
and the data one wants to obtain.

• A graphical user interface, which displays the post-processed data in an user-friendly way.

3
1.3.2 Load Range

The nature and magnitude of the aerodynamic loads are a direct effect of the shape of the model and the
characteristics of the airflow. So, prior to the design phase, it is necessary to clearly state the wind tunnel specs
where the balance is going to be employed, as well as the size of the models and their respective aerodynamic
coefficients estimation. This way, it is possible to determine the maximum design load.
The wind tunnels, for which this equipment is designed, are those from the Fluid Mechanics and
Aerospace laboratories of IST and the closed return wind tunnel of LNEC. They all have very different features:
the maximum airflow velocity ranges up to 40 m/s, and their test sections vary substantially in shape, size,
and type. Table 1.2 summarizes their key specifications.

Table 1.2: Wind tunnels specs

Fluid Mechanics Lab. Aerospace Lab. Wind Tunnel
IST IST LNEC
Max. Airspeed [m/s] 20 40 40
Test Section Type Opened Chamber Opened Chamber Closed Chamber
Section Dimensions [mm] 400 x 600 I 1400 1200 x 1000
Type of Wind Tunnel Open Return Close Return Close Return

Usually, wind tunnel balances have a very narrow range of application. They are optimized to enhance
the data quality of a given kind of tests, and very often, this improvement comes with a decrease of their
range of applicability.
This device will be mostly used for research work and undergraduate teaching, but it might also be
required by student groups, such as Projecto FST Novabase. There is then a large number of potential
test articles which makes rather difficult to predict, accurately, the load cases at which it will be subjected.
Nevertheless, table 1.3 shows a set of articles which might be tested and their respective force and moment
coefficients. It gives an idea of the order of magnitude of the loads, allowing to identify possible maximum
combined load conditions.

Table 1.3: Predicted load cases relative to the airflow axis system

@ 40m{s
Ref. Area Moments
Coefficients Forces [N ]
[m2 ] [N.m]
CX CZ CM y FX FZ MY
FSAE Series 1:5 -0.65 2.40 0.052 -33.8 124.7
Round Disk -1.17 0.010 -11.7
NACAa 2410 0 -0.0106 -0.0204 0.0208 0.080 -0.847 -1.630 0.166
NACA 2410 9.5 -0.0554 -1.1176 0.0200 0.080 -4.431 -89.317 0.160
a Coefficients of National Advisory Committee for Aeronautics (NACA) airfoils predicted at Re  5  105 by XFoil

The maximum combined load is obtained for the Formula Society of Automotive Engineers Series (FSAE
Series) model; roughly 125N in the normal direction and 35N in the axial direction.

1.3.3 Dynamic Loading

The device in question must be able to provide feedback on the nature of the aerodynamic loading, which
may be steady or unsteady. Although both can show an uniform or non-uniform behavior, whether there is or

4
not a different condition in every point in the flow, the steady flow is not time dependent while the unsteady
is. It can show a periodic or a transient behaviour.
Vortex shedding behind bluff bodies is a non-uniform unsteady flow phenomenon, and its intensity is
a function of the Reynolds number (Re). It creates alternating low-pressure vortices on the downstream side
of the model which in turn cause it to oscillate at a determined rate. For this reason, resonance can occur if
the vortex shedding frequency matches the structural vibration frequency of the test rig This would spoil the
measurements by inducing inertial loads on the balance. Having this in mind, as well as the good conditioning
of the acquired data, the loads to be measured must be also characterized in the frequency domain in order
to provide design inputs for the mechanical structure and DAQ.
Vortex shedding is a very complex physical phenomenon, so the majority of the reported data are in
respect to flows around well known standard bluff bodies. Given its shape and wide range of application, long
rigid circular cylinders are often used as test article in boundary layer separation research.
The Strouhal number (St) is a dimensionless number which describes oscillating flow mechanisms and
it is defined by

St 
fL
(1.1)
U

Where f is the main shedding frequency, L is the reference length, and U is the free-stream velocity. Lienhard
(1966) gives an estimate of the Strouhal-Reynolds relationship for eddies on rigid circular cylinders, accurate
within 5%, figure 1.3.

Figure 1.3: The Strouhal-Reynolds number relationship for circular cylinders. Data from Lienhard (1966)

Figure 1.3 shows a reasonably steady behavior of the Strouhal number over a wide range of Re, from
3  10 to 2  105 . Given the maximum airspeed of the wind tunnels where this balance is going to be employed
2

(see table 1.2), and considering a test model with a reference length of 0.1m, the maximum expected Re is
in the order of 105 . This gives a Strouhal number in the order of 0.18 which corresponds, in this conditions,
to a vortex shedding frequency of 70 Hz. Although it may not be representative of the dynamic loading this
balance is going to experience, this value is still a good estimate of the limit of the flow frequency range.

5
1.3.4 Data Quality

The quality of the acquired data is a direct function of the cost of the equipment, so it must be carefully
addressed. Since this balance is not specially designed and built for one particular set of wind tunnel tests,
which is not usually the case, it is a challenge to set a goal for the quality of the acquired data. Moreover, it
highly depends on the design stage of a given prototype - the early phase rough iterations are not as demanding
as the late optimization tests. Despite all that, an estimate of the maximum required accuracy to get valid
data may be done.
The best way to understand the strong requirements demanded to a wind tunnel balance is by studying
the following example, where rough numbers involved in the problem are estimated. Generally, wind tunnel
balances are able to measure a difference of 1 to 2 drag counts1 , which is equal to less than 0.07% up to 0.1%
of the full scale load, in Tropea et al. (2007). Based on this information, and considering the drag coefficient
of the FSAE Series model, an accuracy of 1 drag count corresponds to 5.1 g (0.04% of the full scale load).
On the other hand, if this balance was designed, for instance, to test the 24XX NACA profiles family, the
maximum design load would be around 100 N. In this case, 1 drag count corresponds to 7.8 g (0.06% of the
full scale load) which is one and a half times larger. Late design iterations, by their turn, require accuracies
greater than 0.5 drag counts, according to Lerner & Boldes (2011), which means that the accuracy of the
balance in the drag direction must be higher than 2.5 to 4.0 g.
This analysis gives a good estimation of the order of magnitude for the accuracy to which this project
should aim. It is important to state, however, that this is not a requirement but a reference value instead,
since cost is a heavy limiting factor.

1.4 Structure of the Thesis

The present document is divided in 8 chapters, introduction included, and it is structured as follows:

• Chapter 2: Presents a quick review of the technologies inherent to wind tunnel balances and the existing
commercial solutions;

• Chapter 3: Gives an insight on the many subsystems of the balance and it addresses the early design
decisions;

• Chapter 4: Describes the mechanical design of the balance and following manufacture;

• Chapter 5: The DAQ is broken apart and each subsystem design (force transducers, master and slave
modules) is detailed, hardware and software wise;

• Chapter 6: The potential sources of errors are quantified. The static calibration is described including
the procedure to access the calibration uncertainty;

• Chapter 7: The experimental results, with particular emphasis on a flat plate facing the oncoming flow
test;

• Chapter 8: Summarizes the work and explains how the accuracy of the balance could be improved.

1 A drag count is a dimensionless unit used by aerospace engineers where 1 drag count is equivalent to a CD of 10
4 .

6
Chapter 2

State of the Art

In the path of mastering flight, the early aeronauts tried to follow the example of birds. Leonardo da Vinci, for
instance, sketched the Ornithopter in 1485. Also, numerous inventors came up with bird-inspired machines,
only to watch them flop to the ground. It was clear that in order for humans to fly, they needed a better
understanding of the interplay between wings and airflow. They had to carry out, under controlled conditions,
aerodynamic tests. First, relatively steady natural wind sources were searched out. Models were mounted
above windswept ridges and in the mouths of blowing caves but even so that could not offer helpful design
feedback.
In 1746, Benjamin Robins (1707-1751), an English mathematician and scientist, attached an horizontal
arm to a vertical pole, which rotated, sending the arm spinning in a circle. This is considered to be the first
ever designed weighting mechanism of aerodynamic loads. At the end of the arm, he affixed a variety of objects
and his tests immediately confirmed that the shape of objects had a tremendous effect on air resistance. The
whirling arm provided most of the systematic aerodynamic data gathered up to the end of the nineteenth
century. It has, however, a serious side effect; since it chopped the air as it spun, models at the end of the
arm in effect flew into their own wakes, with so much turbulence, experimenters could not determine the true
relative velocity between the model and air. Scientist needed cleaner artificial air flows.
Frank H. Wenham (1824-1908), a Council Member of the Aeronautical Society of Great Britain, con-
vinced the organization to help finance the construction of the first ever wind tunnel, which debuted in 1871.
Wenham’s tunnel was 3.7m long and 0.46  0.46m square. It produced 65km/h air-streams by means of
a steam-powered fan at the end of the tunnel. The great capabilities for controlled and systematic testing
quickly rendered the whirling arm obsolete, more than a century after its creation, and gather the conditions
for the development of more accurate measurement methods.

2.1 Mechanical Configuration

Nowadays, wind tunnel balances can be distinguished according to the number of aerodynamic load components
they can measure simultaneously (one to six are possible), as well as, by the location at which they are placed.
If they’re placed outside of the model or out of the test section, they are referred to as external balances,
otherwise they are referred to as internal balances.

7
2.1.1 Internal Balance

By rule, there is limited space inside the model itself, so internal balances have to be relatively small and
compact in comparison to the external ones. There are two main types of internal balances, most commonly
known as sting balances. The monolithic type, in which the balance body consists of a single piece of material,
is designed in a way such that certain areas are primarily stressed by the applied loads. The other internal
balance type uses small transducers which are oriented with their sensing axis in the direction of the applied
loads.
Internal balances are generally placed at the center of gravity of the model, which is built around it.
As a result, no interferences are introduced in the wind flow by the components of the balance, nevertheless,
a mechanical support for the model is always needed. This approach significantly increases the complexity
of the test models, since the balance has to be embodied and it also does not provide the same flexibility as
the external concept does. Internal balances are almost universally used for measurements in supersonic and
transonic tunnels. However, they’re also becoming popular in subsonic tunnels, in Aerolab (May 2015).

2.1.2 External Balance

There are two main types of external balances. The side-wall balances, used in half-model tests, are made from
one single piece of material. The other type comprises a set of single force transducers which are connected
by a framework. The wind tunnel balance herein presented, follows the latter concept. Such solution can be
built very stiff but needs more space compared with the one-piece design. There usually is, however, plenty of
space available around the wind tunnel test section for such a balance. Therefore, it is possible to explore a
large number of configurations with respect to measurement requirements, like the sensitivity and decoupling
of load interactions.
Some of the existing solutions are Wire, Platform, Yoke and Pyramidal balance designs. The wire
balance is the most ideal but the hardest to control. The model is suspended by wires connected to separate
force transducers. This particular method is not favorable as wire balances have to be kept still and this can
prove difficult in a running test section given the large tare drag associated with the wires. The Platform
type takes care of this issue by using a fixed platform to mount the model. Usually three or four columns are
used for support, and the aerodynamic load is computed as the sum of the components acting thereon. This
design is naturally orthogonal, but the resolving center of the balance is not at the center of the tunnel and
further data processing must be done. The Yoke balance design, in its turn, has the moment resolving center
near the center of the tunnel, however there is typically larger deflection compared to the Platform balance
configuration. The Pyramidal balance is the most complex, hard to maintain, and its alignment is by far the
most challenging one. It measures moments about the resolving center and is able to separate and read forces
and moments directly instead of summing up components.

2.2 Model Motion Mechanisms

Aerodynamic loads on wind tunnel models are generally measured with them steady, set at a known orientation.
Static tests are the most common form of testing carried out in wind tunnels and there are several useful static
measurements reported in the literature. However, sometimes there is a need to characterize damping effects
associated with the motion of models, such as aircrafts. Dynamic testing allows to do just that by taking
measurements while the model is in motion, undergoing some maneuver. Although this kind of tests may
provide useful data on their maneuverability and control, specially at high angles of attack, reported dynamic
data are somewhat limited compared with static data, due to the inherent practical difficulties of obtaining it,
in Erm (2006).

8
2.2.1 Rotary-balance Test Rig

The rotary-balance test technique was developed, in the 1920’s, to assess the effects of the angular rate on the
overall aerodynamic load acting on aircrafts in flight. Certain motions, such as the developed spin, generate
large differences in local flow directions, and the separated flow characteristics are extremely difficult to predict
or analyze using standard techniques, in AGARD (1996).
The primary goal of a rotary-balance apparatus is to produce a “coning motion”, a continuous rolling
motion of the vehicle about the free-stream velocity vector. In this case, the attitude of the model remains
constant with respect to the air-stream throughout a rotational cycle, hence, the flow is steady. The goal of this
experiment is to determine the overall forces and moments on the model as a function of the rotational rate,
by means of a six-component sting balance, in Bergmann (2007). A typical example of the model-mounting
procedure is shown in figure 2.1a.

(a) Coning (b) Oscillatory Conning

Figure 2.1: Schematic of aircraft model mounted on conning mechanisms at Defence Science and Technology
Organisation (DSTO), in Erm (2006)

Some rotary-balances also allow the axis of rotation to be given an angle of inclination relative to the
free stream, thereby inducing oscillations of angle of attack and side-slip during the model motion, in AGARD
(1996). Figure 2.1b illustrated what is referred to as “oscillatory coning” test.

2.2.2 Forced Oscillation Test Rig

The traditional forced oscillation test consists of moving the model in a sinusoidal motion about each of roll,
pitch and yaw axes independently. The angle of attack, amplitude and frequency of the oscillation are all
typically independent variables during the test.
There are several different mounting configurations, depending mainly on the number of DOF. Test
rigs may be specially designed for a given test or may allow performing different 1-DOF tests by changing the
mounting setup of the model. Detailed information on these methods can be found in Hoe et al. (n.d.) and
Owens et al. (n.d.).
Other types of test rigs, like the Oscillatory Model Support (OMS) of the German-Dutch Wind Tun-
nels (DNW)-Braunschweig Low-speed Wind Tunnel (NWB), take advantage of serial kinematic structures to
achieve a given number of DOF by serial arranging the corresponding number of linear and rotative axes. The
main disadvantage associated comes from the fact that the bottom-most axis of movement has to carry the
weight of all those lying above it. This has a negative impact on the requirement for high stiffness and high

9
dynamics. Moreover, all errors of movement of all axes are added. For these reasons most dynamic wind
tunnel rigs are limited to a small number of DOF, usually three at most, in Bergmann (2007).

2.2.3 Parallel Mechanisms

To overcome all the issues inherent to serial manipulators, mentioned in the previous section (2.2.2), it was
recently suggested at Defence Evaluation Research Agency (DERA) a new 6-DOF dynamic test rig based on
a parallel mechanism. A parallel mechanism is a closed-loop structure in which the end-effector, the mobile
platform, is connected to the base by at least two independent kinematic loops. It is also called Parallel
Platform or a Parallel Kinematic Mechanism (PKM). The most common, known as the Stewart Platform,
short for Stewart-Gough Platform, has six linear actuated legs mounted in pairs to the base of the device,
crossing over to three mounting points on a top plate, in Stewart (1965).
Parallel manipulators have advantages of high stiffness, accuracy and large load carrying capacity - while
the ratio of the mass of the payload over the mass of the device is typically 0.15 for serial 6-DOF industrial
arms, this ratio can be larger than 10 for parallel structures. All this holds true, however, as long as they
operate far from a singular configuration. These singularities in their workspace make the force transformation
matrix loose its rank (number of zero eigenvalues), leading to instantaneous gain of degrees of freedom for
external loads applied in specific directions. The singularity aspects of parallel mechanisms have already been
addressed by several authors. For instance, Zhang & Crane (2004) show different types of singularity conditions
based on the analysis of the Jacobian matrix that is formed by the lines of action of the six leg connectors.
More detailed information and references of theoretical and experimental investigations on this subject can be
found in Merlet (2006) and Tsai (1999).
This mechanism is able to excite the forced sinusoidal in the modes of pitch, roll, yaw, heave and
lateral oscillation about arbitrary axes, despite being usually performed about model-fixed oscillation axes.
Furthermore, it can also take input functions to create desired dynamic motions which simulate realistic flight
maneuvers, in Bergmann (2007). The first analysis to the use of parallel kinematics as a test rig in a wind
tunnel has been made by Greenwell (1996), but to the author’s knowledge, the inherent large number of
singular positions of the proposed kinematics restricted the available workspace too much. In addition, the
first approach to develop a 6-DOF test rig, at DNW-NWB in 2000, revealed that the large number of joints
were challenging for stiffness and precision and the room available was relatively small compared to the size of
the machine. Nevertheless, this kind of apparatus still has several advantages compared with a conventional
multi-axis test rig in serial arrangement:

• Higher dynamics despite identical input power because lower weights are being moved;

• Higher accuracy because errors of single axes are not cumulative as in serial manipulators;

• Higher stiffness regarding weight of the components because only forces in axial direction of the legs are
acting.

These features lead them to develop a new model support called Model Positioning Mechanism (MPM), in
2004, as an improved successor of the previous versions, see figure 2.2. The mechanism is located above the
test section allowing to perform tests with ground effect using rolling devices.

10
 Figure 2.2: The MPM in the open test section of the DNW-NWB, in Bergmann et al. (2008)
.

Formula 1 teams, such as Sauber F1 Team, also take advantage of these kind of mechanism to expedite
their wind tunnel test runs. The model is able to be raised, up and down, and its attitude changed in a quick
and easy way, saving time and energy. Detailed information can be found in Sauber (May 2015).

2.3 Stewart Platform Based Force-Moment Sensor

The advantages of the compact design of these mechanisms, already mentioned in section 2.2.3, leads one
to consider them for force-transducer sensor applications. The sensing task consists in estimating the forces
and torques acting on the upper platform from those measured at its legs. Thus, a Stewart platform with
instrumented elastic legs can be used as a force sensor. The development of the first sensor of this kind was
due to Gaillet & Reboulet (1983), based on the octahedral configuration.
The behavior of the Stewart platform sensor has already been thoroughly studied by many authors,
namely Romiti & Sorli (1992), Sorli & Zhmud (1993) and Dai et al. (1994). Some researchers, like Dasgupta
et al. (1994), also considered the optimality of the condition number of the force transformation matrix for
stability problems. A design methodology for the condition number of the Stewart platform sensor tending
towards structural isotropy, has been considered by Svinin & Uchiyama (1995) while the design of Sorli
& Pastorelli (1995) possesses marked anisotropy, with greater stiffness in preferred directions. The former
approach makes the sensitivity to the externally applied load to be approximately the same in all directions,
therefore, it’s useful for force control with arbitrary loads. The latter, on the other hand, is more suitable for
predetermined loading conditions.
There are already several designs which follow the concept suggested by Sorli & Pastorelli (1995). They
take advantage of spherical and flexure jointed Stewart platforms in near-singular configurations, wherein the
sensitivity is high in certain predetermined directions of applied loads, in Ranganath et al. (2003). In these
directions, there is a significant magnification of forces in the legs, so very small external forces and moments
can be easily sensed. In the other directions, the sensitivity is that of a normal load sensor determined by the
sensitivity of the sensing element.

11
2.4 Force Transducers

Wind tunnel balances can also be classified by their type of weighting mechanisms, namely force compensation
or force transducer. The first system is purely mechanical, based on the classical analytical balance. It has
been used since the early wind tunnel tests and it still is the most simple and cheapest solution. However,
most devices of this kind allow measuring just a single component of the aerodynamic load, narrowing down
considerably its range of application.
Force transducers wind tunnel balances, in their turn, allow measuring several components at the same
time requiring at least one sensing element for each of them. SGs are widely used as sensing elements
in modern, precise force transducers. The electrical SG provides both outstanding technical properties and
cost-effectiveness. Besides the electrical SG, the measurement principle of Fiber Bragg Grating (FBG) based
sensors has recently become established and first applications in experimental mechanics have shown that this
technology offers a lot of potential. On the other hand, in many applications balances based on semiconductors
and piezoelectric force transducers are also a recommended alternative solution for their singular features.
There are several other types of force transducers available, but we find these to be the most commonly seen
in wind tunnel applications.

2.4.1 Metal-foil Strain Gage

Metal-foil based force transducers always comprise a spring element to which force is applied. The elastic
deformation of the measuring body causes their electrical resistance to change linearly, thereby allowing to
monitor strain, . The Gage Factor (GF), K, is a property of the strain gages, and it relates the change in
resistance, ∆R to the change in length ∆L.

 ∆R {R0  ∆R{R0
∆L{L
K (2.1)
0 

The gage factor is constant for a given metal-foil strain-gage and it ranges between 2 and 5, in Hoffmann
(n.d.a). By rearranging the terms, we arrive at the equation for strain.

∆R{R0
 (2.2)
K

This type of technology, due to its inherent linearity, very small mass, wide frequency response (from
zero to more than 100,000 Hz), general versatility and ease of installation, makes it ideal for use in shock and
vibration instrumentation, in Harris & Piersol (2002, Chapter 17th). Furthermore, metal foil based sensors
show almost no drift and are therefore particularly suited for long term monitoring tasks, in HBM (May 2015b).
The output signal of the electric SGs is, however, dependent on temperature variations due to the change
of the electrical resistance of the measuring grid and thermal expansion of the spring element. Although this
error can be largely compensated by using a Wheatstone bridge circuit, perfect error compensation requires
all sensors to behave identically and to be perfectly aligned, which is virtually impossible to guarantee.
Another source of error that must be pointed out is due to strain-gage creep. Experiments have shown
that if a test element is subjected to a suddenly applied load, the strain-gage bridge output decreases with time
from an initial value. The rate of decrease is relatively large for the first few seconds and decreases thereafter
so that the output approaches a constant asymptotic value. This phenomenon is known as creep effect. The
magnitude of the total change with time has been found to be approximately proportional to the unit strain
of the material to which the gage is bonded when subjected to the applied load. Edwards (2000) states that
the creep effect is, apparently, the result of a partial relaxation of the bonding material .

12
2.4.2 Bragg Grating Based Strain Sensor

Bragg sensors are made out of a short segment of optical fiber that reflects particular wavelengths of light
and transmits all others. This reflection is caused by the multiple layers of alternating materials structure,
with varying refractive index, called Bragg grating. Stretching the fiber optic Bragg sensor causes the grating
period to change resulting in a change in the wavelength of the reflected light.
Bragg sensors and electric strain-gages, despite being based on distinct physic phenomenas, share many
features mainly in respect to their package, way of handling and installation. Its long term stability is also
very high.
The drift of metal-foil gages may be explained by their oxidation. Therefore, since Bragg gratings are
made from glass, a material that is highly corrosion resistant, they may be considered free from drift, in Kleckers
(2009). Also, Fiber optic sensors are essentially immune to Electromagnetic Interference (EMI), because the
transducing technique used by the sensors does not involve electrical signals, in Kreuzer (n.d.). High EMI can
be verified in some wind tunnel experiments, such as during plasma torch testing or if a transverse, i.e. model
motion mechanism, is used as in dynamic testing (due to the motors), in Edwards (2000). Apart from these
direct advantages, they benefit from better resolution and accuracy in comparison to metal SG despite having
a lower sensitivity.
There are, however, some properties that show some weaknesses. The temperature has a very strong
impact on the FBG sensors signals, which are dependent on two physical effects. The variation of the refractive
index of the glass and dilatation of the spring element. Since the Wheatstone bridge can not be used to mitigate
these effects, as it happens with electrical SGs, precise measuring can only be achieved with proper temperature
compensation, in Kreuzer (n.d.). Another downside, inherent to FBG sensors, is the difference in sensitivity
between the positive and negative loading. This means that a force transducer has different sensitivities for
tensile and compressive loads, in (Kleckers, 2009). Furthermore, considering the singular features of this type
of technology, it requires a more complex and expensive DAQ, making it less attractive than the other solutions
for standard wind tunnel balance application.

2.4.3 Piezoelectric Sensor

The basis of the piezoelectric effect is that crystals under compressive loading generate an electric charge that
is directly proportional to the applied force. This charge is then converted to a proportional output voltage
with the aid of an amplifier.
This means that piezo-sensors do not rely on spring elements for support. Consequently, the unavoidable
deformations of SG balances are usually one or two orders of magnitude larger than the ones experienced by
piezo-balances. This leads to higher natural frequencies, which is a prerequisite for applications in unsteady
aerodynamics, in Tropea et al. (2007).
Due to their principle of operation, they tend to have an estimated drift of 1N/min. This value does not
vary as a function of the magnitude of the applied force, therefore, the relative measurement error is particularly
unfavorable when small forces are being measured over a long period of time. Moreover, piezoelectric sensors
have a slightly higher linearity error than the other options, in general 0.5% relative to full scale, in HBM
(May 2015b). On the other hand, they can be very compact, being an optimal solution when integration in
very small systems, such as internal balances, is required.

2.4.4 Semiconductor Strain Sensor

The use of semiconductors, such as germanium and silicon, for strain sensing applications has increased
over the past few years. This technology has made miniature strain transducers available that compare with

13
piezoelectric devices in dynamic applications but are also capable of producing Direct Current (DC) signals.
Unlike other strain sensors, semiconductors are based on the piezoresistive effects of semiconductor materials,
and measure the change in resistance with stress as opposed to strain. The physical phenomenon in question
can be better understood in Geyling & Forst (1960).
As in the case of the metal SGs, the basic effect is a change in resistance with the strain of the spring
element. In addition, their resistivity also varies along with the physical dimensions. So the resistance-to-strain
relationship is nonlinear, varying 10 to 20% from a linear equation. Furthermore, compared with metallic foil
sensors, they show a greater sensitivity to temperature variations and tendency to drift.
Nevertheless, semiconductors strain-gages have a lower cost and a much larger gage factor. It may
vary between 50 and 200, which is 25 to 100 times larger than that associated with metal-foil gages. Also,
the signal conditioning is still typically a bridge circuit and by means of more complex computer-controlled
instrumentation, its inherent limitations can be overcome via software compensation, in Chipkin (May 2015).

14
Chapter 3

Project Scope

Given the complexity of the present device and since this is the first prototype of its kind, ever designed and
built at Instituto Superior Técnico (IST), reliability, ease of use and maintenance were of paramount importance
and are, therefore, reflected on the engineering solutions. In addition, the cost of the equipment was a heavy
limiting factor, so it has been kept substantially lower than the commercial alternatives. An overall budget
of 1500 e was established based on a rough assumption which can be found in appendix A. To meet all the
requirements (stated in the section 1.3) within budget, it was absolutely necessary to develop from scratch,
whenever possible, the many subsystems which compose the final product. So an extensive cost to benefit
analysis has been made to determine whether it was best, to purchase ready-to-use commercial solutions or
to purchase raw material and components instead.
A top-down design approach has been carried out in this work. The balance was split up into two main
subsystems, namely the Mechanical System and the Signal Processing System. The former provides the means
to support the model, allows setting up tests and eases the acquisition and decoupling of the aerodynamic load
into six force components. The latter, on the other hand, can be further split in a Data Aquisition System
(DAQ), which holds all the sensing components and signal conditioning module, and a Data Reduction System
(DRS) which digitally processes the data, and is responsible for displaying the values of the measured forces
and associated moments. There is, between both, a one-way communication protocol from the DAQ to the
DRS.

3.1 Mechanical System

3.1.1 Configuration

As mentioned before, the Stewart platform is a promising structure for the intended application (see figure
3.1a). Theoretically speaking, each measurement elastic leg just sustains tensile or compressive strain along its
axis regardless of the gravity and the frictional moment in the spherical pairs, thereby measuring all components
of the applied external load without stress coupling. It has distinguishing advantages of good stiffness and
sensitivity, already addressed in previous sections (2.2.3 and 2.3). Furthermore, it is a symmetric structure and
it has a straightforward mapping expression between the applied load on the upper platform and the forces
acting on its legs, making it particularly suitable for applications in six-axis load sensors.
Even so, the standard Stewart platform shows some disadvantages that must not be overlooked.

• The large number of joints increases the overall complexity of the device. There are twelve spherical
pairs required to be adjusted and pre-stressed separately which is a laborious task since it is difficult to

15
 make the pre-stressing forces to be uniform.

• The contact area of the spherical pairs lead to frictional forces between the ball and the socket joints
which result in stress coupling in the sensing element, namely bending and axial stresses.

• The inherent clearance that exists in the traditional spherical pair causes mechanical hysteresis, damaging
the linearity of the sensor.

There are other solutions that mitigate these negative effects, for instance, hyper-static forms of the Stewart
platform as the one proposed by Hou et al. (2009) or monolithic designs as it is suggested in Sandu et al.
(2011).
Hou et al. (2009) hyper-static design has a “pre-stressing element, which is composed of two spherical
pairs and a screw pair installed at the geometrical center of the upper and lower platforms, respectively. In
other branches, spherical pairs bear a the unilateral constraint instead of the bi-directional constraint” (see
figure 3.1b and 3.2). This way, by abandoning the traditional concept, the above mentioned deficiencies can
be greatly reduced. Anyhow, this is a rather new and somewhat exotic concept that still needs to be proven,
so it does not follow the adopted philosophy towards reliability.

(a) 3-3 Stewart platform (b) Hyper-static Stewart platform

Figure 3.1: Illustration of the Stewart platform-based sensors

Figure 3.2: Ball-and-Socket joint on the left, Point Contact joint (unilateral constraint) on the right

The Sandu et al. (2011) design is a monolithic elastic structure. Despite reducing the nonlinearities
and hysteresis effects, which are mainly induced when assembled parts are used, the decoupling of the applied
load is significantly more sensitive to dimensional and geometric tolerances, as well as, the placement of the
strain sensors. The frame has a complex geometry which makes it particularly difficult to manufacture with
the available equipment. In addition, the data reduction is not a straightforward process since every monitored

16
section is stressed in a different way having coupled loads, such as bending moments and axial forces, acting
there. For these reasons, crossing influences of the load components would most certainly appear.
On the other hand, since this is a monolithic structure, any damage it might suffer would potential
render the device unusable. This balance is meant to be used for research work and undergraduate teaching
so it has to be sturdy and easy to repair, if damaged. A modular solution manages just that, a quick and
inexpensive replacement of damaged parts. It also provides a very good platform for future development,
unlike the monolithic solution which does not have the flexibility to be modified if required later on.

3.1.2 Design Approach

The mechanical design starts with a geometry analysis; For a multi-component load sensor, measurement
isotropy is of key importance to its performance. Isotropy means that the sensor exhibits approximately equal
sensitivity for all components of the applied external load. Ideal spatially isotropic sensors are considered as
maximally informative and most slightly affected by coupling noise, machining tolerances, and other error
sources (Yao et al. , 2011). It also leads to the minimum relative error in the force mapping. Therefore, it is
of most interest taking advantage of this kind of geometry configuration. Equally important is to guarantee
that the framework is far from a singularity configuration. If the mapping matrix is singular or seriously ill-
conditioned, it leads to high divergence among leg forces in response to loads in certain directions. As stated
before, this is an issue inherent to this type of structures so it must also be carefully addressed.
Another crucial aspect for the performance of the balance is the shape of its elastic members. Their
shape allows to enhance the sensitivity, accuracy and quality of the axial force transducers. A sensor with
tubular legs requires very small cross-sectional area to withstand the load. However, if the buckling critical
load is not high enough, an increase in cross-sectional area must be considered. If this area increase comes
over the increase of the wall thickness, the overall sensitivity of the balance is seriously affected. So, to avoid
this effect, a sensing element is usually added to the leg which can take many different shapes. One of them,
that is worth mentioning for its simplicity, is the ring-shaped sensor element, as suggested by Dwarakanath
et al. (2001) (figure 3.3). The geometry of the ring is determined by the mean radius, R, its width, b, and
depth, d.

(a) Stewart platform-based sensor (b) Ring-shaped sensing element

Figure 3.3: Schematic of a Stewart platform-based sensor with ring sensing elements (Dwarakanath et al. ,
2001)

The ring develops a uniform strain distribution in the gage area, which is a condition required for a

17
good measurement, and the output is limited by the maximum allowable strain level in that region. Since its
geometry is parametric, the sensitivity to compliance ratio can be easily optimized according with the design
requirements. “The goal is to choose the geometry such that the strain is maximum and the deflection under
the load is minimum”. Moreover, for a given axial force, the characteristics of bending give rise to stresses
with opposite signs on opposite faces of the ring resulting in the mechanical amplification of the signal by a
factor of two (Dwarakanath et al. , 2001).
Ring-shaped sensing elements, and others, greatly improve the signal-to-noise ratio being particularly
useful when small loads are to be measured such as in this case. However, notwithstanding its intrinsic benefits,
circular tubular elastic members were chosen due to its simplicity. Also, the project design requirements are
not defined to the point of constituting satisfying optimization inputs, thereby preventing to make the most
out of this concept.
A detailed Finite Element Analysis (FEA) was carried out to guarantee that the structure withstands
the loads at which is going to be subjected and to assess its dynamic behavior. Wind tunnel balances are
necessarily very stiff, but it is difficult to avoid oscillatory movements of the model. Those vibrations induce
inertial loads on the balance, and might have a negative impact on the quality of the measurements. These
inertial forces are not desired, but measured together with the aerodynamic loads.
It is possible to do some rough calculations to show the true magnitude of these forces; For a steady
acceleration, Newton’s equation of motion can be expressed as

x  x0
1 2
v0 t at (3.1)
2
from which it follows that the acceleration experienced by the test model is

a
2∆x
(3.2)
t2
where ∆x is the amplitude of the oscillation, and t the time it takes for the model to travel the oscillation
amplitude. Thus,
a  1 2  32∆xf 2
2∆x
p 4f q (3.3)

Assuming that the test article oscillates with an amplitude of 1mm, and at a frequency of 15Hz, the oscillation
of the model produces accelerations in the range of 7.2m{s2 . If the mass of the model is 500g, the induced
inertial forces are in the order of 3.6N p367g q, which is the same order of magnitude as the values that are to
be measured. For this reason, it is of paramount importance to make sure that the natural frequencies of the
structure are well above the expected range of frequencies of the airflow.

3.2 Signal Processing System

Retrieve useful data from a set of sensors requires a framework that is able to comprehend and decode the
information they provide. Consequently, the performance of the balance highly depends on the quality of their
output, as well as on the hardware which receives and processes the signal. Essentially, the design of the signal
conditioning system must be oriented towards making the most out of the output of the sensors. Therefore,
it is of paramount importance to clearly describe it in terms of the measuring range, the type of excitation
(AC or DC), the noise density spectrum, and the number of channels.
Signals may be processed using full analog techniques, full digital methods, or a combination of both
named Mixed Signal Processing (MSP). In this particular case, the choice was clear given the analog nature
of the force transducers that have been considered. There are several ways of achieving MSP, though in most
cases, the main difference between those methods has only to do with the placement of the Analog-to-Digital
Converter (ADC) in the conditioning chain. The schematic presented in figure 3.4 gives a general overview

18
of the signal processing chain divided into several smaller steps according with the nature of the manipulated
signal.

Figure 3.4: Signal Processing Chain

3.2.1 Data Aquisition System (DAQ)

Strain Sensors

Previously (section 2.4), we addressed the many forms of force transducers usually seen in wind tunnel balance
applications. The type of sensor as to be chosen as a function of the measurement requirements, and conditions
under which the balance is going to operate. Figure 3.5 establishes a qualitative comparison between the
sensing technologies, aiding to determine which one of them better fits this project. Another factor that must
also be taken into account is the electrical character of the sensor and its output. They all require different
types of signal conditioning, which might have a major impact on the complexity and cost of the equipment.
Frequency
Response

Stability

Sensitivity

Linearity

Cost
Metal-foil Strain-Gage (SG)
Fiber Bragg Grating (FBG) based SG
Piezoelectric sensor
Semiconductor SG

Figure 3.5: Summary of the Strain Sensor Technologies Features

FBG-based sensors, given the physic phenomenon under which they operate, are the most expensive to
implement. The commercial acquisition modules are very complex equipments and their reproduction is out

19
of question. For these reasons, despite its good stability and linearity features, it is not a viable option.
The piezoelectric cell has a straight forward signal conditioning circuitry which could be easily developed
in-house. Unlike the other options it does not need a spring element for support, since the cell itself consists in
the whole force transducer, and it has an intrinsic high stiffness. These singular features allow balances with
high stiffness designs, providing a good frequency response. On the other hand, piezoelectric sensors are not
suitable for long term measurements, due to its drifting ‘’tendency”. Therefore, their application field usually
complements the use of metal-foil and semiconductor strain-gages, which are limited to the low-frequency
range.
Semiconductors based SGs are the least expensive ones but the most sensitive, and enjoy the same
frequency behavior and stability as the metal-foil sensors. This technology would then be the the most
suitable if it was not for its lack of linearity. This issue could be overcome by means of a more complex
reduction system, however, it is not clear if its advantages would balance potential complications when simple
solutions are paramount in this project. Ultimately, metal SGs have been chosen for their linear behavior and
extensive use in several applications at the university, such as tensile and vibration tests.

Bridge Circuit

As previously explained, it is possible to determine the strain of the spring element by measuring the change
in resistance of metal SG. The Wheatstone bridge circuit allows to do just that. It is a highly sensitive mean
of detecting and measuring the small resistance changes, which characterize SGs. The full- and half-bridge
circuit configurations are given in figure 3.6.

(a) Full Bridge (b) Half Bridge

Figure 3.6: Wheatstone Bridge Configurations

The four branches of the bridges are formed by the resistors R1 to R4 . If nodes 2 and 4, the so-called
excitation diagonal, are connected to the excitation voltage UE , then a voltage, UA appears between the nodes
1 and 3, the so-called measurement diagonal. The value of the output voltage depends only on the ratio of
the resistors according to the following equation.
UA
UE
 pR R1 RR3
qpRR2 R4R q (3.4)
1 2 3 4

If the bridge is balanced, then R1 =R2 =R3 =R4 and

UA
UE
0 (3.5)

On the other hand, if the value of the resistors changes unevenly, Ri  R0i ∆Ri , the bridge becomes
unbalanced. With the additional assumption that ∆Ri Ri it follows,
UA
UE
 41 p ∆R
R
1

∆R
R
2 ∆R3
R3

∆R
R
4
q (3.6)
1 2 4

20
 From equation 2.2, K  ∆R
R , so the output voltage after such change is
0

UA
UE
 K4 p1
2 3
4 q (3.7)

Considering, for the moment, the application of the full Wheatstone bridge configuration to the force
transducers of the balance. The spring elements are going to be subjected essentially to traction or compressive
axial loads as it is shown in the schematic of figure 3.7. Then, the strain in the axial direction is given by
1  σ1 {E and the transverse strain given by 2  υ1 - where σ1 is the stress in the axial direction, E is the
Young modulus of the elastic element, and υ the Poisson ratio. This is also true for strain-gages number 3
and 4, respectively.

Figure 3.7: Complete force transducer

Introducing, υ  0.3 and since 1  3 and 2  4 , the net signal is given by

UA
UE
 K4 p1
p
0.31 q 1
p
0.31 qq (3.8)

UA
UE
 K4 B1 (3.9)

B is commonly called the bridge factor and it is equal to 2.6.

The choice of the best suited configuration to implement, is obviously dependent on the measuring
problem in hand. Force transducers generally have to comply with more demanding accuracy requirements
than those of stress analysis. Specially in this project, the sensitivity must be maximized to improve the
signal-to-noise ratio therefore, all sensing elements of the wind tunnel balance comprise a full-bridge circuit
with similar active gages in all four branches, as it is shown in figure 3.6a. Otherwise, if the half bridge
configuration was used, the B factor would be 1.3, halving the sensitivity. Furthermore, it would only be 1 in
the case of the quarter bridge solution.
There are other inherent advantages in using this kind of configuration. Full and half-bridges are
temperature insensitive. As it fluctuates, the material of the spring element expands or compresses equally
in all directions causing the value of the gages to vary evenly, thereby keeping the bridge balanced and
consequently its output signal, UA , equal to 0. The full Wheatstone bridge has yet another advantage. It is
also able to compensate traces of bending loads that might be superimposed by the friction on the bearings of
the rod-ends. Equation 3.7 shows that the absolute strain values are subtracted if they appear with opposite
signs in oposite bridge arms. Consequently, signals resulting from the normal force are added and signals from
the bending load compensate one another.

21
DAQ Hardware

The data acquisition process can be broken up into two steps: the first one takes place in the analogical
domain and it consists of exciting the bridge and adjusting its output signal to the input of the second step
that is carried out by the Analog-to-Digital Converter (ADC). This device converts a continuous voltage signal
to a digital number with a given bit resolution. The higher the number of bits used to describe the input
signal, the better the resolution is.
There are commercial solutions with all the signal conditioning required to power and measure bridge-
based sensors simultaneously such as the NI 9237 developed by National Instruments. This module is able
to provide a very stable Direct Current (DC) excitation voltage up to 10V and it has a built-in full and half-
bridge measuring system with 24-bit resolution ADCs on four inputs. In addition to these basic features it
can also perform offset/null and shunt calibrations, remote sensing and it has USB connection to simplify the
communication:

• Shunt Calibration - This is an indirect calibration method that involves simulating the input of strain
by changing the resistance of an arm of the bridge, shunting it with a larger resistor. Due to its
simplicity, this is the normal procedure for verifying or setting the output of a SG instrument relative to
a predetermined mechanical input at the sensor.

• Remote Sensing - It continuously and automatically corrects for errors in excitation leads. The resistance
in the wires that connect the bridge excitation diagonal causes a voltage drop, which is a source of gain
error.

The NI DAQ has been specially designed for acquiring bridge signals so it provides lots of very interesting
tools which, at a closer look, may not be particularly suited for this application. Firstly, shunt calibrations are
most useful with quarter-bridge sensors because there may be significant resistance in the wiring to the active
resistor in the bridge. Moreover, a full-bridge shunt calibration is considerably harder to perform since all
resistors have to be bypassed in such a way that the behavior of the spring element is properly simulated. It is
then necessary to know beforehand the ratio of transverse to axial strain in order to pick the right combination
of shunt resistors. Secondly, given the operating principal of this balance, it would be virtually impossible
to calibrate it using indirect methods. In the case of transducers instrumentation, direct calibration methods
are preferable, that is, known mechanical inputs are applied to the transducer and the instrument sensitivity
is adjusted as necessary to register the corresponding output. Finally, direct calibrations estimate the overall
gain of the sensor rendering pointless, as a result, the remote sensing functionality for this application.
Commercial integrated solutions, like the mentioned National Instruments module or other equivalent
products, for instance, the USB-2404-UI by Measurement Computing, can easily cost 1200 e and each of them
is only able to monitor four bridges at the same time. The balance comprises a total of six force transducers so
it would be necessary to purchase two of them in order to fulfill the requirements, thereby drastically increasing
the overall cost of the project. The alternative solution is to built, from scratch, a customized DAQ, which
aims to fulfill only the basic balance requirements following a reduce-to-essentials philosophy. This approach,
even though it increases the complexity of the project, seems to be the most cost-effective solution.
To achieve this goal two methods were considered, namely, the traditional analog front-end design and
the integrated circuit solution - both follow the conditioning chain shown in figure 3.4: The former uses a high
resolution ADC as the core element in the process. This component has to be surrounded by a considerable
amount of analog signal conditioning circuitry, such as gain stages, to enhance the transducer output quality.
When designing gain stages, several factors have to be taken into account. The first and most critical is
noise, both in the measurement environment and in the components that constitute the high gain stage.
Other causes for concern include its common-mode rejection ratio, the offset in the amplifiers along with the
drift performance. Consequently, the performance of said circuitry, rather than the high resolution converter,

22
often determines the system success in measuring low level signals. A fully integrated solution, on the other
hand, incorporates the required signal conditioning, filtering, and ADC on a single chip. This enables a direct
transducer interface without the need for a front-end signal conditioning, thereby reducing the complexity of
the analog circuit design and layout. It also offers a better control of specifications and error budgets than
those of discrete solutions.
Given the clear advantages of using integrated chips, all the DAQ has been designed based on its
application. The chosen architecture consists of a distributed system with each Slave monitoring a force
transducer, hence six slaves are required. The Slave module is a Printed Circuit Board (PCB) responsible
for the excitation and monitoring of each sensor at a given programmable rate. The Master module, in
turn, communicates with all Slaves through Serial Peripheral Interface (SPI). Its main function is to set the
pace at which raw data is acquired, receive it and provide it to the DRS through Universal Asynchronous
Reciever/Transmitter (UART) communication protocol.

3.2.2 Data Reduction System (DRS)

The DRS can be found at the end of the data acquisition chain, see figure 3.4. It is a software, responsible
for post processing the acquired raw data and reducing it to the aerodynamic coefficient form. Concerning its
programming language; it affects the development speed and freedom, but also the final performance of the
application. The chosen ‘software package’ must, above all, include libraries to access serial ports and have
graphical libraries that allow the creation of a fairly complex graphical interface.
The options taken in consideration were LabVIEW, C++ and MatLab. The first one is commonly used
for data acquisition, instrument control, and industrial automation on a a variety of operating systems. The
data reduction of the National Instruments acquisition board, addressed in the previous section, is performed
by a LabVIEW integrated software named LabVIEW Signal Express. It is an intuitive application that allows
to work with all the features of the acquisition board in a user-friendly environment. LabVIEW is mainly
a graphical programming language, which makes it simple to create small applications, however, complex
algorithms or large-scale codes may require an extensive knowledge of the special syntax and the topology of
its memory management. Otherwise, small changes can force deep restructures of the whole code, making it
harder to build an application from the ground up. C++, for its turn, is a high-level language with low-level
features achieving a good performance and versatility. MatLab also manages all the requisites and includes
a complete graphic library, although it tends to become heavier speed-wise. Nevertheless, it is an intuitive
language and it provides a large database of built-in algorithms as well as a complete technical documentation.
MatLab is also part of the curricula of the IST Masters in Mechanical Engineering. For these reasons, it has
been chosen as the base programming language for the DRS.
To accurately compute the aerodynamic coefficients, it is also necessary to monitor, in addition to
force transducers, airflow conditions, particularly its velocity. To do so other peripheral sensors are necessary,
such as, differential pressure transducers, digital accelerometers and compasses to measure roll, pitch and yaw
attitudes of the model. These equipments are not part of this system, however it cannot properly work without
them.

3.3 Static Calibration

“[...] Let us start by making clear the immensity of a wind tunnel balance calibration: with a competent
crew the first calibration of a new balance will take three months at least. This time frame supposes that
adequate shop facilities for sorts of changes are available and recognizes that the first ‘calibration’ will almost
inevitable involve many adjustments. Many balances have, however, subsequently served very well for decades

23
with modest additional calibration efforts. [...]”. In Barlow et al. (1999)
As mentioned in section 3.2.1, the direct calibration process, which suits this device, involves the
application of known calibrated weights to the balance and recording strain-gages readings at each force and
moment combination. There are a number of different sets of equations that can be used to represent the
calibration data, but most of them assumes that the balance output voltages are functions of the calibration
coefficients and the applied loads (Leung & Link, 1999).
For the six-component balance, the calibration is always described by six different equations, but the
number of terms in each of them can vary depending on whether the equations are first-order, second-order
and so on. The calibration procedure using higher order approximations is significantly more complicated since
it requires the application of two or more load components simultaneously. Moreover, the calibration of the
balance is a manual process naturally prone to random errors, so more accurate descriptions of the balance
behavior can actually lead to poorer results. For a well-designed balance, second- and third-order terms are
generally small so, under these circumstances, a first-order calibration has been carried out in this project.
After acquiring all calibration points, the full decoupling matrix of the balance can be computed and its
uncertainty evaluated following the method proposed by Reis et al. (2013): the data points are interpolated by
a fitting function and the uncertainties of both the applied loads and the readings of the sensors are considered
in the regression.
Usually, it is strongly recommended to do the calibration with the balance in place. This would ensure
its alignment with the tunnel, and thus, having more accurate test results. In this particular case though,
such procedure is not as relevant since the facilities, where this device is going to be employed, may serve
other purposes which do not require its use. It may, therefore, be often installed in and removed from the test
section. For this reason, the calibration system has to be able to perform its task without the aid of reference
points provided by the test section. The framework of a ‘standalone’ static calibration system provides this
flexibility. In addition, it also eases its setup since the workspace is not confined to the test section.

24
Chapter 4

Mechanical Design & Manufacture

4.1 Requirements

Prior to the conceptual stage of the balance, it is important to clearly state its geometrical and functional
requirements:

1. For all the advantages it enjoys, the mechanical design should be based on a spatial parallel mechanism;

2. It should be able to allow an easy decoupling of the aerodynamic load into its six components without
large interferences;

3. Leg connectors should have the same length and round tube section;

4. The mechanism should be symmetric in relation to the vertical mid plane of the wind tunnel test section;

5. It should have a spatially isotropic geometry;

These requirements have been established towards maximizing the performance of the balance having in
mind the conjuncture of its development. Essentially, they are meant to ease the manufacture and maintenance
of the device, and are a result from a qualitative assessment of potential technical obstacles that could have
arisen during its development. Although at first thought, they are limiting the performance of the initial
concept, we find these to be necessary measures to ultimately enhance the potential of the prototype regardless
the adopted strategy be or not the best one. If all these requirements are fulfilled then all the others, end-user
oriented requisites mentioned in Product Scope (section 1.3), will also be achieved.
Engineering this prototype, in particular coupling both the mechanical and electrical systems, demanded
a series of iterations that are not described in this document. As such, even though the mechanical design is
covered in the first place, there is not any parallelism between the development cycle and the order of how
contents are presented in the next chapters. Take the force transducer design, for instance. It could have been
easily covered, in part, at this stage since it has a mechanical element. Yet, this subject is only addressed in
the next one as part of the signal conditioning chain. Consequently, for clarity reasons, subsequent chapters
are often mentioned.

4.2 Mapping Matrix Expression

In the conceptual design stage, a ‘best’ design is selected among alternatives. The 3-3 octahedral structure
was the first to be considered since it is the simplest structure of its kind.

25
 Grübler established a criterion to compute the mobility of general spatial mechanisms, in Tsai (1999)

¸
j
M  N pn
1q
pN
fi q (4.1)

i 1

M is the mobility or number of degrees of freedom of the system.

N , number of degrees of freedom of the space in which a mechanism is intended to work, N  6.
where, n, number of links in the mechanism, including the fixed link.
j, number of joints in a mechanism, assuming that all the joints are binary.
fi , degrees of relative motion permitted by joint i.

Figure 4.1: Octahedron in the central configuration, plan view

Figure 4.1 illustrates the plan view of a 3-3 parallel mechanism with an equilateral triangular base of side
b, and an equilateral triangular moving platform of side a. Each leg has ball-joints at both ends and is treated
as 2 bodies connected by a prismatic joint, thereby simulating the elastic behavior of the force transducers.
Summing up, there is a grand total of n  14 bodies, so for this case, the mobility equation becomes

¸
12 ¸
6
M  6p14
1q
p6
3q
p6
1q  78
36
30  12 (4.2)

i 1 
i 1

Six out of the twelve degrees of freedom are trivial since each leg connector can rotate about its own axis.
By eliminating these from consideration, the number of degrees of freedom is reduced to six, proving that
in-parallel mechanisms provide a mean to determine the aerodynamic load components by sensing the axial
load acting on each leg.
The static equilibrium equation for a parallel mechanism with six leg connectors can be written as

ŵ  f tS; S0 u  f1 tS1 ; S0L1 u f2 tS2 ; S0L2 u ... f6 tS6 ; S0L6 u (4.3)

where f is the magnitude of the external wrench and tS; S0 u are the Plücker line coordinates of the
screw associated with the wrench. fi represents the reacting force produced on the i
th elastic leg and
tSi ; S0Li u are the Plücker line coordinates of the line along connector i. The latter equation can also be
26
expressed in matrix form as follows

 
f1
f 
   
 2

S6 f3 

S
f  S1 S2 S3 S4 S5
S06 

 (4.4)
S0 S01 S02 S03 S04 S05 f4 
 
f5 
f6

   
J  S1
S01
S2
S02
S3
S03
S4
S04
S5
S05
S6
S06
 JF
JM

J is a 6  6 matrix formed by the Plücker coordinates of the lines along the six leg connectors. JF is the force
mapping matrix and JM is the torque mapping matrix. A derivative of equation 4.4 will yield a relationship
between the changes in individual leg forces to the change in the externally applied wrench. The details of
obtaining the transformation matrix from equations 4.4 are given in appendix B.
Note: The coordinates of a line joining two finite points with coordinates px1 , y1 , z1 q and px2 , y2 , z2 q,
Ŝ1 , are given by the second-order determinants of the array
 
1 x1 y1 z1
1 x2 y2 z2

therefore,
 T
Ŝ1  x2
x1 y2
y1 z2
z1 y1 z2
y2 z1 z1 x2
x1 z2 x1 y2
y1 x2 (4.5)

That is, the first three Plücker coordinates describe the direction of the line and the last three coordinates
describe the plane containing the line, the origin and the distance from the origin to the line.

4.3 Singularity Analysis

4.3.1 Singular Geometries

If a parallel structure is well designed, the external wrench applied is statically balanced by the six leg forces
and these are linearly independent. If not, then the system matrix J is singular and it can not sustain any load.
Even when the sensor is near singularity, J is ill-conditioned and would result in highly uneven distribution of
leg forces in response to loads in certain directions.
Lee (2000) analyzed the singular configurations of these devices and defined the problem of ‘closeness’
to a singularity by defining what is known as the Quality Index (QI), λ, for in-parallel devices. The quality index
is the ratio of the determinant of the matrix formed by the Plücker line coordinates of the six leg connectors
of the platform in some arbitrary position to the maximum value of the determinant that is possible for the
in-parallel mechanism. It is then a dimensionless value with a maximum of 1 at an optimal configuration and
a minimum of 0, indicating an uncontrollable singularity state.

λ
|J |
|J |m (4.6)

Lee (2000) showed that for a 3-3 platform with a top equilateral triangle of side a and a similar bottom
triangle of side b, the maximum determinant of J occurs when the top platform is parallel to the base and is

27
rotated as illustrated in figure 4.1. The determinant of the matrix J is given by
?
3 3a3 b3 h3
|J |  (4.7)
4r pa
3bq h2 s
2

b to the base. In addition, Lee proved that the maximum

where h is the height of the top platform with respect
value for the determinant of J occurs when h  13 pa
bq2 and b  2a, therefore

b  2a  2h (4.8)

Although the geometric configuration of the 3-3 parallel platform has the simplest geometry, it is a
concept hard to implement in practice due to the pair of concentric spherical joints, which are not only very
hard to implement but also induce unwanted interference between legs.
There are other alternatives to this type of configuration. One of them is the 6-3 platform, which
has six points of connection on the base and three on the upper platform. This solution despite having less
concentric joints, it still shares the same difficulty of arranging the joint connections on the top platform. So,
in order to completely eliminate the need for coincident connections, the 6-6 in-parallel mechanism has been
considered, also known as the classical Stewart platform. See figure 4.2.

Figure 4.2: Spatial 6-6 in-parallel mechanism, plan view

The moving platform joints are separated by αa where 0 ¤ α ¤ 1 and the base joints are also separated
by βb where 0 ¤ β ¤ 1 from their original double joints. The top platform is raised to a distance h and each
of the six legs, EAi Ai , like the previous configuration are treated as two bodies connected by a prismatic joint.
By following the same procedure as for the 3-3 parallel mechanism, Lee (2000) showed that all the leg
lengths, which are the same for all six, and the determinant of the Jacobian matrix are
c
l pa p3α2
3α 1q abp3αβ
1q b2 p3β 2
3β 1q 3h2 q
1 2
(4.9)
3
and ?
81 3a3 b3 h3 p3αβ
2α
2β 1q3
|J |  4pa2 p3α2
3α 1q abp3αβ
1q b2 p3β 2
3β 1q 3h2 q3
(4.10)

Note that |J |  0 when 3αβ
2α
2β 1  0. This relationship always yields similar base and top
platforms and some of these cases are illustrated bellow in figure 4.3.

28
 Figure 4.3: Configurations with singular matrix of line coordinates, from Lee (2000)

By differentiating J with respect to h, the |J |m occurs when

c
h pa p3α2
3α 1q abp3αβ
1q b2 p3β 2
3β 1qq
1 2
(4.11)
3
then substituting equations 4.11 and b  γa into 4.10 yields

27a3 p3αβ
2α 1q3

|J |m  32p 1 p3α2
3α 1q γ1 p3αβ
1q p3β 2
3β 1qq3{2
(4.12)
γ2

The absolute maximum value of |J |m can be determined by taking the derivative of equation 4.10 with
respect to γ which leads to the relation

 ab 
6α3αβ
6α1
2
2
γ (4.13)

Substituting equation 4.13 into 4.12 results in

?
|J |am  a p3α2
3α 1q3{2
3 3 3
(4.14)
4
|J |am? depends only on the value of α, it is independent of β.? When α  0 or 1 it takes the maximum value
of 3 4 3 a3 and when α  21 , it becomes minimum, |J |am  3323 a3

4.3.2 Prototype Geometry

After choosing the desired configuration and addressing the necessary condition for structural integrity of
in-parallel structures, λ ¡ 0, it is possible to finally specify in more detail the physical dimensions of the
mechanism, summarized in table 4.1.
The notation commonly adopted to describe classical Stewart platforms is illustrated bellow in figure
4.4. The frame of reference, Oc
XY Z, has its origin located at the geometrical center of the upper platform.
The X axis is perpendicular to the line connecting the two joints U1 and U6 ; The central angles between u1

29
and u3 and between u3 and u5 are 2π {3, and the central angles between U1 and U3 and between U3 and U5
are also 2π {3;

Table 4.1: Geometry of the Balance

Symbol Dimension Units
R1 275 mm
R2 150 mm
Initial
ϕ1 110 
ϕ2 10 
l 287.3 mm
Best Hc 203.2 mm
λ 1

Figure 4.4: Classical Stewart platform notation

The first set of rows specify the radius of the lower and upper platforms, R1 and R2 respectively, as
well as the positions for the pick-up points of the leg connectors, defined by the central angles ϕ1 and ϕ2 .
These dimensions were initially established by weighing the size of possible test articles and their estimated
generated load against the stability of the device itself. Then, based on the optimal dimensional study of 6-6
in-parallel manipulators described in the previous section, the length of the connectors and distance between
both platforms is calculated to achieve optimality. Equations 4.9 and 4.11 yield the values of l1 and Hc1 ,
respectively.

4.4 Isotropy Analysis

Section 3.1.2 addressed the impact of spatial isotropy on the performance of this kind of structures. In a nut
shell, isotropic modes achieve uniform measurement sensitivities yielding approximately the same accuracy for
all force components, which is clearly a highly desired feature.
As the mapping matrix J has three dimensionless rows and the other three with linear dimensions,
the condition number would be a ratio of two quantities of different dimensions. As such, in general, it
is not possible to derive a condition number which is a true representative of the isotropic behavior of the
transformation associated with the matrix. For this reason, usually force and moment isotropy’s are accessed
separately. A sensor configuration is considered to be force isotropic, if the following condition is satisfied.

JF JFT  m1 I3x3 (4.15)

Similarly, moment isotropy is achieved if

T
JM JM  m2 I3x3 (4.16)

Therefore, the sensor configuration is considered to be spatially isotropic if the mapping matrix J satisfies the
following condition    
JF JFT JF JM T
JJ   0 mI
T m1 I3x3 0
(4.17)
JM JFT JM JM T
2 3x3

30
where m1 and m2 are scalars and I3x3 is the identity matrix.
Yao et al. (2011) showed that for the classical Stewart platform-based sensor, figure 4.4, the matrix
JJ T can be expressed as
 
F1 0 0 0 R 0
0 F 0
R 0

 0 
 1

0 0 
JJ     1
T 0 F2 0 0

 L2 (4.18)
 0 R 0 M 0 0 
 
1
R 0 0 0 M1 0 
0 0 0 0 0 M2

F1 3pL2
Hc q; F2  6Hc2 ; M1  3Hc2 R22 ; M2  3R12 R22 p1
cospϕd qq

b
R  3Hc pR1 R2 cosp ϕ2
R22 qq; L  R12 R22 Hc2
2R1 R2 cosp ϕ2 q; ϕd  ϕ1
ϕ2
d d

where L denotes the length of the leg, and ϕd is the angle difference between ϕ1 and ϕ2 .
To guarantee spatial isotropy, the matrix JJ T must be a diagonal matrix, hence the expression of R in
equation 4.18 must be equal to zero. Furthermore, the following conditions must be satisfied. F1  F2 and
M1  M2 . Through algebraic manipulation it can be easily shown that these conditions can be reduced to
?
 3sinp
ϕd
2
q0 (4.19)

This equation implies both platforms to be aligned as ϕ1
ϕ2  0. Unfortunately, under this con-
dition, the classical Stewart platform becomes architectural singular, see figure 4.3. Therefore, it cannot
possibly realize a spatially isotropic configuration, and consequently, to provide a framework that maximizes
the performance of the balance, as initially expected.
In order to workaround this issue, Stoughton & Arai (1993) proposed a modified version of the classical
Stewart device with two different sized rings on the bottom platform, figure 4.5.

Figure 4.5: Modified Stewart platform

The modified solution illustrated above does improve the sensitivity isotropy of the balance, but such as
the classical configuration, it does not fully complies with the design requirements stated in the beginning of

31
this chapter. Both geometries are 6-6 in-parallel mechanisms, hence both satisfy the first criteria. With each
connector being composed of two spherical joints, the second requirement is also fulfilled. However, when it
comes to the last three criteria, the solutions at stake diverge. While the classical configuration is symmetric
and comprises leg connectors with equal lengths, it is not spatially isotropic. On the other hand, the modified
version provides sensitivity isotropy at the expense of having different leg lengths and an asymmetric geometry.
In this regard, a more extensive analysis should be made in order to access the real advantages of
isotropic configurations. Anyhow, given the goals and adopted design philosophy towards reliability, concepts
leading to simple solutions are best suited. For this reason, the classical Stewart platform described in figure
4.4 and table 4.1 has been chosen as the mechanical configuration for the balance. Since all legs have the
same length, it is less prone to manufacturing errors and it eases the replacement of damaged parts. The
calibration process is also simplified, since it has a symmetric geometry its behavior is more predictable.

4.5 Structural Analysis

Having established the geometry of the balance, its structural integrity and dynamic behavior can be analyzed.
Figure 4.6 shows a picture of the mechanical assembly; the adjustable vertical strut, already attached to the
upper platform, is 300mm long at most. Its function is to place the model at the center of the wind tunnel
test section, where the velocity profile is more uniform. The technical drawing of the assembly is available in
appendix B.
A set of static FEA were carried out with ANSYS APDL. Even though several iterations have been
made, only the final design and corresponding results are given in this document. Finally, a modal analysis was
performed to identify the inherent natural shapes and frequencies. It is of paramount importance to assure
they are not within the estimated vortex shedding frequency spectrum - section 1.3.3.

Figure 4.6: Balance Structure

32
 The material and dimensions of the sensing elements were specified in the first place. These features
have a major impact on the performance of the balance, mechanical and electrical wise. Specifically, a key
aspect should be held in mind; the accuracy of the acquired data is tightly related with the stiffness of the
structure. If its geometry changes under load, so does the force transformation matrix, thereby drifting from
the one estimated through the calibration process. In addition, the induced structure compliance causes the
attitude of the model to change in relation to its initial setup. On the contrary, if the frame is built very stiff,
the frequency response of the balance is privileged over its sensitivity, damaging the signal-to-noise ratio.
Ideally, the elastic members would be the only cause for compliance; their stiffness would be the
highest possible that fulfills the accuracy and resolution requirements, and all other parts would be infinitely
rigid. In practice, however, stiff designs lead to heavier solutions, which induce higher inertial loads on the
balance. In this regard, the bottom square platform was not an issue, since weight improves the stability of
the understructure. The core concern is with the suspended frame, which should be as light as possible to
minimize inertial loads and, at the same time, stiff enough to minimize deformations.

4.5.1 Sensing Elements

The chosen material for the leg connectors is aluminum. Among the most used materials for force sensing, it
has the particular properties of high thermal conductivity, low density and relatively low stiffness. Aluminum
also provides good surface features to glue the SGs, in addition to its good workability and market availability.
The section of the rods should have, at least, the minimum dimensions that assure the structural
integrity of the mechanism, and secondly, they must allow a proper attachment of the SGs to the surface.
Electrical gage sensors are available in a wide range of sizes and configurations covering a variety of applications.
Smaller footprints are suited to measure local strains and larger footprints average the value over a wider area.
As it is expected an uniform strain on the rods, the smaller ones would be suitable, but having in mind their
future installation and visual inspection, the medium sizes have been considered (see table 5.1).
The force transducers are meant to measure forces in the direction of their axis, and at the same time,
to mitigate bending strains resulting from eccentric loadings and frictional moments. To do so, each one has
been fitted with a full Wheatstone bridge placed in accordance with figure 4.7 - SGs must be symmetrical
both with regard to their position and specifications.

Figure 4.7: Bar fitted with longitudinal and transverse SGs

Considering the tubular shape of the spring elements, their minimum allowed diameter can be estimated
as follows c
8pLSG ∆P q
Dmin  (4.20)
π
where, LSG is the length of the SGs and ∆P the gap left in between.
The longest gage among the candidates has 13.3mm; thus, if a gap of 5mm is left in between, it yields
a minimum of I7mm rounded up. Ideally, since the outer diameter is already set, the wall thickness of the
elastic members would be just enough to guarantee the structural integrity and compliance limit. However,
a quick market search shows that these are not standard dimensions and there are not many drawn tube

33
manufacturers specialized in small quantity production. As a result, prices are usually very high. Precision
Tube Production and GoodFellow provide, among other forms, an extensive range of small diameter precision
tubes in stock. The following table summarizes a few options available close to the minimum allowed outer
dimension.

Table 4.2: Drawn tubes - price comparison

Seller OD rmms Thickness rmms Length rmms Area rmm2 s Price [e]
Specialized 9 0.5 1000 13.4 76.2a
shop 12 0.5 1000 18.1 74.2
Non-specialized 8 1 1000 22.0 3
shop 10 1 1000 28.3 3.5
12 0.5 1000 18.1 2.5
a If 5 units were bought

Obviously, this project could not possibly afford to spend over e350 just on aluminum tubes, so the
solution was to look for equivalent alternatives in non-specialized shops where prices are incomparably smaller.
As shown in the bottom of table 4.3, the section dimensions are roughly the same, but these stores do not
provide any information regarding the aluminum alloy nor the manufacturing tolerances. Despite not being
beneficial, this lack of information is not harmful either. The expected loads are not substantially high, and if
necessary, it is possible to run tensile tests in order to access its mechanical strength. In the end, the 12mm
outer diameter rod has been chosen for having the smaller section area.

4.5.2 Finite Element Analysis (FEA)

To expedite the structural analysis of the balance a simplified model has been considered - figure 4.8. The
leg connectors have been modeled by 1D-link elements (LINK180 ), and the vertical strut and top platform by
4-nodes shell bilinear elements (SHELL181 ).

Figure 4.8: FEA model - boundary conditions

34
 The concern for a high stiff and lightweight suspended frame lead us to consider the use of composite
materials, namely fiber compounds. There are several types of fibers with very different features and applic-
ations. Ferreira (2013) gives a short and clear description about fiber-reinforced polymers, covering some of
the manufacturing processes, their properties, and the commonly used failure criterion. Particularly in this
project, carbon fiber has been used given its high strength to weight ratio and availability.
The layup of the upper platform is the following: [ 45 /
45 / AIREX C70.7510mm /
45 / 45 ].
It comprises a 10mm thick closed cell foam as core (AIREX C70.7510mm ); this layer increases the second
moment of area, thus improving the flexural rigidity of the part. On either sides, a thin layer of bi-diagonal
carbon fiber provides the necessary strength.
The initial design of the vertical strut planned for a carbon fiber one, sectioned in three parts with
successively smaller diameters to reduce flow interferences (see figure 4.6). This solution turned out to be
insufficiently rigid, inducing large relative displacements of the model at higher loads. For this reason, the final
design uses an aluminum round tube instead, with I30mm and 2mm thick. The detailed material properties
can be found in appendix B.

Deformation

Figure 4.9a shows the deformation contour and shape of the balance under the worst loading case scenario -
testing the FSAE Series 1:5 model at 40m{s, FX  30N and FZ 
130N . The forces are applied at the
tip of the vertical strut, where the model is attached.
At first glance, most of the displacement at the free-end of the beam comes from its own deflection,
3.3mm, and the layup of the upper platform seems to provide a fairly rigid base, deflecting less than 0.5mm.
However, a scrutiny to the deformed shape of the balance reveals a different behavior; in fact, rather than the
deflection of the vertical strut, the upper platform is responsible for most of the observed drift. Figure 4.9b
shows the contour of the nodal rotation vector sum, in radians. As it can be seen, the model support nodes
display approximately the same rotation along its length showing that it does not bend as much as it tilts.

(a) Displacement vector sum (b) Rotation vector sum

Figure 4.9: FEA results - static analysis

35
 The reason behind this tilting motion lies on the deformed shape of the upper platform; the bending
moment at the bottom of the cantilever beam, associated with drag or any other force lying on the XY
plane,
causes the anchoring region to deform as illustrated in figure 4.10. The nodes in this area are rotated roughly
0.55 inducing a displacement of 300 sinp0.57q  3.1mm at the mounting point of the model. The remaining
drift is due to bending and it can be easily verified by equation 4.21, which expresses the tip deflection of a
cantilever beam, δ, with an end load, P .
3
δ  P3EI
L
; I  π4 pRe4
Ri4 q (4.21)

where, E is the modulus of elasticity, l its length, and I the moment of inertia of a round tubular beam with
an outer radius Re and inner radius Ri .
Considering the dimensions and material from which the vertical strut is made of, its maximum estimated
deflection, under the given load case, is about δ  0.2mm. The total displacement is then a result of the
added effect of the mentioned deformation mechanisms, hence 3.1 0.2  3.3mm.

Figure 4.10: FEA results - static analysis detail, rotation vector sum

Failure Criterion

Aluminum is an elastic and isotropic material, so the principal stresses of the stress tensor can be easily
identified allowing to compute the von Mises equivalent stress as follows.

pσ1
σ2 q2 pσ2
σ3 q2 pσ1
σ3 q2 2

σyield (4.22)
2
If this condition is not verified then the structural integrity is compromised.
In the case of orthotropic materials the principal stress directions do not coincide with the orthotropic
directions. Therefore, the von Mises criterion is no longer suitable and a different one must be considered. The
Tsai-Wu criterion for the case of plane stress has been used in this project, mainly due to its good agreement
with test results and for being available in ANSYS APDL; its formula is given by equation 4.23.

C11 σ12 C22 σ22 C66 σ62 C1 σ12 C2 σ22 2C12 σ1 σ2 1 (4.23)

where,

C1 σ 1

σ 1
; C2 σ 1

σ 1
; C11 σ 1
;
l_tension l_comp t_tension t_comp l_tension σl_comp

C22 σ 1
C12  12 ?σ 1
; C66  τ12
t_tension σt_comp l_tension l_comp σt_tension σt_comp
σ lt

36
 (a) von Mises eq. stress - vertical strut (b) Tsai-Wu index - upper platform

Figure 4.11: Failure criterion

As expected, the region around the contact between the cantilever beam and the upper platform is
the most critical one for either parts, figure 4.11. Still, at a closer inspection the former displays a maximum
vonMises equivalent stress of only 97M P a, which is well below the aluminum yielding strength, and the latter
a maximum Tsai-Wu index of 0.11, also reasonably lower than the rupture point. In addition, it is important to
point out that the assembly between both parts is not properly modeled, leading to higher stresses than what
is actually verified. Figure 4.12 illustrates the platform-strut assembly; the vertical beam has an aluminum
insert at the bottom-end which significantly increases its foot size, thereby distributing the cleavage loading
over a wider area and thus reducing stress.

Figure 4.12: Detail of the platform-strut assembly

Modal Analysis

To access the performance of the balance under dynamic loading, it was necessary to identify the frequencies
associated with the vibration modes of the structure, and to foresee their impact on the load measurements
through a qualitative analysis. Two studies have been carried out, and for each one of them all shape modes
with natural frequencies below 200Hz were investigated. The first set of results, figure 4.13, corresponds to
the natural frequency modes of the structure alone. On the second one, figure 4.14, a lumped mass of 0.5kg
has been added on top of the vertical strut, simulating thereby the effect of having a test model attached.

Table 4.3: Vibration modes

Mode 1 2 3 4

Frequency rHz s
no model 63 65 177 220
0.5kg model 21 22 120 177

37
 (a) mode 1 (b) mode 2 (c) mode 3 (d) mode 4

Figure 4.13: FEA results - modal analysis without test model, displacement vector sum

(a) mode 1 (b) mode 2 (c) mode 3 (d) mode 4

Figure 4.14: FEA results - modal analysis with test model, displacement vector sum

The first two modes shapes seem to show the cantilever undergoing free vibration while the upper
platform stands still. At a closer look though, such as the static analysis, it can be clearly seen that this
motion is in fact driven by the mode shape of the upper platform and not only by the vertical strut deflection
- figure 4.15. Despite having a very small amplitude, these deformations cause the model support to oscillate
back and forth. As it was previously discussed, this induces negative inertial loads on the balance whose
intensity can easily top the aerodynamic loads that are being monitored. Particularly, the lower natural
frequency modes are the most prejudicial; not only they are within the expected vortex shedding frequency
range but also lead to larger displacements. According with the observed deformed shapes, the inertial forces
and moments driven by these modes and mixed together with the acquired signal are FX , FY , MX , and MY .
By its turn, the pattern of motion of the highest frequency modes causes the cantilever to oscillate up
and down, giving rise to inertial vertical loads instead. These, however, are not cause for concern since the
maximum estimated frequency of the vortex shedding is well below the necessary excitation levels for resonance
to occur.

Figure 4.15: FEA results - mode 2 detail

The increase of the lumped mass at the free end of the beam, as expected, significantly lowered the
natural frequency of the first two modes - from 63 and 65Hz to 21 and 22Hz, respectively. Their shape
resembles the first natural vibration mode of a cantilever anchored to an infinitely rigid support - figure 4.16.
In fact, it can be easily shown that they follow the same behavior. Equation 4.24 is an approximate formula

38
for the natural frequencies of systems having both concentrated and distributed mass, where k is the inherent
stiffness.

c
ωn  M
k
0.23m
; (4.24)

Figure 4.16: Uniform cantilever beam with lumped

mass, in Harris & Piersol (2002)

Firstly, consider for the moment an equivalent system which includes the vertical strut and the anchoring
region as well. Its stiffness is therefore given by the combination of stiffnesses of the mentioned subsystems.
Secondly, note that between the modal analysis carried out, the only difference is the increase of the lumped
mass at the tip of the vertical strut. All other features such as the geometry of the structure, the fiber layup
of the top platform, and material properties have been kept constant. Consequently, so as the stiffness of the
system in question. Since the stiffness does not vary, the reduction factor becomes a simple ratio of masses
given by

c c
0.02 0.23  0.152
r  M1
M2
0.23m
0.23m
 0.5 0.23  0.152
 0.32
which is in agreement with the decrease of the first modes natural frequencies.

4.6 Manufacture Considerations

The structural design of the balance was heavily influenced by the materials and processing technologies made
available by Projecto FST Novabase, to keep the production cost as low as possible. As such, some design
decisions that were taken specifically in this context, are not justified in others.
For example, the bottom platform is a composite part of carbon fiber, steel and structural foam, in the
following order [ 45 /
45 /Steel Sheet4mm / Rohacell IG7120mm /
45 / 45 ]. The steel sheet provides
stability by adding mass, and similarly to the top platform, the closed cell foam and carbon fiber layers provide
flexural rigidity and mechanical strength. These materials and processing techniques employed are not usually
oriented towards minimizing the cost, but rather increasing the performance of a given system, which in this
case does not fully apply.
Still, they allowed to benefit from a plain understructure. Otherwise, it would have been necessary to
explore more complex designs, such as steel or aluminum frames with ballast. These would most certainly
worsen the manufacturing tolerances, since they are an assembly of many parts, and do not provide the same
flexibility concerning the position of the leg connectors pick-up points (brackets), figure B.1a. Furthermore,
and most important, it made possible to reduce the number of manufacturing processes and the lead time. All
composite parts were manufactured by Vacuum Infusion Process (VIP) (see Ferreira (2013)) and then waterjet
trimmed to the desired final dimensions by TS Waterjet, an important sponsor of Projecto FST Novabase. The
brackets were also initially waterjet cut from an aluminum block, and subsequently, milled. The manufacturing
sequence is illustrated in figure 4.17.
The choice of the rod-end bearings also had a great impact on the performance and cost of the balance.
Plastic rod ends from IGUS were chosen over metal ones, figure B.1 Plastic produces lower friction, even when
running dry, are maintenance free, and the cost is incomparably lower (it was verified a budget reduction

39
of 850 e). Obviously, given the way they are manufactured (plastic injection), the dimensional tolerance is
inevitably worst; those used in this prototype are from the dimensional series E, according to DIN ISO 12240.
Since both the top and bottom platforms were waterjet trimmed, it was necessary to vacuum infuse
them in the early design phase, so that TS Waterjet had enough time to process them. Consequently, the
layup of the upper platform was established before performing the modal analysis, compromising at first the
dynamic performance of the balance, since the first frequency modes lie within the expected vortex shedding
frequency spectrum. It is, however, still possible to get around this misstep by reinforcing the anchoring region
of the vertical strut with backing plates, thereby mitigating the observed local deformation.

(a) Outline of the brackets after waterjet cut (b) Brackets after the multi-pass milling process

(c) Brackets final form

Figure 4.17: Manufacture of the bottom and top brackets

40
Chapter 5

Data Acquisition Design &

Manufacture

The acquisition system as to be designed according to the type of sensors employed (electrical SGs), the
conditions under which they operate, and data one wants to attain. As such, taking into account the arguments
of section 3.2.1, its requirements are:

1. The acquisition module should rely on Chip integrated solutions, as a mean to achieve reliability, and to
follow a Master-Slave configuration;

2. It should have at least six input channels;

3. It must be designed to fit the operating range of the balance;

4. It has to be a plug-and-play solution to improve functionality;

5. It should be capable of providing information regarding its operating condition and errors.

5.1 Strain Gage

Figure 5.1 illustrates the interface between the electrical and mechanical systems. The attachment of the
sensor to the elastic member is a critical aspect in this project leading to unsuccessful readings if poorly
placed. For this reason, the preparation of the bonding surface must be carefully addressed. The lead wire
system includes the wire from the gage to the tab, or solder terminal, and the heavier wire from the gage tab to
the instrumentation box. This facilitates faultless solder connections and protects the lead wires from tension
forces that could otherwise cause the SG to detach. Finally, the protective coating provides an efficient shield
against damage due to environmental factors, such as oxidation, which could in time affect the measurement
quality of the force transducer.

1. Specimen surface
2. Bonding adhesive
3. Grid carrier
4. Strain sensing grid
5. Lead wire system
6. Protective coating
Figure 5.1: The strain gage system, L Hannah & E Reed (1994)

41
 HBM provides a variety of strain gages that ease the installation and improve the output data quality
according to the application. The first step in selecting the best suited among them is to identify in which
direction the principal stress runs. The force transducers are going to be strained the most in their axial
direction, so linear and rosette SGs are possible configurations. Secondly, the size of the sensing elements
has to be chosen with regard to the strain field and space requirements. In this case, since it is expected an
homogeneous strain field, SGs with small grid lengths from 3 to 6mm are large enough. Nevertheless, the
medium sizes have been preselected to facilitate their correct positioning along the principal stress directions -
see appendix C. Table 5.1, summarizes the relevant features of the selected candidates, based on the mentioned
criteria.

Table 5.1: HBM strain gages features, HBM (n.d.)

no. Reference Type R 0 r Ωs Gage Factor [K] Grid Carrier rmms Umax rV s
1 1-LY43-6/120 Linear 120 2.07 13.9 x 5.9 8
2 1-LY43-6/350 Linear 350 2.07 13.9 x 5.9 15
3 1-XY13-3/120 T Rosette 120 2.07 10.5 x 8 5.5
4 1-XY13-6/120 T Rosette 120 2.07 23.3 x 10.5 12
5 1-XY13-3/350 T Rosette 350 2.07 10.5 x 8 10
6 1-XY13-6/350 T Rosette 350 2.07 23.3 x 10.5 20

(a) Linear (b) T Rosette

Figure 5.2: Considered configurations

Configurations such as the T Rosette, figure 5.2b, which guarantee the orthogonality between gages
by sharing the same grid carrier, are particularly suited to measure axial and transverse strain. As already
discussed, an ideal orthogonall full-bridge is able to cancel out the effects of temperature fluctuations, and
bending moments superimposed by the friction on the spherical pairs. So, even though, all options share the
same electrical features, the linear type ones have been put aside given the practical difficulties of a correct
installation.
The excitation level depends on the measuring task in hands. The change in output voltage for a
given level of strain increases in direct proportion to the excitation voltage. Increasing excitation voltage thus
improves signal-to-noise ratio, but a practical limit is reached when the ill effects of gage self-heating become
predominant. This is particularly important in this case of low-force transducers, where the thin section can
give rise to long warm-up times and apparent gage instability, regardless the good heat-sink properties of the
mounting surface. Based on empirical studies, HBM computes the maximum permissible excitation voltage,
Umax , to avoid incorrect measured values - the usually result is performance degradation rather than gage
failure, in HBM (May 2015a). They claim that the measurement error generally exceeds reasonable values for
a temperature gradient of ∆{d  0.75 C {mm, or higher, in the area close to the SG. Assuming that all the
electric heat generated is dissipated completely via a measuring body of infinite thermal capacity, then
c
Umax  RAλ
∆T
d
(5.1)

42
where, R is the SG nominal resistance and λ is the thermal conductivity of the measuring body.
The maximum excitation voltages provided on the catalog take steel as the material of the measuring
body, but since aluminum is going to be used instead, a correction factor CF should be applied to determine
the true excitation limits. c
CF   2.17
λal
(5.2)
λsteel
Given its value, the excitation limits more than double ranging from 11.9 to 40V . Nonetheless, as
a safety measure they have been kept slightly lower than the maximum rating to avoid self-heating effects,
at minus 3V . Based on the SG alone, the 6th option would be the most suitable candidate, however other
aspects must be taken into account. Most acquisition modules provide a 5V excitation source, and the
solutions capable of higher levels of excitation are substantially more complex and expensive to develop. For
this reason, the acquisition module has ultimately been designed to excite at 5V . Therefore, despite its low
excitation limit, the 3rd option is still one of the best candidates, as well as the 4th .
Both alternatives have suited electrical features and configuration. So the final decision focused mainly
on their size and price. Table 5.2, points out what sets these options apart; essentially, the higher the sensing
area, the higher the price. To instrument all force transducers with full-bridges a total of 12 units are required,
so 2 packages of 10 units have to be purchased, leaving the remaining 8 SGs as replacement parts. As a result,
the 3rd option turns out to be e75 cheaper, which is a significant amount given the tight budget assigned to
this project. In addition, the expected uniform strain distribution on the leg connectors renders the difference
between the measuring grid areas irrelevant at these sizes. So, under these circumstances, option 3 has been
chosen has the most suited strain sensor for this application.

Table 5.2: HBM strain gages Price

no. Reference Type Grid Carrier rmms Measuring Grid rmms Price [e/10units]
3 1-XY13-3/120 T Rosette 10.5 x 8 3 x 2.8 128.00
4 1-XY13-6/120 T Rosette 23.3 x 10.5 6 x 5.8 166.00

5.2 Slave Module

The Slave module has been designed based on a fully integrated solution that incorporates the required signal
conditioning, filtering, and ADC on a single chip. The direct advantages of this approach are addressed in
section 3.2.1.
Traditional precision analog integrated circuits were designed for operation from supplies of 15V , and
many would operate within specification over quite a wide range of supply voltages. Nowadays the processes
used for many, but no all, modern converters have low breakdown voltages and absolute maximum rating of
only a few volts, thereby often limiting the excitation levels, in Sheingol & Kester (2005). Solutions, such as
the ADS8684 from Analog Devices for instance, still allow higher input ranges at the expense of providing
lower resolution ADCs, or having higher costs.
This module uses the AD7195; a complete analog front end for high precision weight-scale and pressure
measurements from Analog Devices (May 2015), figure 5.3. It is able to drive a bridge with a 5V source
and comprises an ultra-low noise 24-bit sigma-delta ADC. The device accepts low level signals directly from
the transducer and outputs a serial digital word. The input signal is applied to an inbuilt Programmable
Gain Amplifier (PGA) based around an analog modulator. The modulator output is processed by a low pass
programmable digital filter, allowing adjustment of the cutoff frequency, output rate and settling time. The
AD7195 is housed in a 32-lead LFCSP package. Schematics for the slave module and its production masks
are documented in appendix C.

43
 Figure 5.3: Integrated solution schematic

5.2.1 Bridge Output

As shown in equation 3.9, the bridge output signal, UA , varies proportionally with the gage factor, K, the
excitation voltage, UE , and the strain induced by the axial force acting on the elastic member, 1 .
UA
UE
 0.65K1
since σ1  E1 and σ1  F1 {A, it can be rewritten as follows
UA pF q  0.65
KUE
F (5.3)
EA
where, F is the axial load and EA is the product between the modulus of elasticity and the cross sectional
area ogf the spring element, respectively.
The same method used to set the maximum design load in section 1.3.2, can be employed as well
to estimate the magnitude of the bridges output signal. This time the mapping matrix of the mechanism
computes the axial load at each transducer and equation 5.3 determines the corresponding bridge output.
Table 5.3 shows the predicted results if the same models were tested at 15m{s, and the bridges excited by a
5V source.

Table 5.3: Bridge output voltage estimation

Test Article Aero Loada Bridges UA [µV ]
@15m{s x [N] z [N] My [N.m] 1 2 3 4 5 6
FSAE 1:5 Series -4.66 17.2 -18.89 -31.67 -9.47 -9.47 -31.67 -18.89
Disk -1.61 0.39 -4.03 3.64 3.64 -4.03 0.39
NACA 2410 0 -0.12 -0.22 -0.2 3.41 -1.71 -0.93 -0.93 -1.71 3.41
NACA 2410 9.5 -0.61 -12.32 -0.2 17.61 11.14 14.26 14.26 11.14 17.61
a According with the reference frame of figure 4.4

The bridges output voltages are very low, in the order of µV . This underlines the importance of assuring
that the minimum change in resistance/strain produces an output large enough to overcome the noise in the
system, which leads to the next topic.

5.2.2 Accuracy

All ADCs have a certain amount of noise. That includes both input-referred noise, which is inherent to the
ADC, and quantization noise generated during the conversion of the signal. Specifications such as noise,

44
effective number of bits (EN OB), effective resolution, and noise-free resolution define in large part the
performance of the acquisition system, in IEE (2000).
The accuracy of the balance has been established based on the considerations discussed in section
1.3.4. Accordingly, the balance should be accurate to at least 5g (1 drag count) in each direction of the
reference frame. Note that the aerodynamic load is not mechanically decoupled as in a pyramidal balance,
where drag, lift, side force, and associated moments are measured directly. Instead, they are a combination
of output signals from six linear force transducers, whose resolution has to be computed through the force
transformation matrix.

Table 5.4: Force transducers resolution requirements

Force Transducers Load rg s
Load Case
1 2 3 4 5 6
FX  5g 0.23 -2.35 2.13 2.13 -2.35 0.23
FY  5g -2.59 -1.10 1.48 -1.48 1.10 2.59
FZ  5g 1.09 1.09 1.09 1.09 1.09 1.09

The total number of binary quantization levels, #counts, required is equal to the full-scale load, FF S ,
divided by the desired resolution. The latter can be easily determined from the results of table 5.4. Ideally,
it should be at most of the same order of magnitude as the lowest load acting on a force transducer, that is
0.23g. However, as it will become clear further ahead in this section, such target can only be achieved if the
dynamic behavior of the balance is compromised. For this reason a more reasonable value of 1g has been set,
which still guarantees the initial design accuracy for the normal and side forces without damaging the drag
component too much. The full-scale load can be obtained from the output function of the force transducers,
equation 5.3. Since this module integrates a PGA up to 128, this is factored into the voltage range by making
UA  UE {Gain.
1 p70  109 qp18.1  10
6 q
FF S  Gain
1 EA
0.65K
 128 0.65  2.07
 7356.7N (5.4)

The full scale load is 7356.7N , so the necessary #counts  7357{0.01  735700 and the sensitivity, C, of
the force transducers computed as follows

 UUF S  0.65 KF
C
FS
EA
 7.8mV {V (5.5)
E

With a 5V excitation voltage, the transducer full-scale output is 7.8  5  39mV . The required noise-
free, Vp
p , is therefore given by Vp
p  39mV {735700  0.053µV . This defines the required bit width,
and so the peak-to-peak noise must be less than 0.053µV . The value of the rms noise can be calculated
directly from the peak-to-peak value, dividing it by the coverage factor k. For normally distributed noise,
one can use k  6.6 for a 99% level of confidence, thus the corresponding allowable rms noise is given by
Vrms  Vp
p {4  0.008µVrms .
According to Holub & Vedral (2004), the noise-free and effective resolutions of the ADC are calculated
as follows:

Noise-free resolution  log2  log2 0.053µV  19.5

UF S 39mV
(5.6)
Up
p

Effective Resolution  log2  log2 0.008µV  22.2

UF S 39mV
(5.7)
Urms

The system performance can be determined by examining the AD7195 data sheet. Tables 5.5 and 5.6
show, respectively, the effective resolution of the ADC and peak-to-peak noise as a function of the output
data rate and amplification factor, with the chopping scheme enabled for the SIN C 4 filter.

45
 Table 5.5: Effective Resolution (Peak-to-Peak Resolution) vs. Gain and Output Data Rate
Output Data Settling
Gain of 1a Gain of 8 Gain of 16 Gain of 32 Gain of 64 Gain of 128
Rate [Hz] Time [ms]
2.5 55.3 24(22.2) 23.8(21.3) 23.8(21.3) 23.8(21.3) 23.8(21.1) 23.2(20.6)
12.5 55.3 23.8(21.2) 23.5(20.9) 23.4(20.9) 23.2(20.6) 22.7(20.2) 21.9(19.3)
15 133 23.6(20.9) 23.3(20.6) 23.3(20.5) 23(20.5) 22.6(19.9) 21.8(19.1)
37.5 55.3 23.2(20.7) 22.9(20.2) 22.8(20.2) 22.5(20) 22(19.3) 21.1(18.4)
75 26.7 22.8(20.3) 22.6(20) 22.3(19.8) 22.1(19.5) 21.5(18.9) 20.6(18)
240 8.33 22.2(19.4) 21.8(19.2) 21.63(18.9) 21.3(18.6) 20.7(18.1) 19.8(17.2)
600 3.33 21.4(18.8) 21.1(18.4) 20.9(18.2) 20.6(18) 20(17.3) 19.1(16.5)
1200 1.67 19.9(17.3) 19.8(17.1) 19.8(16.9) 19.6(16.9) 19.3(16.5) 18.5(15.8)
a The output peak-to-peak (p-p) resolution is listed in parentheses.

Table 5.6: Peak-to-Peak Noise (nV) vs. Gain and Output Data Rate
Output Data Settling
Gain of 1 Gain of 8 Gain of 16 Gain of 32 Gain of 64 Gain of 128
Rate [Hz] Time [ms]
75 26.7 7780 1202 707 424 311 304
240 8.33 13142 2121 1273 778 573 502

In section 1.3.3 we estimated the maximum expected vortex shedding frequency. According to the
Nyquist sampling theorem, the aliasing effect can be easily avoided if the acquisition rate is set higher than
twice the maximum frequency of the signal, that is 70  2  140Hz. However, it can be quickly inferred
from table 5.5(green shade) that the AD7195 is not capable of satisfying both the resolution and dynamic
requirements at the same time. An output data rate of 150Hz yields an effective resolution of only 20.3bits,
when 22.2bits are necessary to achieve the wanted resolution. On the other hand, from table 5.6 the peak-to-
peak noise is 394nV and thus the number of noise-free counts is UF S {Up
p  39mV {394nV  98984.8. The
ultimate force transducer accuracy for a full-scale load of 735700N is then 735700{p2  98984.8q  3.7g for
a level of confidence of 99% and roughly 2.2g for a level of confidence of 95%.
Although it was not possible to achieve the target accuracy, it is important to remind that the main ob-
jective was to design the force transducers and acquisition system towards maximizing the overall performance
taking the initially established value as a reference rather than a requirement. In addition, despite not being
possible to satisfy both requirements at the same time, it is still possible to achieve the reference accuracy over
a decrease of the dynamic performance of the balance. Table 5.5(blue shade) shows that if the acquisition rate
is set to 10Hz, the AD7195 is capable of providing the necessary resolution to achieve the wanted accuracy
of 1g.

5.2.3 Driving Bridge

Of particular concern is maintaining the accuracy and stability of the bridge excitation voltage. The bridge
output it directly proportional to the excitation voltage, and any drift in the excitation produces a corresponding
drift in the output signal. For this reason, most all-elements varying bridges are six-lead assemblies: two leads
for the bridge output, two leads for the bridge excitation, and two sense leads. This configuration, shown in
figure 5.4, is called Kelvin or 4-wire sensing.
Essentially, the same reference is used for both the transducer excitation and the ADC. So, a given
percentage change in excitation is countered by the same percentage change in the conversion process, and

46
vice versa. The ADC output code, Dout , is then a digital representation of the ratio of the input signal, AIN ,
to its reference, Vref . This way, since the input of the converter and its reference come from the same source,
changes in the excitation do not introduce measurement error.

Figure 5.4: Ratiometric operation functional diagram

Ratiometric operation is a very powerful tool; it enables measurement and control, using the on-chip
analog supply, to obtain an accuracy independent of the stability of voltage references or excitation sources.
This is a very simple solution yet it has reduced the circuitry complexity and improved its conditioning efficiency.

5.2.4 Bridge Excitation

Strain-gages can be excited using AC or DC sources and both have pros and cons. The inherent advantages
to DC excitation include simplicity of implementation and low cost, which is in agreement with the reduce-to-
essentials philosophy. The downside is the difficulty in separating the actual signal from unwanted DC errors
due to both thermal- and pink-noise, parasitic thermocouples, and amplifier offset errors. Unfortunately these
are not fixed, they vary unpredictably, thus can not be compensated.
Ideal differential amplifiers would have an infinite Common-mode Rejection Ratio (CMRR), that is, the
signal common to both input leads would be fully suppressed and only the difference between them amplified.
In practice though, this is not achievable. Despite the rather high CMRRs, inherent to high performance
amplifiers, traces of DC offsets are always expected. Parasitic thermocouples, by their turn, can be found in
normal circuit wiring. Junctions between tin-lead solder and PCB traces can introduce errors of 3 to 4µV { C,
if thermal gradients exists. Thermocouple junctions also exist between the copper traces of the circuit board
and the IC amplifier pins, creating voltage errors as large as 35µV { C, in O’Grady (2000).
Although the AC excitation technique is usually more complex to implement, it offers many performance
benefits. It is able to remove effectively offset errors, to average out pink noise and eliminate effects of parasitic
thermocouples, enjoying much higher signal-to-noise ratios. It is, thereby, capable of producing discernible
output signals with much lower excitation voltages in comparison with a DC source.
The AD7195 provides a mean to excite the bridge with an AC square wave, reversing the polarity of the
excitation signal between measurements, thus canceling out effectively the induced DC errors. This chopping
scheme is also able of removing pink noise, which is dominant at low frequencies in these kind of applications.
Figure 5.5 outlines the connections for the AC configuration.

47
 Figure 5.5: AC excitation configuration schematic

Using transistors Q1 to Q4 to perform the switching, the polarity of the excitation voltage to the bridge
is reversed on alternate cycles. Every time the polarity changes, the ADC takes a sample. Therefore, since the
input and the reference voltages are reversed on alternate cycles, the AD7195 must synchronize itself with this
reversing process. To do so, it provides the logic control signals for the switching of the excitation voltage.
These ports are the non-overlapping CMOS outputs ACX and ACX
To better understand the chopping scheme behind the AC excitation, consider the cycle timing diagram
shown in figure 5.5, and note that all induced DC and low frequency errors haven been lumped together as
EOS . During phase 1, Q1 and Q4 are active while Q2 and Q3 are inactive; the output Vout1 , is given by
pVA EOS q. During phase 2, Q2 and Q3 are active while Q1 and Q4 are inactive; the output Vout2 , is given
by p
VA EOS q. The error voltage is then removed by averaging both results in the digital filter, yielding
Vout  VA .
In order to simplify the circuitry performing the AC excitation, all the discrete MOS transistors present
on the schematic above have been replaced by the dedicated bridge driver chip MIC4427, from Micrel Semi-
conductors (May 2015). Figure 5.6 illustrates its configuration and respective cycle timing diagram.

(a) Functional diagram (b) Cycle timing diagram

Figure 5.6: MIC4427 test circuit

48
 The supply input Vs , corresponds to the excitation signal VDD provided from the AD7195, and the
control inputs INA and INB, are driven by the ACX and ACX ports, respectively. The outputs A and B, in
turn, are connected to the Wheatstone bridge excitation diagonal. During stage 1, INA is set to high while
INB is set to low; the output OUTA takes the supply input value of 5V , and the output OUTB is tied to
ground. During stage 2, INA is set to low while INB is set to high; the output OUTB takes the supply input
value of 5V , and the output OUTA is tied to ground.
The sampling rate is synchronized with the reversing cycles of the excitation signal, consequently, it has
been set to 150Hz. The acquisition system is capable of providing higher frequencies sources, up to 1200Hz,
but that would severely damage the quality of the acquired signal.

5.3 Master Module

The Master module manages all communications with the Slaves through SPI, it reports the acquired data
to the user, as well as any fault it may occur. It is also responsible for keeping the Slave modules working
below a critical temperature level. As such, this module requires a fairly good processing power and several
I/O functions. Its architecture is shown in figure 5.7. To fulfill these requirements, the dsPIC FST Module
has been considered, figure 5.8.

Figure 5.7: Data Acquisition Architecture

The dsPIC FST Module employs the 16-bit micro-controller dsPIC30F6012A from Microchip (May
2015) and was initially designed by Professor Dr. Moisés Piedade (DEEC-IST) to be a multi-functional and
flexible system. Since then, it has been improved by several MSc students and thoroughly used by Projecto
FST Novabase as a development platform for all electronic systems.
Although it has been designed to operate from a 12V supply, to match the low voltage level of the
FST prototypes, the microprocessor works at 5V . So it comprises a power converter which not only adjusts
the output voltage but also filters traces of electric noise from the power source. Since this is a base system
to which several expansions can be added, the embedded DC-DC converter has a very high power capacity,
being capable of providing a constant 5V supply up to 3A. The several components of the acquisition system
such as the Slave modules, cooling fans, and LED’s could then be powered by this source, thereby simplifying
the distribution system.

49
 Figure 5.8: dsPIC FST Module, v1.3

5.3.1 Software

Each force transducer has its own dedicated converter, so the SPI bus operates with a single Master and six
Slave devices working independently. This type of configuration uses an independent Chip Select (CS) line for
each Slave, as illustrated in figure 5.9. The Master module starts every communication by selecting the Slave
device with the logic level 0 on the corresponding CS line. Similarly, the communication ends by setting it to
high, 1.

Figure 5.9: Independent Slave configuration

One of the most important requisites to the development of the acquisition system is that it should be
a plug and play solution. Bearing this in mind, it has been designed to monitor each force transducer and send
the acquired data to a computer via USB. This configuration enabled to work on a flexible user interface, and
data manipulated and displayed in real time (see appendix C). Another alternative was to develop an isolated
system that would store the acquired raw data in a SD card instead, for later reduction. This solution allows
higher acquisition rates, since it does not depend on the serial communication performance of MATLAB. On
the other hand, it would be significantly more laborious to work with.
The Flowchart in figure 5.11 illustrates the work-flow of the Master module. The normal routines are
represented by white blocks, whereas the red ones denote errors handling and the green ones user inputs.
To improve robustness and ease the initial debugging and development of the system, several measures
have been taken to ensure its proper operation. The AD7195 has a series of programmable functions controlled
using a set of on-chip registers that listed in appendix C. These functions determine the operation features of
the signal conditioning, that is, the amplification factor, the acquisition rate, type of excitation, and filtering

50
options, among others. Each Slave has to be programmed every time it is turned on, therefore it is of at most
importance to check their configuration prior to perform any conversion. To do so, they report their set of
on-chip registers to the Master module, which in turn compares it with the programming sequence. If there
is not a match, the corresponding error flag is set to high. After the programming cycle, if a single error flag
is set, the Master module reports to the user which Slaves have not been properly programmed, via LED, and
halts itself.
While in the main cycle, prior to sending the bridge output readings, all Slave devices send a status
report every second providing information regarding their operating conditions:

• Reference detect - It detects whether a valid reference for conversion is available. If the voltage
between the reference pins is within 0.3 and 0.6V , the AD7195 detects that it no longer has a valid
reference. If this happens, the conversion output is clamped to 11 s and the Master module lights on the
corresponding Slave LED (see figure 5.10);

• Under- or over-range - If a large enough compressive or tensile load is applied on the force transducers,
the readings of the ADC are clamped to all 01 s or 11 s, respectively. If so, the master module detects it
and lights on the respective LED. This is, however, a very unlikely scenario given the load range of the
force transducers;

• Temperature sensor - Each Slave has its own embedded temperature sensor with an accuracy of
2 C. If any chip exceeds the maximum operating limit, set to 80 C, then the cooling fans are turned
on permanently.

It is important to point out that these symptoms may have very different natures, in particular the first
two. One of the most likely causes for the absence of a reference voltage is when the excitation line is short
circuited to ground, and consequently the reference too. Although it may seem to be a far fetched possibility,
specially after running a few wind tunnel tests, the sensing bridges and their wiring to the acquisition module
can be easily damaged. If so, this allows to quickly diagnose and to fix the problem.
The under- or over-range warning has a very similar role. In a normal situation, it would be only
triggered by the excess of loading, still it may be also caused by several system malfunctions. For instance, if
the wiring from from the bridges to the acquisition module is not properly connected, or if there is a wiring
fault in the bridges.
Finally, the acquired data is sent to a computer through UART, and each complete message has a total
of 4 bytes; the first one carries the Slave identification followed by the 24-bit word from the ADC. Taking into
account the acquisition frequency of 150Hz, the data transfer rate is roughly 230kbit{s.

Figure 5.10: Data acquisition box

51
Figure 5.11: Flowchart

52
Chapter 6

Static Calibration

The balance has been manually calibrated using the experimental setup shown in figure 6.1a. The calibration
rig is an orthogonal aluminum structure placed on top of the bottom platform (4). Its correct alignment with
the reference frame of the balance is assured since the screw of the leveling feet match four holes on the
bottom of the structure, one in each corner, as illustrated in figure 6.1b. This mechanism comprises a set
of pulleys (1) supported by the upper frame (2), and cables are then treaded around them in the grooves.
Depending on which loading component is to be applied, a different pulley setup is required. The pulleys
can slide along the the XX or YY directions and their height can also be adjusted, whether by sliding the
whole upper frame along the vertical beams (3) or by fine-tuning each one individually. Precise alignment
of the pulleys, cables and rig is essential in order to ensure ‘true’ loads are applied on the balance and in
the desired direction. The calibration procedure is described in Ferreira (2015). The maximum forces and
moments applied to the balance during the calibration were chosen to correspond to the range of loads likely
to be experienced by a model during the subsequent tests.

(a) Calibration Rig - 1 Pulley; 2 Upper Frame; (b) Balance-Rig assembly detail
3 Vertical Beam; 4 Bottom Platform

Figure 6.1: Calibration Setup

53
6.1 Calibration Methodology

Commonly, the mathematical model of the calibration relates the estimated aerodynamic components (ŷi , i 
1, 2, ..., 6) to second-order polynomial functions of the force transducers output (rj , j  1, 2, ..., 6). However,
a sufficient number of load cases have to be applied to the system so that all coefficients of the calibration
equation may be determined. The complexity of this procedure increases exponentially with the order of the
functional relationship, so to improve accuracy by simplifying it, a first-order calibration has been carried out
instead. The form of the calibration equation is

¸
6
ŷi  rj cj,i i i  1, 2, ..., 6 (6.1)

j 1

where cj,i are the linear calibration coefficients, and i a random variable, usually regarded as a measurement
error for ŷi .
The conventional calibration procedure is to apply pure component loads independently. Although
straightforward, this method does not represent the real, practical loadings that the model encounters. It goes
against the basic principal that the loadings of the balance during the calibration should be representative of
that experienced by the model during wind tunnel tests.
Theoretically, if the calibration equation describes exactly the relationship between the applied load
and the force transducers output, the way the calibration is conducted should not affect the outcome of the
results. Though in reality, the calibration equation given in 6.1 is only an approximation and does not describe
exactly the load-output relationship. The response of the balance can be quite different under different loading
distributions which causes different deflection characteristics in the balance. The calibration coefficients thus
calculated are dependent of how the balance is loaded during the calibration.
Ramaswamy et al. (1987) suggested a Least Squares (LS) method for the calibration of wind tunnel
balances. This method does not require the balance components to be loaded independently. In fact, any
combination of component loads may be applied simultaneously. The distribution of loads on the balance
during the calibration can be chosen to match as closely as possible those likely to be experienced by the model
during wind tunnel tests. Virtually any random set of component loads may be applied to the balance, in any
sequence, provided that all the data obtained generate enough independent information for the calculations
of the coefficients. This method is be described in the subsequent sections.

6.1.1 Formulation of the Weighted LS Method

Given that for a set of loading combinations, yi , where i  1, 2, ..., 6, the corresponding force transducers
output are rj , where j  1, 2, ..., 6. If there are N sets of yi , with N sets of corresponding rj , denoted by
yp,i and rp,j where p  1, ..., N , then the calibration coefficients may be found such that the sum of squared
residuals is a minimum. That is
¸
N
si  wp pyp,i
ŷp,i q2 (6.2)

p 1

¸
N ¸
6
si  wp pyp,i
rp,j cj,i q2 (6.3)
p 1  
j 1

should be minimum. wp is the weight assigned to the pth observation that reflects the uncertainty of the
measurement. Generally this is a function of the variance of the observations, denoted by σ 2 . A straightforward
strategy is to define w  σ
1 , however, a more complete weighting scheme, described in section 6.1.2, as
been used.

54
 Minimizing the residuals involves taking the partial derivative of each si with respect to each of the
coefficients and equating the resulting expression in zero. The result is a set of 6 linear equations and 6
unknowns (c ,i ) for each component i.
Bsi
Bcj,i  0 (6.4)

¸
N ¸
6
wp ryp,i
rp,j cj,i s rp,1 0 (6.5)

p 1 
j 1

¸
N ¸
6
wp ryp,i
rp,j cj,i s rp,2 0

p 1 
j 1

r...s
¸
N ¸
6
wp ryp,i
rp,j cj,i s rp,6 0

p 1 
j 1

It is perhaps more convenient developing this approach in the matrix form having in mind its imple-
mentation. Starting from equation 6.1,
Ñ
ÝY  RÑ
ÝC Ñ
Ý (6.6)
Ñ
ÝC P IR6 , Ñ
ÝY and the residual vector Ñ
Ý P IRN . R is a N  6 matrix and is known as the design matrix; its
rows are the readings of the force transducers for each calibration loading condition yp .
     
y1 c1 r1,1 r1,2 ... r1,6
y  c  r 
Ñ
ÝY   2  ; Ñ
ÝC   2  ; R   2,1 r2,2 ... r2,6 
 ..   ..   .. .. .. .. 
 .  .  . . . . 
yN c6 rN,1 rN,2 ... rN,6

The least-squares solution is the one with the smallest misfit to the measurements, that is, the euclidean
length of Ñ
Ý given by
Ñ
Ý .Ñ
Ý  Ñ Ý  ¸ 2p
Ý T Ñ
N
(6.7)

p 1

should be minimum. So C
Ñ
Ý  Ĉ such that

Ýc rÑ
∇Ñ
Ñ
Ý
Ý p C q.Ñ Ñ
Ý
Ý p C qs  0 Ñ
Ý (6.8)

Equivalently, and introducing the weighting matrix WpN N q ,

Ñ
Ý
RÑ ÝC qT W pÑ
Ýc rp Y
∇Ñ
ÝY
RÑ
ÝC qs  Ñ
Ý0 (6.9)
Ý rÑ
∇Ñc
ÝY T W Ñ
ÝY
Ñ
ÝY T W RÑ
ÝC
Ñ
ÝC T RT W Ñ
ÝY Ñ ÝC T RT W RÑ
ÝC s  Ñ
Ý0 (6.10)

Ñ
ÝT Ñ
Ý Ñ
Ý Ñ
ÝT
Since Y W R C  pR C qT p Y W qT Ñ
ÝC T RT W T Ñ
ÝY and given the symmetry of the weighting matrix,
W T  W , this becomes
Ñ
ÝT Ñ
Ý
2pRT W Ñ
Ýc r Y W Y
∇Ñ
ÝY qT Ñ
ÝC ÑÝC T RT W RÑ
ÝC s  Ñ
Ý0 (6.11)

2RT W ÑÝY 2RT W RÑ ÝC  0 (6.12)

The calibration coefficients matrix is therefore determined as follows

Ĉ  pRW Rq
1 RT W Y (6.13)

55
6.1.2 Weighting Scheme

The LS method assumes that each data point provides equally precise information, that is, the standard
deviation of the error is constant over all values of the predictor. However, this assumption may not hold in
this application, not even approximately. Since the calibration process is mostly manual, it is not reasonable to
assume that every observation should be treated equally. For this reason, a weighted LS has been considered
in order to maximize the efficiency of the parameter estimation, which still shares the ability with the other
methods to provide different types of easily interpretable statistical intervals for estimation and prediction.
To make the most out of this strategy, the weighting matrix, W , must reflect the behavior of random
errors in the model. Usually weights that are inversely proportional to the variance at each level of the
explanatory variables, in this case the readings of the force transducers, yield precise parameter estimates. Yet,
if they are estimated from a small number of replicated observations, the results can be badly conditioned. It
is, therefore, important to assess afterwards the true effects of this type of calibration.
The weighting matrix, in this project, is related to the uncertainties of the applied loads. It is made up
of the contributions of the sources of errors due to the application of weights in the calibration system VW
and the uncertainties in the readings of the bridges VR .

W  pVW DVR DT q
1 (6.14)

Matrix VW

VW is a diagonal matrix of dimensions N  N . Ideally, its elements are determined based on the uncertainties
declared in the calibration certificates of the weights used in the calibration loading process, and also in an
estimation of the errors caused by the calibration system. Unfortunately, the weights employed are not certified
so the associated uncertainty is not available. The second contribution, on the other hand, is mostly random
errors due to misalignments of cables and pulleys as well as friction between the wheels and their axis.
The difficulty of experimentally estimating these uncertainties led the author to explore two distinct
approaches:

1. The first model consists in statistically quantify the contribution of these errors to the overall uncer-
tainty. It takes two steps: Initially, the fitting method described in the previous section 6.1 is employed,
but setting the covariance matrix equal to the identity in equation 6.13. This corresponds to first assign
an uncertainty value equal to 1 to the data points.

C  pRT IRq
1 RT IY  pRT Rq
1 RT Y (6.15)

Following this, the uncertainty associated with the set of measurements is approximated by the standard
deviation of the fit σi , which is the positive square root of the expression:

¸
N
σi2  N
1 m pyp,i
ŷp,i q2 (6.16)

p 1

where m is the number of parameters to be fitted, in this case six.

Finally, based on an educated guess, the contributions assigned to the diagonal matrix VW due to the
calibration procedure errors are computed as 70% of the standard deviation σ. It is then assumed that
the calibration process contributes this amount to the dispersion of the data.

2. The second model follows a rather new procedure that has been recently introduced by Reis et al.
(2013), based on a type B evaluation of the error sources: Firstly, the contribution of the uncertainties
of the weights applied to the calibration cross are promptly verified with a precision balance. Secondly,

56
 an estimation of the uncertainties in the calibration system is obtained based on the quantification of the
sources of errors that affect the resolution of the calibration system itself, such as frictional forces and
misalignments between cables and pulleys. It involves applying small weights to the trays and verifying
when the load readings of the aerodynamic balance change their indication in a detectable way. The
downside of this method is that in order to fully characterize the system behavior it is necessary to carry
out this procedure for all six aerodynamic components.

Matrix VR

The replication of the calibration procedure supplies the required information to build VR that is diagonally
symmetric. Several calibrations should be carried out under repeatability conditions over a short period of
time. The diagonal elements of this matrix are represented by the variances and the off-diagonal elements are
the covariances between the readings of the force transducers. The first N lines are elements related to the
first force transducer, R1 , followed by the N lines related to R2 and so on, resulting in a 6N  6N matrix -
the first subscript in the equation 6.17 identifies the calibration loading condition, 1 to N , and the second is
related to each force transducer, 1 to 6.
 
up r , r q upr1,1 , r2,1 q ... upr1,1 , rN,1 q upr1,1 , r1,2 q ... upr1,1 , rN,6 q
 upr1,1 , r1,1 q upr2,1 , r2,1 q upr2,1 , rN,1 q upr2,1 , r1,2 q

upr2,1 , rN,6 q 
 2,1. 1,1 ... ...


 .. ..
.
..
.
..
.
..
.
..
.
..
. 


VR  uprN,1 , r1,1 q uprN,1 , r2,1 q uprN,1 , rN,1 q uprN,1 , r1,2 q uprN,1 , rN,6 q

 ... ...
 (6.17)
 upr1,2 , r1,1 q upr1,2 , r2,1 q ... upr1,2 , rN,1 q upr1,2 , r1,2 q ... upr1,2 , rN,6 q 

 .. .. .. .. .. .. 

. . . . . . 
uprN,6 , r1,1 q uprN,6 , r2,1 q ... uprN,6 , rN,1 q uprN,6 , r1,2 q ... uprN,6 , rN,6 q

Consider, for the moment, that 3 calibrations have been performed to provide enough information for
statistical analysis. VR is calculated by the DRS, implemented in MATLAB, from a 6N  3 matrix named
Rglobal , where the first N elements of column 1 correspond to the readings of the load cell R1 from the first
calibration loading condition, followed by the N elements of the readings of the load cell R2 and so on. The
same procedure is repeated for columns 2 and 3, using the data set from the second and third calibration
loading conditions, respectively. The matrix is then transposed and input in the MATLAB function “cov”.

Matrix D

The matrix DVR DT , in equation 6.14, corresponds to the variances and covariances of the aerodynamic loads
due to uncertainties in the bridge readings and it is N  N . D is N  6N and its elements are the sensitivity
coefficients evaluated by taking the partial derivatives in equation 6.1. For the drag force, y1 , matrix D is
given as:

 By 1,1 By1,1 By1,1 By1,1 BY1,1 

 BByr Br2,1 BrN,1 Br1,2 BRN,6
... ...
1,1
By2,1 By2,1 By2,1 By2,1 

 Br. 
2,1
Br2,1 ... BrN,1 Br1,2 ... BrN,6
D   ..
1,1
.. ... .. .. ... .. 

 By
ByN
1,1
. .
ByN
1,1 ByN
1,1
.
ByN
1,1
. 

(6.18)
 Br N

B
1,1
yN,1
1,1
Br2,1
ByN,1
... BrN,1
B yN,1
Br1,2
B yN,1
... BrN,6
B yN,1

Br1,1 Br2,1 ... BrN,1 Br1,2 ... BrN,6
The first and second subscripts in equation 6.19 identify the loading number and the aerodynamic
component, respectively. Note that once again the first block of N columns is related to the first force
transducer R1 , the second is related to R2 and so on. In this equation, only the sensitivity coefficients whose

57
second subscripts match, i.e. corresponds to the same loading, have values different from zero. Consequently,
the matrix D is reduced to

 By 1,1 By1,1 
 Br 1,1
0
By2,1
... 0 Br1,2 ... 0


 0. Br2,1 ... 0 0 ... 0 

D   .. .. .. .. .. .. ..

 . . . . . . 

(6.19)
 0 0 ... 0
ByN,1
0 ... 0
ByN,1

0 0 ... BrN,1 0 ... BrN,6

The elements B yp,1 {B rp,j of the matrix D are given as

Byp,1  c Byp,1  c Byp,1  c

Brp,1 1,1 Brp,2 2,1 ...
Brp,6 6,1 (6.20)

These parameters are estimated by performing the least squared regression to the calibration data, as described
in the previous section, by initially making W equal to VW
1 . This means that the fitting is necessarily applied
twice to the calibration data; firstly W  VW
1 and in the second time W  pVW DVR DT q
1 .

6.1.3 Uncertainty Assessment

Matrix pRT W Rq
1 , in equation 6.13, is called the error matrix and it contains information about the uncer-
tainties of the estimated parameters. Its diagonal elements correspond to the variances (squared uncertainties
u2c ) and the off-diagonal elements are the covariances of the fitted parameters, upci , cj q.
Thus, the uncertainty in the predicted aerodynamic component is the positive square root of the diagonal
elements of

VF̂  RpRT W Rq
1 RT (6.21)

which is equivalent to applying the law of propagation of uncertainty to equation 6.6, according to the Joint
Committee for Guides in Metrology JCGM (2008).

6.2 Calibration Results

The developed Data Reduction System (DRS), described in appendix C, provides a dedicated platform to aid
in the calibration process. It is a fully dynamic environment that allows the user to fill-in the calibration loading
condition table prior to calibrating the balance, or at the same time. The reduction system averages the bridges
output values over a user-preselected acquisition period. While acquiring, data is displayed in real time to
facilitate the detection of badly conditioned information, avoiding the corruption of the calibration data. When
the output readings for all calibration loading conditions have finally been retrieved, the coefficients matrix
C is automatically computed following the method described in the previous section and whose work-flow is
represented below in figure 6.2. All the intermediate results are provided, as well as the uncertainty values of
the transformation matrix and other quality-of-fitting indicators.

58
 Elements Calibration data Sensors output

make W  I Model 2
C  pR IRq
1 RT IY
T

Model 1

Estimate VW , Estimate VW ,
type A evalutaion type B evaluation

make W VW
1


C  pR VW Rq
1 RT VW Y
1
T

Compute D matrix

make W  p VW DVR DT q
1

C  pRT W Rq
1 RT W Y
Figure 6.2: Calibration algorithm flowchart

On the first attempts to calibrate the wind tunnel balance, the matrix VW , which represents the
component of the error induced by the calibration system, has been estimated conducting a type-A evaluation.
The calibration results following this procedure can be found in appendix D and a detailed description is
given in Ferreira (2015). The alternative strategy, a type-B evaluation, is considerably more complex and
time-consuming, so it has only been employed after acquiring some ability in operating the equipment in
question.

6.2.1 Loading Combinations

Thirty-six loading combinations have been applied to calibrate the balance and the process has been repeated
seven times under similar conditions1 , to achieve statistical significance. A set of 20 weights of 100g, and
two trays weighting 1N each, were used in different configurations according with the calibration loading
conditions. As already mentioned, these are not standard calibrated weights, so, in order to assess the
uncertainty associated, their mass has been measured in a precision balance with a resolution of 0.01g and
accurate down to 0.05g - see appendix D. The results of this assessment are the following: taking into
account a rectangular distribution, the maximum error in the applied weights is 1.6g whereas for the trays is
only 0.6g. For a confidence interval of 99% this leads to an uncertainty of 0.45g and 0.2g, respectively.
Table 6.1 shows part of the loading plan that is fully available in the appendix D.

1 The facility where the calibration took place does not provide controlled room conditions, so to ensure repeatability all calibrations
have been performed on the same day, between 14h00 and 23h00 (a period 9 hours).

59
 Table 6.1: Applied calibration loads and associated uncertainties
Loading FX uF x FY uF y FZ uF z MX uM x MY uM y MZ uM z
No. rg s rgs rg s [Link] [Link] [Link]
1 0 0 0 0 0 0 0 0 0 0 0 0
2 202.4 0.9 0 0 0 0 0 0 0 0 0 0
3 402.9 1.9 0 0 0 0 0 0 0 0 0 0
.. .. .. .. .. .. .. .. .. .. .. .. ..
. . . . . . . . . . . . .
6 1004.6 4.5 0 0 0 0 0 0 0 0 0 0
.. .. .. .. .. .. .. .. .. .. .. .. ..
. . . . . . . . . . . . .
31 0 0 0 0 0 0 0 0 0 0 0 0
32 0 0 0 0 202.4 0.9 0 0 10.1 0.2 0 0
33 0 0 0 0 402.9 1.9 0 0 20.1 0.4 0 0
.. .. .. .. .. .. .. .. .. .. .. .. ..
. . . . . . . . . . . . .
36 0 0 0 0 1004.6 4.5 0 0 50.2 1 0 0

Figure 6.3: Eccentric loading setup, FZ MX

The calibration system has been thoroughly studied in order to minimize random errors. Several sets of
loading combinations have been tested to identify the factors to which the calibration data is most sensitive,
and how their effects can be mitigated. The ultimate loading plan can be broken apart in 6 blocks: the first
one corresponds to pure loading in the axial direction, FX ; the balance is initially loaded by a tray holding a
single weight and additional sets of two weights (weighting a total of 200g) are successively incremented up
to 1000g. The same strategy is followed for the normal and side forces, one at each time. By their turn, the
bending moments are applied together with force components. Despite the practical difficulties in quantifying
its impact, the human factor is probably the most responsible for the error induced by the calibration system.
Given the way it operates, pure moments can not be applied using just a single tray, at least two are necessary,
further increasing the complexity of the loading setup and, as a result, the error. So the solution encountered
was to apply an eccentric load on the balance that induces a moment around its resolving center proportional
to the length of the lever arm, as illustrated in figure 6.3. As such, MX and MY are applied together with an
eccentric normal-force component and MZ with a side or axial force.
Table 6.2 summarizes the bridge sensors output, R1 to R6 , under the calibration loading conditions
specified in table 6.1. Figure 6.4 shows the wind-axis reference frame adopted in this project, previously
established in section 1.2.

60
 Table 6.2: Force transducers output readings rmV s
No. R1 R2 R3 R4 R5 R6
1 0.001 0.007 0.011 0.005 -0.001 0.003
2 1.418 -0.089 -1.432 -1.354 -0.129 1.389
3 2.840 -0.177 -2.847 -2.702 -0.254 2.773
4 4.321 -0.267 -4.306 -4.068 -0.380 4.182
5 5.761 -0.341 -5.782 -5.434 -0.495 5.596
6 7.246 -0.436 -7.266 -6.838 -0.614 7.028
.. .. .. .. .. .. ..
. . . . . . .
36 2.234 2.322 0.650 0.602 2.194 2.125

Figure 6.4: Frame of reference

6.2.2 Parameters Estimation

Performing the LS fitting for the second time, that is, considering the uncertainties in the data points estimated
through equation 6.14, results in the calibration coefficients shown in table 6.3. The remaining coefficients
and respective uncertainty can be found in appendix D.

Table 6.3: Calibration coefficients and uncertainty estimations of the load components FX and MX
c1,j rN {mV s uc1,j rN {mV s uc1,j {c1,j c4,j rN.m{mV s uc4,j rN.m{mV s uc4,j {c4,j
c1,1 0.6653 0.0147 45.11 c4,1 -0.2521 0.0010 249.58
c1,2 -0.6376 0.0147 43.38 c4,2 -0.0234 0.0013 17.86
c1,3 -0.0185 0.0148 1.25 c4,3 0.2210 0.0015 145.52
c1,4 0.0363 0.0138 2.63 c4,4 -0.2287 0.0010 226.67
c1,5 -0.6740 0.0144 46.65 c4,5 0.0226 0.0010 21.96
c1,6 0.7062 0.0152 46.28 c4,6 0.2660 0.0016 165.43

Table 6.4: Estimated calibration loading conditions and associated uncertainty

Loading FX uF x FY uF y FZ uF z MX uM x MY uM y MZ uM z
No. rgs rg s rg s [Link] [Link] [Link]
1 -0.4 0.0 -0.5 0.0 1.1 0.0 0.1 0.0 0.0 0.0 0.0 0.0
2 201.1 4.1 -0.6 1.4 -0.9 2.7 0.4 0.4 -0.2 0.4 0.0 0.3
3 403.9 8.1 -0.4 2.8 -0.3 5.3 0.3 0.7 -0.2 0.8 0.0 0.6
4 609.3 12.2 0.3 4.2 0.8 8.0 0.2 1.1 0.0 1.2 0.1 0.9
5 813.0 16.3 0.9 5.6 0.6 10.7 0.1 1.5 -0.1 1.6 0.1 1.2
6 1002.2 20.1 2.4 7.0 -0.2 13.4 -0.3 1.8 0.1 2.0 0.0 1.5
.. .. .. .. .. .. .. .. .. .. .. .. ..
. . . . . . . . . . . . .

6.2.3 Quality-of-fitting

Chi-squared goodness-of-fit test

The LS method finds the optimum description of a set of data that minimizes the weighted sum of the squares
of the deviation of the calibration data yp,i from the fitting function ŷp,i prp,j q. In this section, the relationship
between the expected ‘amount of scatter’ and the one that is actually observed s2 , given by equation 6.16, is

61
reduced to a chi-square value.
¸
N
s2i  N
1 m pyp,i
ŷp,i q2
p 1 
The parent variance of the data σ 2 is a characteristic of the parent dispersion of the data about a
parent distribution so it is not descriptive of the fit. On the other hand, the estimated variance of the fit s2
is characteristic of both the spread of data and the accuracy of the fit. Therefore, χ2 , given as the ratio of
the estimated variance s2 to the parent variance σ 2 times the number of degrees of freedom ν  N
m, is a
convenient measure of the goodness of fit, in (Bevington & Robinson, 2003).
N !
¸ )
χ2i  1
σ2
ryp,i
ŷp,i s2 (6.22)

p 1 i

Taking into account the covariances, in the matrix form,

 pY
Ŷ qT W pY
Ŷ q
χ2 (6.23)

where W is the inverse of the covariance matrix pVW DV RDT q
1 .

s2 and σ 2 can be better compared if the reduced chi-square is used instead.

 pY
Ŷ q W pY
Ŷ q
T
χ2ν (6.24)
ν
If the fitting function is a good approximation to the parent function, then the estimated variance s2 should
agree well with the parent variance σ 2 , and the value of the reduced chi-square should be approximately unity.
A value lower than 1 does not necessarily indicate a better fit. It is simply a consequence of the fact that
there exists an uncertainty in the determination of s2 . A value of χ2ν that is very small may indicate an error
in the assignment of the uncertainties in the measured variables. Table 6.5 summarizes the chi-squared and
its reduced form for the calibration performed.

Table 6.5: Goodness-of-fit χ2

Force Transducer χ2 χ2ν P pχ2 ; ν qr%s
R1 38.9 1.30 13.1
R2 31.4 1.05 39.2
R3 32.6 1.09 33.8
R4 54.7 1.82 0.6
R5 45.4 1.51 3.9
R6 41.1 1.37 8.8

The reduced chi-squared values for all force transducers are close to unit. However, at a closer look,
the fourth and fifth are still slightly higher than desired.
The P
values presented above were determined using the χ2 cumulative distribution table. If it is
equal or smaller than a given significance level α, it suggests that the observed data are inconsistent with a
normal distribution. By convention if the p-value is smaller than α  0.05, the fit is not accepted. So it can
be promptly seen that the calibration data from the fourth force transducers, in particular, do not fulfill the
desired quality standard.
If all force transducers shared the same symptom, this could be an indication that the adopted math-
ematical model might not suit this calibration This is not the case though, which implies that the uncertainties
of the data points are in fact larger than stated or the scatter of the calibration data points does not follow a
Gaussian distribution as initially thought it would.

62
Chapter 7

Experimental Methods

7.1 Facility

The experiments herein presented were conducted in the open-return blow-down wind tunnel of the fluid
dynamics laboratory at IST.

Figure 7.1: Open-return wind tunnel

The maximum airflow speed is 20m{s and it is controlled manually by means of a trimmer potentiometer
with 10{0.2  50 adjustable positions. The fan is approximately 1m diameter; it is located downstream air
filters and it pushes the air through the flow conditioners and test section. This type of configuration, blow-
down wind tunnels, tend to cause twist in the flow that is significant enough to exist past the flow conditioning
sections and into the test section. The motor is mounted outside the circuit, parallel to the fan, and delivers
15kW through a driving rubber belt which mechanically links both components.
The stagnation chamber, located upstream contraction, is the part that has the largest cross section
area and it has a major influence on the flow quality in the test section. Because of the high local pressure
drop generated by quality flow improving devices it is important to position them in the location with the
lowest speed and thereby minimize their contribution to the total pressure drop of the wind-tunnel circuit.
This plenum chamber is equipped with one honeycomb and anti-turbulence screens. The honeycomb is a very
effective flow straightening device, reducing cross-stream components, and has a relatively low pressure drop.
On the other hand, the screen layer is very effective in breaking up larger eddies and acts primarily to reduce
mean non-uniformities and fluctuations of the stream-wise component. The screens are ought to be used in
a cascade configuration with subsequently finer mesh but there is no information regarding the number or
specifications of those used in this installation.

63
 The final and the most important part in the wind tunnel circuit is the contraction through which the
flow passes before reaching the test section. In the contraction the flow is accelerated rapidly; this results in
a large stream-wise strain that reduces mean flow variations, and the larger the contraction ratio, CR, the
larger the strain and consequent reduction. In this wind tunnel the vertical and horizontal contraction ratios
are 1.3 and 2.75, respectively, and the cross section area of the test section is 0.4  0.6m2 .
To determine the length of the test section is necessary firstly to analyze the flow coming out of the
contraction section which is comparable to that in an ideal free jet - see figure 7.1. A free jet flows into an
air space where there are no solid boundaries to influence its development. It can be divided in two regions:
the potential core region is immediately downstream the supply opening where the mixing of the jet fluid with
the surrounding is not complete. Its length usually extends 5 up to 10 equivalent opening diameters. In this
region the centerline velocity is constant and equal to the supply velocity. Test models should be placed within
that region. The other is the axisymmetric decay region; it is dominated by a highly turbulent flow generated
by viscous shear at the edge of the shear layer. The centerline velocity decreases inversely with the distance
from the opening. Baturin (1972) provided the following fit to experimental data.

um
u0
 ax{d 0.480.145 (7.1)
0
where um is the mean centerline velocity, u0 is the supply velocity, x is the distance from the supply, d0 is the
supply diameter and a is a constant whose value lies between 0.076 and 0.080 for cylindrical tubes.
Figure 7.2a shows the evolution of the centerline velocity for a 400  400mm square duct with equivalent
diameter deq  437mm. The maximum length of the potential cone is then given by the x-coordinate of the
intersection between the red and blue curves, x  1.8m, from which the centerline velocity starts decreasing.
Having estimated the potential cone length, it was possible to assess the dimensions of the test section. Figure
7.2b shows the boundary of the potential cone. The red shaded region corresponds roughly to the wake that
would be developed by a bluff body 150mm wide, a flat plate for instance. In this case the test section is
approximately 1.1m long. If instead a larger body is used, say 300mm wide, then its wake is represented by
the blue shaded rectangle and the length of the test section is reduced to less than half.

um {u0  0.078x{0.437
0.48 0.2 0.2
x

0.2
9
um {u0  1
0.145 x
3 9
z [m]

2
um
u0

0.2
0 1 2 3 4 5 0 0.5 1 1.5
x [m] x [m]

(a) Jet relative centerline velocity (b) Jet potential cone boundary

Figure 7.2: Test section dimension analysis

One of the main characteristics of jet flows is that the constant ambient pressure is imposed on the flow
field, and, as a result, the static pressure along its axis is constant. This is specially beneficial for experiments
with longer bodies. The greatest drawback, as it has been shown, is its limited usable length. The jet coming
out of the nozzle quickly mixes with the quiescent air in the surroundings and the width of its potential
core, in which the velocity profile is more uniform, decreases rapidly. Furthermore, turbulent mixing processes
at the boundary of the jet are more intense than in a turbulent boundary layer adjacent to a wall, and it

64
causes higher losses too. These turbulent structures modulate the flow noise around the test article and can
potentially damage the measurement of transient phenomena, in (Tropea et al. , 2007).

(a) Isometric view (b) Top view

Figure 7.3: Wind tunnel test apparatus

Figure 7.3 shows the wind tunnel test apparatus. The balance is placed as close as possible to the exit
of the contraction to make the most out of the usable test section length. All models were mounted on top
of the vertical strut of the balance on the upstream end of a squared tubular beam with variable length and
parallel to the wind tunnel axial direction. A symmetric streamlined profile was mounted around the model
support to reduce flow-structure interactions. In addition, an aluminum plate had to be installed at the exit,
which largely mitigates the vertical force induced on the balance by the pressure differential imposed on both
sides of the top platform.

7.2 Peripheral Systems and Test Conditions

Prior to wind tunnel testing, it is of at most importance to fully characterize the airflow and test conditions.
The quality of the flow and experimental techniques have a great impact on the test results; for example, the
aerodynamic performance of streamlined models is highly dependent on the behavior of the laminar boundary
layer, so low turbulence levels within the test section are necessary to ensure that laminar flow does not
prematurely transition to turbulent flow. The variation and angularity of the free-stream velocity are also
aspects of concern, as well as the model positioning and traverse systems, among others.
It is recommended to check the turbulence level, using for instance hot-wire anemometry, for the whole
range of Re. If the intensity is higher than a threshold value, say 0.1, then the anti-turbulence screens and
honeycomb setup should be revised. The variation of the free-stream velocity in the test section can be
obtained by comparing the dynamic pressure at a Pitot-static probe mounted near the exit of contraction with
that measured by a downstream probe. Barlow et al. (1999) states that the variation in dynamic pressure,
pdyn , in the working range of the test section should be less than 0.5% from the mean (rule of thumb). If
the probes are traversed in the YZ plane perpendicular to the airflow it is possible to draw a series of contour

65
maps of the dynamic pressure variation across the test section. In the same way, the uniformity and symmetry
of the velocity profile can be evaluated.
The dynamic pressure is given by the difference between the total and static pressures, respectively
ptotal and pstatic . A Pitot-static tube, illustrated in figure 7.4, manages to measure that difference.

Figure 7.4: Pitot-static tube schematic

pdyn  ptotal
pstatic  12 ρU 2 (7.2)

where, U is the free-stream velocity and ρ is the air density calculated from the ideal gas law.

ρ
Patm
(7.3)
RT

Just as important to have uniform flow velocity in the test section, is to have the flow aligned with
the axial direction. For the most part, Pitot-static probes are insensitive to flow angles in the range of 12 ,
so a large flow angle is required to introduce a discernible error in the dynamic pressure and total head
measurements. On the contrary, small changes in the flow direction contribute to a change in the attitude of
the model, which can skew the measurements when they are plotted as a function of pitch, roll or yaw. The
flow angularity can be measured using a five-hole probe, and according to Barlow et al. (1999) a flow angle
variation of 0.2 is acceptable.
Main flow disturbances are another aspect inherent to wind tunnel testing, which arise from the presence
of the test model in the limited test section. In this particular case, the jet expansion, the jet deflection, and
nozzle blockage, illustrated in figure 7.5, are the most important. Their effects in the pressure field can be
accounted for in the final result by applying correction factors, but this would require further investigation so
it has not been considered in the data reduction.

(a) Jet expansion (b) Jet deflection (c) Jet blockage

Figure 7.5: Expected major flow disturbances of the flow past a model. In Tropea et al. (2007)

The horizontal buoyancy is also not taken into account. According to textbooks, a free jet has the
same pressure as the plenum into which it is ejected. However, experimental measurements disclosed that in
fact this is not true. It is known that horizontal buoyancy adds a value ∆CD to the drag coefficient, but how
it is affected by the near wake of the body is still a matter of discussion. Further information regarding this
phenomenon can be found in Tropea et al. (2007).

66
 Wind tunnel testing is indeed a highly complex and time-consuming experiment. It employs a series
of peripheral systems to ensure that test conditions are within the desirable range. Ideally, all wind tunnel
test reports would include a detailed description of the procedures and test conditions under which they were
obtained, in addition to the uncertainties associated with the equipments employed. Very often though, as in
this case, it is not possible to control the test environment to such extend given the high cost of the equipment
in question. For this reason, it is noted that all data herein presented is compromised from the start since
none of the procedures mentioned above have been carried out. Nevertheless, it is important to keep in mind
that the main purpose of these experiments is only to access the performance of the balance and not to fully
characterize the airflow around the test articles.
A single pressure differential sensor was used to monitor the airflow velocity. A simple Pitot tube was
positioned at the exit of contraction, eccentric to the centerline 70mm vertically and horizontally. Since this
type of Pitot tubes include only one pressure inlet, it has been assumed that the static pressure in the test
section is constant and equal to the atmospheric, pstatic  patm . Therefore, the total pressure was measured
relatively to the atmospheric pressure. Equation 7.2 becomes

ptotal
patm  pdyn  12 ρU 2 (7.4)

The air density ρ was estimated using the table of the thermodynamic properties of air.
It is used the ultra low pressure differential sensor DC002NDC4 from Honeywell - the spec-sheet is
available on the website Honeywell (May 2015). It operates from a DC 5V supply and it has a maximum
pressure range of 5mBar. The offset voltage is 2.25V and the output span 2V . The accuracy in percentage
of Best Fit Straight Line (BFSL) is 0.25% (percentage of full-scale output), which includes non-linearity,
hysteresis, and repeatability.
The relation between the output voltage, V [mV ], and the dynamic pressure, Pdyn [P a], is expressed
by the following linear relationship.
U  2250 3.4036Pdyn (7.5)
This relationship was obtained from a least squares fit to the calibration data of the pressure sensor with a
coefficient of determination R2 of 0.9996. The calibration procedure and results are outlined in appendix E.
Equation 7.5 can be rewritten to be given as a function of the air-stream velocity instead.
1
3.4036 ρV 2
2
U  2250 (7.6)

From equation 7.6, and considering the performance specifications of the sensor, it is possible to plot its accur-
acy in measuring the airflow velocity against the airflow velocity. As illustrated in figure 7.6 the fundamental
relationship between pressure and velocity yields an inversely proportional evolution. As such, in terms of
velocity, the accuracy provided by this sensor is significantly deteriorated for lower values.
Airflow velocity accuracy [m{s]

1.5

0.5

0
0 5 10 15 20
Airflow velocity [m{s]

Figure 7.6: Pressure sensor performance

67
7.3 Description of the Tests

The experiments can be divided in three groups:

1. The first set consists in determining the drag coefficient of a flat plate facing the oncoming flow with an
aspect ratio of 1.5 and area 150cm2 . It is noted that since the plate is always normal to the air stream,
having an aspect ratio of 1.5 or 1{1.5 yields identical results. Two different setups were employed: initially
the test article was fixed so that its geometric center had the same Y Z coordinates as the resolving
center. This way, the balance only experiences force in the drag direction. On a second approach, the
flat plate was placed eccentrically to the resolving center, so in addition to the drag force, a bending
moment is also expected. The performance of the balance in decoupling the several components of the
aerodynamic load is then evaluated by not only statistically analyzing the results but also by estimating
the position of the pressure center of the plate, which, in this particular case, coincides with its geometric
center. Figure 7.7 illustrates both mounting setups.

(a) Centered with resolving center (b) Eccentric to resolving center -
50mm in
the ZZ direction

Figure 7.7: Flat plate mounting setup

2. A cylinder with 80mm diameter which spans the whole width of the wind tunnel, 400mm, was also
tested to assess the dynamic performance of the balance. The same fixation scheme was used to ensure
all time-averaged aerodynamic components, except for drag, would have a mean value of 0 (see figure
E.3 in appendix E). The main objective of this experiment is to verify if whether or not the balance is
able of measuring the dynamic loading resulting from the vortex shedding mechanism.

3. The drag coefficient of two bluff bodies in three configurations, usually tested by undergraduate students
for the course of Fluid Mechanics using a mechanical drag balance, are compared against those obtained
using the balance described in this document. The test procedure and results using the mechanical
balance can be found in Sarmento (2011). To establish a valid comparison between both methods, all
test conditions have been reproduced as well as the test articles whose fixation system does not suit the
digital balance. One of them is a I120mm and 20mm thick disk, to be tested facing the oncoming
flow, and the other is a I120mm and 150mm long cylinder oriented with the axial direction (see figures
E.4 and E.5). A third configuration is tested which consists in the latter body with an appendix placed
in front of it, a small cone.

68
7.4 Results - Rectangular Plate

This article was tested following two distinct approaches. Both hold data originated from a set of seven similar
runs: a test run consists in scanning the air-stream velocity between 4 and 20m{s, acquiring the output of all
six force transducers at predetermined steps. In this particular case an interval of 2m{s was used yielding a
total of 9 data points per run. At the beginning of each run all channels of the balance are zeroed out. The
first approach takes the first data-point at 4m{s and goes up to 20 in one go, that is, without turning off the
wind tunnel and taring the balance between measurements. Oppositely, the second one starts at 20m{s and
goes down to 4. In the end, a grand total of 126 data points have been acquired, over a period of 10 seconds
each. The test procedure is described in appendix E.
The Reynolds number, Re, based on a characteristic dimension of the test article, L, is given by

Re 
UL
(7.7)
ν
In the case of the rectangular plate, the square root of its projected area normal to the airflow is a suitable
characteristic length. ν is the dynamic viscosity of air and it is estimated using the table of its thermodynamic
properties.

7.4.1 Drag Coefficient

Figures 7.8 and 7.9 show, respectively, the scatter plots of the drag component data points in the coefficient
form as a function of the Reynolds number for the Up and Down approaches.

Run 1
1.06
Run 2
Drag cofficient, CD

Run 3
1.04 Run 4
Run 5
1.02 Run 6
Run 7
1

0.98
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Reynolds number, Re  10
5

Figure 7.8: Rectangular plate - Up approach scatter plot

1.2
Run 1
Run 2
Drag cofficient, CD

1.1 Run 3
Run 4
1 Run 5
Run 6
Run 7
0.9

0.8
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Reynolds number, Re  10
5

Figure 7.9: Rectangular plate - Down approach scatter plot

69
 The drag coefficient is
CD  ρU
2FX
2A
(7.8)

where, A is the reference area and FX is the mean value of the drag force over the acquisition period, during
which a total of 1500 samples were acquired.
Figure 7.10 shows the average of the drag coefficient of all seven runs for each Re, µCD i , as well as
the associated standard deviation, σCD i .
¸
7
µCD i  17 CD j,i ; i  1, 2, ..., 9 (7.9)

j 1

and g
f ¸
f 7
σCD i  e 17 pCD j,i
µCD i q2 ; i  1, 2, ..., 9 (7.10)
j 1

Assuming a normal distribution of the data points around the mean value, table 7.1 summarizes the
estimated uncertainty for a level of confidence of 95%. It is computed as the standard deviation multiplied by
the coverage factor k  1.96.

Table 7.1: Cd uncertainty 1.06

Up
as a 95% confidence range
Down
- drag counts 1.04
Drag cofficient, CD

Re  10
5
uCD 1.02
Up Down
0.40 34.5 217.8 1
0.52 28.2 95.5
0.98
0.70 21.2 54.1
0.90 8.0 39.6
0.96
1.07 7.8 25.9
1.26 8.0 21.6 0.94
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
1.42 5.3 13.7
1.60 4.5 10.6 Reynolds number, Re  10
5
1.79 7.4 5.5
Figure 7.10: Rectangular plate - drag coefficient results

Eyre et al. (1959) carried out a series of experiments on isolated flat plates perpendicular to the
airflow. It is shown, on their technical report, how drag, base pressure and flow pattern are affected by the
aspect ratio. The vortex shedding mechanism is also analyzed. Rectangular plates shed turbulent eddies at
particular frequencies; generally there are two shedding frequencies for each plate, one associated with the
smaller dimension of the pate, and a lower frequency associated with the larger dimension. For a rectangular
flat plate with an aspect ratio of 1.5, which is the same as the one tested, they state its drag coefficient lies
between 1.143 and 1.150.
By inspection, it can be promptly seen that the observed drag coefficients largely drift from the reference
values. The most likely cause has to do with the location of the Pitot tube used to estimate the airflow velocity.
It is located upstream of the test article at the exit of contraction, that is, in its vicinity where the velocity
field is significantly affected; the presence of the test model accelerates the oncoming flow as it is deflected
around it. So in fact, the undisturbed flow velocity is slightly lower than the measured value. The aerodynamic
coefficients are computed taking the undisturbed airflow velocity as a reference. Since the measured quantity
is inflated, it yields lower values than expected.

70
 1.5 Rect. Plate 100  150
y  0.0498x
0.0648

Velocity decrease without

Disk 120
y  0.0411x
0.0608
model rm{ss
1
Cylinder 80
y  0.0798x
0.0281

0.5

0
4 6 8 10 12 14 16 18 20
Airflow velocity with model rm{ss

Figure 7.11: Influence of the test articles on the airflow velocity

1.15
Up
Down
Drag cofficient, CD

1.1

1.05

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Reynolds number, Re  10
5

Figure 7.12: Rectangular plate - corrected drag coefficient

In order to workaround this problem the influence of each test article has been determined by measuring
the decrease in airflow velocity at the probing point when the model is removed from the test section. Figure
7.11 summarizes the results of all test articles as a function of the measured air-stream velocity when the model
is mounted on the balance. As expected, the cylinder test model has the greatest impact on the velocity field,
since it has the larger projected area. The sample data sets could be approximated by linear curve fits, which
are also represented in the chart, and the respective expressions provided a mean to correct the acquired data
relative to the air-stream velocity. For the case of the rectangular plate,

Ucorrected  p1
0.0498qU 0.0648 (7.11)

Figure 7.12 shows the corrected values of the drag coefficient for both approaches as a function of Re.
According to textbooks, the drag coefficient of bluff bodies gradually decreases with the Reynolds
number. This results from the reduction of the width of the wake developed behind it. The experimental
data, specially relatively to the Up approach, shows a good agreement of the drag coefficient results with the
mentioned behavior for higher Re. On contrary, the initial evolution does not follow the expected tendency,
and given the conditions under which the experiment was carried out, it is also complicated to assume or make
any suggestion on what could be the cause of this discrepancy.

71
7.4.2 Pressure Center

Given the adopted frame of reference and position of the model, the vertical coordinate of the pressure center
CP relative to the resolving center of the balance would be simply computed as the ratio between the pitch
moment and drag coefficients, respectively, CMY and CD . However, since it is not possible to guarantee its
perfect alignment with the airflow direction, it is also necessary to consider traces of vertical force that are
measured together. The force-moment static equilibrium equation around the resolving center of the balance
for the XZ plane is
CMY  CD z CZ x (7.12)

where, CZ is the vertical force coefficient, x is the horizontal coordinate of the pressure center and z is the
vertical coordinate.
According to the Joint Committee for Guides in Metrology JCGM (2008), the combined uncorrelated
standard uncertainty associated with a given function f pxi q is given by the positive square root of the combined
variance u2f , which is expressed as follows.

N !
)
u2 
¸ Bf 2
u2 pxi q

i 1
B xi (7.13)

Accordingly,



Bz
Bz
 BBCz Bz
2 2 2 2
u2z u2CD u2CM u2CZ u2x
D BCM Y
Y BCZ Bx (7.14)

where,

B z CZ x
CM Bz Bz Bz
BCD  CD2 BCM  CD ; B CZ 
CD ; Bx 
CD
Y
1 x CZ
; (7.15)
Y

The vertical coordinate of the pressure center is plotted in figure 7.13 as a function of the Reynolds
number. As expected, all data points from the Up approach except the first one can be found within a
relatively narrow band, between 52 and 55mm. On the other hand, those from the Down approach seem to
be in agreement with the reference value for higher Re, however they start drifting away from the resolving
center as the airflow velocity decreases.

Table 7.2: CP uncertainty 100

Down
as a 95% confidence range
Up
[mm]
Cp position rmms

80
Re  10
5
uz
Up Down
0.39 37.7 245.8
60
0.50 27.3 101.1
0.67 22.8 46.9
0.86 19.6 27.0
40
1.02 13.8 15.9
1.20 12.6 10.2
0.4 0.6 0.8 1 1.2 1.4 1.6
1.35 9.3 6.1
1.52 8.8 5.1 Reynolds number, Re  10
5
1.70 7.6 2.6
Figure 7.13: Rectangular plate - Pressure center -Z coordin-
ate

72
7.4.3 Experimental Uncertainty Estimation

The uncertainty associated with the experimental setup was estimated considering the error contributions of
all equipments employed. The method described in the previous section was used in a similar fashion to
determine the combined uncertainties associated with experimental measurements of pressure, temperature
and lengths among other physical quantities.
From equations 7.13 and 7.11, the following relation is developed for the uncertainty in the coefficient
of drag.

BCD 2
BCD 2
B CD 2
BCD 2
u2CD  u2FX u2U u2A 2
B FX BU BA Bρ uρ (7.16)

where,
B CD  2 ; BCD 
2FX ; BCD 
2FX ; BCD 
4FX
BFX ρAU 2 Bρ Aρ2 U 2 BA ρA2 U 2 BU ρAU 3
(7.17)

Each component ui of equation 7.16 was estimated. The error induced by the drag force measurement is
mainly due to calibration. If the balance is poorly calibrated, the accuracy of the balance is seriously damaged.
The fitting method, addressed in the previous chapter, provides information regarding the uncertainty of the
calibration parameters ci,j in the matrix form, cui,j .
 
14.75 4.47 8.66 1.01 1.07 0.87
14.70 
 4.44 7.48 1.30 1.29 1.08
 
14.81 1.26
Ĉu  
5.12 9.70 1.52 1.84   10
3
13.78 0.94
(7.18)
 2.99 6.60 1.01 0.85 
 
14.45 4.34 7.38 1.03 0.69 0.87
15.26 6.68 12.43 1.60 1.73 1.07

In the same manner that the calibration coefficients matrix C transforms the force transducers readings r into
aerodynamic load components, its corresponding uncertainty matrix Ĉu allows to compute the uncertainty
associated uŷ as follows.
uŷ  absprqĈu
1
(7.19)
2
The rest of the quantities involved are summarized in table 7.3.

Table 7.3: Uncertainty of measured variables

Quantity Uncertainty
Atmospheric Pressure rP as not monitored, 101325
Temperature r C s 1
Ruler rmms 1
Dynamic Pressure rP as 1.25
Air Density rkg {m3 s 5%a
a of the value @20 C and standard atmospheric pressure, considering
the air as an ideal gas

Based on the minimum resolution of the ruler and considering the dimensions of the rectangular plate, 150 
100mm, the uncertainty of the area is
a
uA  p0.150  0.001q2 p0.100  0.001q2  0.00018m (7.20)

73
 Finally, the air-stream velocity is expressed by equation 7.4 as a function of the dynamic pressure and
air density.
pdynamic  ρU 2
1
2
Therefore, according to the law of propagation of uncertainty,



 BBUρ BU
2 2
u2U u2ρ u2Pdyn
BPdyn (7.21)

where,
? ?
BU 
2 P
a dyn ;
BU  2
a
1
Bρ 2 ρ Pdyn {ρ
2 Bpdyn 2 ρ Pdyn {ρ
(7.22)

The multiple Y-axis chart in figure 7.14 shows five graphs; the greenish ones, on the left side in log
scale, deal with the four components of the combined uncertainty that is, in its turn, represented on the right
side by the blue graph.

ρ parcel
10
2 A parcel
140 FX parcel
U parcel
uCD

uCD [Drag counts]

u2xi

120
10
3
2
B CD
Bxi

100

10
4
80

6 8 10 12 14 16 18 20
Air-stream velocity rm{ss

Figure 7.14: Experiment Uncertainty Estimation

This analysis shows that for lower air-stream velocities, below 8m{s, the error induced by the velocity
measuring system, which involves air density estimation and accuracy of the pressure sensor, clearly dominates
the uncertainty associated with the measurement of drag. In this interval its value drops rapidly, and then
the U variance looses importance in relation to ρ and FX variances, which are mostly constant across the
whole domain. As a result, the combined uncertainty keeps decreasing as the U component decreases, but at
a much slower rate. The contribution of the area uncertainty, variance of A, is reduced in comparison so it
does not play an important role on the evolution of the combined quantity.
From the results of figure 7.14 it is possible to identify the aspects which hold the greatest potential
for improvement of the whole system. It can been seen that the uncertainty at lower air-stream velocities is
more effectively reduced if the accuracy of the velocity measuring system was improved. It should be noted
again that this is a function of both the pressure sensor accuracy and air density estimation. So, given the
importance of the ρ parcel, the most reasonable initial strategy would be simply to improve the accuracy of
the latter. Besides this change, at higher velocities the uncertainty of the balance could only be reduced by
recalibrating the balance with an improved calibration system.

74
 Figure 7.15 shows the values read by the balance after each test run for both approaches. As it can be
seen, it is not possible to discern a zero-load pattern in the readings showing a negligible hysteresis effect.

2
Down
Up

Drag component rg s 0

2

4

Run 1 Run 2 Run 3 Run 4 Run 5 Run 6 Run 7

Figure 7.15: Rectangular plate - offset at the end of runs

75
76
Chapter 8

Conclusion and Future Work

8.1 Conclusion

The main objective of this master thesis was to design and manufacture a six-component external wind tunnel
balance that could stand has an economical alternative to the existing commercial solutions, and provide
support for research work and undergraduate teaching. The total cost of the prototype is roughly 1100 e, which
was financed by the Scientific Area of Thermofluids and Energy Conversion Technologies and the Scientific
Area of Computational and Structural Mechanics. The mechanical structure of the device was sponsored by
Projecto FST Novabase. It should be highlighted that an equivalent product, such as the pyramidal balance
developed by Aerolab (May 2015), which is a company specialized in preparing complete wind tunnel facilities,
in particular for educational institutions, costs over 105.000 e.
In order to reduce the manufacturing cost it was necessary to design from scratch every subsystem,
which significantly broadened the scope of the project. Also, as this is the first device of its kind ever designed
and built at IST, reliability, ease of use and maintenance were privileged and are therefore reflected on the
design decisions. As a result, despite providing satisfactory results, there is still a large margin for improvement
in almost every aspect. More complex alternatives to the many solutions employed are suggested throughout
the document which can ultimately improve its overall performance.
A calibration system based on a set of pulleys and weights was developed to calibrate the balance. The
calibration data points acquired were then successfully interpolated by a first-order weighted fitting function
and the uncertainty of the calibration parameters estimated. Since the DAQ was designed so that the balance
was capable of achieving an accuracy of 2 drag counts (roughy 0.1% of the full-scale load), and given the
values presented in section 8.2, the calibration process is in fact what is currently limiting its performance.
The DRS proved to be a valuable tool to expedite both wind tunnel tests and the calibration process by
keeping the acquired raw data organized and providing a friendly user interface; it outputs, for each acquisition
period, a set .txt files to be later post-processed. The serial port reading function of MatLab revealed not to
be very efficient speed-wise limiting the data transfer from the DAQ to the PC. To workaround this issue, it
would be necessary to turn to other software packages, more suited to serial communication such as LabVIEW.
The analysis to the uncertainty of the wind tunnel data system, which includes the balance and other
peripheral systems, showed that it is significantly larger than the perceived through experimental investigation.
Furthermore, it provided information regarding the weight of each contribution to the error and how they vary
as a function of the Reynolds number. From this, its was possible to conclude that in order to reduce the
combined uncertainty the most efficient solution is to improve the air density monitoring system. Alternatively,
it would be required to work on an improved version of the calibration rig, which is clearly a more expensive

77
and time-consuming approach. The pressure differential transducer from Honeywell offered an acceptable
accuracy for higher airflow velocities, at lower speeds, however, the inherent uncertainty becomes too large in
comparison with the other components.
It was found a good agreement of the experimental data of the rectangular flat plate with values
published by Eyre et al. (1959), specially for higher Re. It was also possible to estimate the position of the
center of pressure of the test model. On the other hand, the test of the rigid circular cylinder transverse to the
airflow, which was meant to determine if whether or not the balance is able to measure the dynamic loading
induced by the vortex shedding mechanism, did not yield meaningful results. The frequency spectrum of the
acquired signal is heavily polluted with white noise and shows a strong peak at 20Hz that might be related
to the first natural frequency of the top platform, as predicted in the modal analysis. To overcome this, it is
necessary to revise the top platform design, specially the anchoring region of the vertical strut - rebuild it with
a different fiber layup or reinforce it by including a set of backing plates on each side.

8.2 Features of the Balance

• 6-6 in-parallel mechanism balance system:

- 6-component measurement ability;
- ‘Single strut design, m6  20mm male thread at fixation point;
- Equipped with Pitch, Yaw and Roll angle sensor, accurate to 0.1 ;
- Overall dimensions: 700mm long, 700mm wide, 715mm high (the vertical strut is 300mm long
and the device body is 415mm high).
- Floor space required, 800  800mm;
- Minimum required space below test section - 430mm;
- 4 leveling feet with 120mm long thread;

• Load limits (maximum single loading, absolute values):

- Normal force: 50N  0.4%
- Axial force: 30N 2.1%
- Side force: 30N  0.7%
- Roll moment: 2.5N.m  5.9%
- Yaw moment: 2.5N.m  2.0%
- Pitching moment: 2.5N.m  3.6%
The uncertainty values are given as a 95% interval of confidence and are a result from the calibration
process described in this document.

• Data acquisition
- 6 force transducers;
- 6 input channels, RJ-12 6P6C receptacles, with dedicated ultra-low noise 24-bit sigma-delta ADC;
- Acquisition frequency: 150Hz;
- Type of excitation: AC square wave, 150Hz
- Amplification gain: 128;

78
8.3 Future Work

This section suggests a series of modifications to the current design of the balance which enhance its per-
formance and workability. These are not meant to improve the already built prototype, instead they imply the
construction of a whole new version. Most of them do not come over an increase of the cost, in fact, if all
suggestions were implemented the latter would be marginally lower.
The structure of the balance should be redesigned as an hyper-static form of the Stewart Platform,
illustrated in figure 3.1b. This concept does not rely on spherical pairs at the end of each force transducer,
thereby eliminating the source of stress coupling in the sensing bridge, as well as hysteresis effects. In addition,
it drastically reduces the cost of the equipment since rod-ends are no longer required. An improved version of
the balance could also benefit from spatial isotropy which leads to approximately the same accuracies for all
aerodynamic load components.
Different shapes for the sensing elements can also be explored. The ring-shaped ones seem to be a
promising alternative; their design can be easily parameterized and the sensitivity to compliance ratio optimized
according with the design requirements. It also allows to mechanically amplify bridge signals by a factor of
two in comparison to tubular elastic elements, resulting in a direct improvement of the signal-to-noise ratio.
The DAQ follows a reduced to essentials philosophy therefore, it was built with the sole purpose of
conditioning the signal from the force transducers. There are, however, a series of other peripheral sensors
necessary to conduct experiments on the wind tunnel, such as dynamic and atmospheric pressure transducers,
temperature and humidity sensors to monitor both the air inside and out the wind tunnel, and an Inertial
Measurement Unit (IMU) to estimate the attitude of the model in relation to the wind tunnel frame of
reference. It would be then a major asset if an integrated and more versatile acquisition unit was developed
which could handle the signals of all sensors mentioned above.
Without changing any aspect of the balance it is still possible to increase its accuracy if the calibration
system is reviewed. A semi-automated calibration rig comprising dedicated step-motors, to move and position
the pulley carriers in space, would most certainly provide higher precision than the current design, which
is based on visual inspection. Even more because the calibration loading combinations have to be applied
repeatedly under similar conditions to achieve statistical significance.
Finally, the DRS should be implemented on a more powerful platform such as LabVIEW, suited to serial
processing. As it is, there is no room left to expand the reduction system, as suggested above for the DAQ,
without inducing a delay in relation to the stream of data. A bi-directional communication protocol between
the acquisition system and the user interface, instead of the one-way configuration presently employed, would
also provide more flexibility to the user allowing to control acquisition parameters, such as the type of excitation
signal, filtering and sampling frequency.

79
80
Appendices

81
Appendix A

Budget

Table A.1: WTB_Ampv1.0 list of components

Entry Comment Description Designator Footprint Quantity Distributor Price per unit Price [ e]
1 1MOhm Resistor R1 SMD 0603 6 Digikey 0.1 0.6
2 1MOhm Resistor R2 SMD 0603 6 Digikey 0.1 0.6
3 1MOhm Resistor R3 SMD 0603 6 Digikey 0.1 0.6
4 1MOhm Resistor R4 SMD 0603 6 Digikey 0.1 0.6
5 100R Resistor SDI_con SMD 0603 6 Digikey 0.1 0.6
6 100R Resistor SDO_con SMD 0603 6 Digikey 0.1 0.6
7 100R Resistor SCK_con SMD 0603 6 Digikey 0.1 0.6
8 100R Resistor SS_con SMD 0603 6 Digikey 0.1 0.6
9 0R Resistor Shunt SMD 0603 6 Digikey 0.1 0.6
10 BMB2A1000LN2 Ferrite Bead Ferrite Bead SMD 0805 6 Digikey 0.14 0.84
11 10uF Capacitor Cap_S5_1 SMD 0603 6 Digikey 0.19 1.14
12 100nF Capacitor Cap_S5_2 SMD 0603 6 Digikey 0.10 0.60
13 100nF Capacitor Cap_S5_3 SMD 0603 6 Digikey 0.10 0.60
14 4.7uF Capacitor Cap_S5_4 SMD 0603 6 Digikey 0.1 0.6
15 10uF Capacitor Cap_S3_1 SMD 0603 6 Digikey 0.19 1.14
16 100nF Capacitor Cap_S3_2 SMD 0603 6 Digikey 0.10 0.60
17 100nF Capacitor Cap_S3_3 SMD 0603 6 Digikey 0.10 0.60
18 4.7uF Capacitor Cap_S3_4 SMD 0603 6 Digikey 0.1 0.6
19 1uF Capacitor C1_Ref SMD 0603 6 Digikey 0.1 0.6
20 10nF Capacitor C2_Ref SMD 0603 6 Digikey 0.1 0.6
21 10nF Capacitor C3_Ref SMD 0603 6 Digikey 0.1 0.6
22 1nF Capacitor C1 SMD 0603 6 Digikey 0.1 0.6
23 10nF Capacitor C2 SMD 0603 6 Digikey 0.1 0.6
24 1nF Capacitor C3 SMD 0603 6 Digikey 0.1 0.6
25 1nF Capacitor C4 SMD 0603 6 Digikey 0.1 0.6
26 10nF Capacitor C5 SMD 0603 6 Digikey 0.1 0.6
27 1nF Capacitor C6 SMD 0603 6 Digikey 0.1 0.6
28 10uF Capacitor DS_C1 SMD 0603 6 Digikey 0.19 1.14
29 100nF Capacitor DS_C2 SMD 0603 6 Digikey 0.10 0.60
30 10uF Capacitor AS_C1 SMD 0603 6 Digikey 0.19 1.14
31 100nF Capacitor AS_C2 SMD 0603 6 Digikey 0.1 0.6
32 ADP33035V IC ADP33035V SOIC8 6 Farnell 2.79 16.74
33 ADP33033.3V IC ADP33033.3V SOIC8 6 Farnell 2.78 16.68
34 MIC4427 IC MIC4427_i SOIC8 12 Farnell 1.05 12.6
35 AD7195 IC AD7195 32Lead LFCSP_WQ 6 Farnell 12.73 76.38
36 Header 5X2A Connector Bridge 5Pin, Dual row, Male 6 Digikey 0.27 1.62
37 Header 6A Connector SPI 6Pin, Male 6 Digikey 0.64 3.84
38 35um Cu, 0.11dm2 PCB WTB_Ampv1.0 42.7x24.99mm 15 Multi Circuit Boards 6.08 91.25

82
 Table A.2: WTB_PDModulev1.0 list of components

Entry Comment Description Designator Footprint Quantity Distributor Price per unit Price [ e]
1 Header 5X2A Connector Bridge 5-Pin, Dual row, Female 6 Digikey 0.85 5.1
2 Header 6A Connector SPI 6-Pin, Female 6 Digikey 0.7 4.2
3 Header 10X2A Connector dsPIC, Pan 10-Pin, Dual row, Female 1 Digikey 1.55 1.55
4 Header 10X2A Connector dsPIC, PDIG1 10-Pin, Dual row, Female 1 Digikey 1.55 1.55
5 Header 8X2A Connector dsPIC, PDIG2 8-Pin, Dual row, Female 1 Digikey 1.26 1.26
6 Header 8X2A Connector dsPIC, PCAN 8-Pin, Dual row, Female 1 Digikey 1.26 1.26
7 Header 4X2A Connector Feed12 4-Pin, Dual row, Male 1 Digikey 0.86 0.86
8 Header 2X2A Connector Buttons 2-Pin, Dual row, Male 1 Digikey 0.16 0.32
9 Header 6X2A Connector Leds 6-Pin, Dual row, Male 1 Digikey 0.96 0.96
10 Header 4X2A Connector Feed12 4-Pin, Dual row, Female 1 Digikey 0.85 0.85
11 Header 2X2A Connector Buttons 2-Pin, Dual row, Female 1 Digikey 0.5 1
12 Header 6X2A Connector Leds 6-Pin, Dual row, Female 1 Digikey 1 1
13 Header 5A Connector PIC Prog 5-Pin, Male 2 Digikey 0.5 1
14 Header 3A Connector UART 3-Pin, Male 2 Digikey 0.27 0.54
15 Header 2A Connector Connector to Leds 2-Pin, Male 3 Digikey 0.49 1.47

83
16 Header 5A Connector PIC Prog 5-Pin, Female 2 Digikey 0.67 1.34
17 Header 3A Connector UART 3-Pin, Female 2 Digikey 0.66 1.32
18 Header 2A Connector Connector to Leds 2-Pin, Female 3 Digikey 0.5 1.5
19 Diode, Schottky, 1A, 20V Diode Diode SOT-23 1 Farnell 0.129 0.645
20 JACK SMT, SIDE ENTRY, 6/6 Connector Con_StrainGages RJ-25 6 Farnell 1.99 11.94
21 JACK, SMT, SIDE ENTRY, 6/6 Connector PIC Prog connector RJ-25 1 Farnell 1.99 1.99
22 FTDI UB232R FTDI PIC-PC com 1 Farnell 16.63 16.63
23 Multicomp switch, IP67, Black Push Button Push buttons DAQ 12mm diameter 2 Farnell 3.97 7.94
24 Switch toggle Toggle Button on/off button 1/4 diameter 1 Farnell 1.92 1.92
25 Red LED through-hole LED Warning LED’s T-1 3/4 5mm diameter 6 Farnell 0.173 1.038
26 Green LED through-hole LED On/off, Error T-1 3/4 5mm diameter 2 Farnell 0.738 1.476
27 Blue LED through-hole LED Acquiring T-1 3/4 5mm diameter 1 Farnell 0.212 1.06
28 Schurter IEC Connector AC supply plug 1 Farnell 7.4 7.4
29 Fan 12VDC Fan Cooling 40x40x10 mm 2 Farnell 5.7 11.4
30 TXM 035-112 AC/DC Power supply 1 Farnell 23.95 23.95
31 35um Cu, 2.35dm2 PCB WTB_PDmodulev1.0 258x91mm 1 Total Electrica 60.00 60
 Table A.3: WTB Force transducers list of components
Comment Description Quantity Distributor Price per unit Price [ e]
1-XY13-3/120 Strain Gage T-Rosette 2 HBM 128 256
1-NG150 Strain Gage protector 1 HBM 50 50
1-Z70 Strain Gage Adhesive 1 HBM 50 50
PLUG, IDC, RJ12, FLAT, 6P6C Connectors 6 Farnell 0.254 2.54
CABLE, 28AWG, 2 PAIR, PER M Wiring 7 Farnell 6.13 42.91

Table A.4: WTB Mechanical structure list of parts

Item no. Name Material Quantity Distributor Price per unit Price [ e]
1 WTB_Bottom_A 1 Projecto FST
1.1 WTB_Bottom_Foam RohaCell 71WF 1
1.2 WTB_Bottom_SteelSheet 1
1.3 WTB_Bottom_Insert 4
1.4 WTB_Bottom_CFLayer 150 g/sqm +45/-45 2

2 WTB_LevelingFeet_LevelingFeet 4 Reiman LV.A 28.7 114.8

-70-14-M16x78-ELK
2.1 WTB_LevelingFeet_Base 1
2.2 WTB_LevelingFeet_Screw 1
2.3 WTB_LevelingFeet_Rubber 1
3 WTB_Bottom_Bracket 6 Projecto FST
4 WTB_Bottom_BackingPlate 2 Projecto FST
5 WTB_Top_A 1 Projecto FST
5.1 WTB_Top_Foam Airex 1
5.2 WTB_Top_CFLayer CF 150g/sqm 0/+45/-45 2
5.3 WTB_Top_BackingPlate 6
6 WTB_Top_Bracket 6 Projecto FST
7 WTB_Rod_A 6 Projecto FST
7.1 WTB_Rod_Rod 1
7.2 WTB_Rod_Insert 2
7.3 WTB_Rod_EARM_05_3 1 IGUS EARM_05 4.44 26.64
7.4 WTB_Rod_EALM_05_3 1 IGUS EALM_05 4.44 26.64
8 WTB_Support_A.SLDPRT 1 Projecto FST
8.1 WTB_Support_Beam19 1
8.2 WTB_Support_InsertM19 1
8.3 WTB_Support_InsertF14 1
8.4 WTB_Support_Beam14 1
8.5 WTB_Support_InsertM14 1
8.6 WTB_Support_InsertF10 1
8.7 WTB_Support_Beam10 1
8.8 WTB_Support_InsertM10 1
8.9 WTB_Support_InsertF19 1
8.10 WTB_Support_Pointer 1
9 WTB_Shield_A 1 Projecto FST
9.1 WTB_Shield_Rod 1
9.2 WTB_Shield_InsertBot 1
9.3 WTB_Shield_Beam19 1
9.4 WTB_Shield_InsertTop 1
9.5 WTB_Shield_BackingPlate 1
9.6 WTB_Shield 2

Table A.5: Cost of the balance

with VAT 23% Cost of sub-assembly [ e] Quantity Cost per part [ e]
WTB_Ampv1.0 240.11 6 40.0
WTB_PDmodulev1.0 174.47 1 174.5
Force Transducers 401.45 6 66.9
Mechanical structure 168.1 1 168.1
Calibration System 130 1 130.0
Total cost 1114.1
Number of parts 390

84
Appendix B

Structural

Transformation Matrix
To compute the Plücker line coordinates of the leg connectors of the balance, Ŝi :
Matrices of the parameterized Plücker coordinates of the spherical pairs located at the bottom and
top platforms, respectively, Ui,j and ui,j . The subscript i identifies the leg connector and j the coordinate
number.
! )
U1,j  ϕ1
2
1, R1 cosp
q, R1 sinp q, 0
ϕ1
2
! )
U2,j  1, R1 cosp
q , R1 sinp
q
2π ϕ1 2π ϕ1
,0
3 2 3 2
! )
U3,j  1, R1 cosp q , R1 sinp q
2π ϕ1 2π ϕ1
,0 (B.1)
3 2 3 2
! )
U4,j  1, R1 cosp
q , R1 sinp
q
4π ϕ1 4π ϕ1
,0
3 2 3 2
! )
U5,j  1, R1 cosp 3 3
2 q, R1 sinp 3 3
2 q, 0
4π 2π ϕ1 4π 2π ϕ1
! )
U6,j  1, R1 cosp
q, R1 sinp
q, 0
ϕ1 ϕ1
2 2

u1,j  t1, R2 cosp ϕ22 q, R2 sinp ϕ22 q,
Hc u

u2,j  t1, R2 cosp 2π3
ϕ22 q, R2 sinp 2π3
ϕ22 q,
Hc u
u3,j  t1, R2 cosp 2π3 ϕ2
2
q , R2 sinp
2π ϕ2
3 2
q,
Hc u (B.2)

u4,j  t1, R2 cosp
q , R2 sinp
2 q,
Hc u

4π ϕ2 4π ϕ2
3 2 3
u5,j  t1, R2 cosp 4π3 2π3
ϕ22 q, R2 sinp 4π3 2π3
ϕ22 q,
Hc u
u6,j  t1, R2 cosp
q, R2 sinp
q,
Hc u
ϕ2 ϕ2
2 2
Ŝi  rSi S0i s is given by the second order determinants of the following matrix
 
1 ui,1 ui,2 ui,3
(B.3)
1 Ui,1 Ui,2 Ui,3

 T
Ŝi  Ui,1
ui,1 Ui,2
ui,2 Ui,3
ui,3 ui,2 Ui,3
Ui,2 ui,3 ui,3 Ui,1
ui,1 Ui,3 ui,1 Ui,2
ui,2 Ui,1

85
 Material Properties
Table B.1 gives the properties of all materials employed. In the case of the carbon fiber, the elastic limit
matches the rupture limit and the rupture strength is different when loading is applied along the longitudinal
direction or along the transversal direction (Ferreira, 2013).

Table B.1: Material properties

Carbon HT Rohacell IG71 Al 2024 T3

Specific Mass [kg {m3 ] 1530 75 2700
Tensile Yield strength [M P a] 290
Longitudinal tensile fracture strength [M P a] 1270
Longitudinal compressive fracture strength [M P a] 1130
Transverse tensile fracture strength [M P a] 42
Transverse compressive fracture strength [M P a] 141
Shear strength [M P a] 1.3 283
In plane shear strength [M P a] 63
Interlaminar shear strength [M P a] 90
Elastic modulus [M P a] 92 73100
Longitudinal elastic modulus [M P a] 134
Transverse elastic modulus [M P a] 7000
Shear modulus [M P a] 4200 29 28000
Poisson Ratio 0.25 0.37 0.33

(a) CNC drilling of the holes on the bot- (b) Detail of the rod-end assembly
tom platform at Fablab

Figure B.1: Manufacture

86
 8.8
8.5
8.7
8.4
8.6
9.6

300
8.3
8.1

A B ITEM NO. PART NUMBER DESCRIPTION QTY.

1 WTB_Bottom_A 1
1.1 WTB_Bottom_Foam RohaCell 71WF 1
730

1.2 WTB_Bottom_SteelSheet 1
1.3 WTB_Bottom_Insert 4

1.4 WTB_Bottom_CFLayer
150 g/sqm 2
+45/-45

300
2 WTB_LevelingFeet_LevelingFeet 4
2.1 WTB_LevelingFeet_Base 1
3 2.2 WTB_LevelingFeet_Screw 1
2.3 WTB_LevelingFeet_Rubber 1
1.1 3 WTB_Bottom_Bracket 6
4 WTB_Bottom_BackingPlate 6
2.2 5 WTB_Top_A 1
5.1 WTB_Top_Foam Airex 1
315 2.1 350 CF 150g/sqm
5.2 WTB_Top_CFLayer 2
350 0/+45/-45
9.4 9.3 9.5 9.2 5.3 WTB_Top_BackingPlate 6
2.3
9.1 6 WTB_Top_Bracket 6
7 WTB_Rod_A 6
5.1 7.1 WTB_Rod_Rod 1
8.9
7.2 WTB_Rod_Insert 2
7.3 7.3 WTB_Rod_EARM_05_3 2
8 WTB_Support_A.SLDPRT 1
7.2 10 8.1 WTB_Support_Beam19 1
6 20 8.2 WTB_Support_InsertM19 1
8.3 WTB_Support_InsertF14 1
12

10
7.1 8.4 WTB_Support_Beam14 1
8.5 WTB_Support_InsertM14 1
8.6 WTB_Support_InsertF10 1
DETALHE A 8.7 WTB_Support_Beam10 1
Escala 2 : 5 8.8 WTB_Support_InsertM10 1
DETAIL B 8.9 WTB_Support_InsertF19 1
SCALE 2 : 5
8.10 WTB_Support_Pointer 1
9 WTB_Shield_A 1
9.1 WTB_Shield_Rod 1
300
550

5.3
600
700
60°

9.2 WTB_Shield_InsertBot 1
9.3 WTB_Shield_Beam19 1
12

9.4 WTB_Shield_InsertTop 1
4 9.5 WTB_Shield_BackingPlate 1
9.6 WTB_Shield 1
NOME ASS. DATA
PROJ: Manuel Ferreira 66318 14/05/2015
DES: Manuel Ferreira 66318 29/07/2014
12 REV:

MATERIAL: MASSA: 16200 g

0°
12

WTB_A Tipo de Documento:

WTB_assembly
600
Qtd: Número Desenho: Escala:
700
1 0 ISO 1:5
Appendix C

DAQ and DRS

PCB Production Masks and Schematics

(a) Top layer (b) Bottom layer

Figure C.1: WTB_Amp v1.0 production masks (not to scale)

Figure C.2: WTB_Ampv0.2 mounted on test module v0.1

88
(a) Top layer (b) Bottom layer

Figure C.3: WTB_PDmodule v1.0 production masks (not to scale)

89
 1 2 3 4

Amp1 Amp2 Amp3

SDO SDO SDO
SDO SENSE2 (+) SDO SENSE2 (+) SDO SENSE2 (+)
SDI SDI SDI
SDI AIN4 SDI AIN4 SDI AIN4
SCK SCK SCK
SCK SENSE2 (-) SCK SENSE2 (-) SCK SENSE2 (-)
CS1 CS2 CS3
SS AIN3 SS AIN3 SS AIN3
A GND SENSE(-)_1 GND SENSE(-)_2 GND SENSE(-)_3 A
GND SENSE1 (-) GND SENSE1 (-) GND SENSE1 (-)
VCC5_1 WB(-)_1 VCC5_1 WB(-)_2 VCC5_1 WB(-)_3
VCC AIN2 VCC AIN2 VCC AIN2
SENSE(+)_1 SENSE(+)_2 SENSE(+)_3
AINCOM SENSE1 (+) AINCOM SENSE1 (+) AINCOM SENSE1 (+)
WB(+)_1 WB(+)_2 WB(+)_3
NC AIN1 NC AIN1 NC AIN1
WTB_Ampv1.0 WTB_Ampv1.0 WTB_Ampv1.0

Amp6
Amp4 Amp5 SDO
SDO SDO SDO SENSE2 (+)
SDO SENSE2 (+) SDO SENSE2 (+) SDI
SDI SDI SDI AIN4
SDI AIN4 SDI AIN4 SCK
SCK SCK SCK SENSE2 (-)
SCK SENSE2 (-) SCK SENSE2 (-) CS6
CS4 CS5 SS AIN3
SS AIN3 SS AIN3 GND SENSE(-)_6
GND SENSE(-)_4 GND SENSE(-)_5 GND SENSE1 (-)
GND SENSE1 (-) GND SENSE1 (-) VCC5_2 WB(-)_6
VCC5_2 WB(-)_4 VCC5_2 WB(-)_5 VCC AIN2
VCC AIN2 VCC AIN2 SENSE(+)_6
SENSE(+)_4 SENSE(+)_5 AINCOM SENSE1 (+)
AINCOM SENSE1 (+) AINCOM SENSE1 (+) WB(+)_6
WB(+)_4 WB(+)_5 NC AIN1
NC AIN1 NC AIN1
WTB_Ampv1.0
WTB_Ampv1.0 WTB_Ampv1.0
B B

AN10

AN12

GND
AN2

AN4

AN6

AN8
dsPIC_FST *
UART
GND
AN1/VREF-
AN2
AN3
AN4
AN5
PGC/AN6
PGD/AN7
AN8
AN9
AN10
AN11
AN12
AN13
AN14
AN15

AGND
AVCC
AN0/VREF+

3
TXD
2
RXD
CS1 1
RD0 RC1
RD1 RC2 Header 3H
CS2
RD2
PAN SCK1
RD3 RG13
CS3 RXD
RD4 SDO1
RDY_R TXD
RD5 SDI1
CS4
RD6 RG15
Prog_R
RD7 RG12
PDIG2
C CS5 C
RD8 SRX
PDIG1

INT2 dsPIC FST

RD9 STX
CS6 interface
RD10 SCL
INT4
RD11 RG14
SCK
SCK2 SDA
SDO
SDI2 VPP
SDI GND
SDO2 GND
VCC5_2
SS2 PVCC
CANI TX
PGC
CANI RX PGC/AN6
ICD2

GND PGD
GND PGD/AN7
VCC5_1 GND
PVCC GND
PCAN VCC Title:
VCC
VPP
CANH1

CANH2
CANL1

CANL2

VPP
WTB_PDmodulev1.0
VBAT
VBAT
VBAT

VBAT
VBAT
VBAT
PM
PM
PM

PM
PM
PM

Size: Number: Designed by:

D
A4 1 Manuel Aguiar Ferreira D
VCC_diode VCC_diode
GND GND
Date: Sheet: 1 of
April 2015 2
File: Revision:

1 2 3 4
 1 2 3 4

Con1 Con2 Con3

GND GND GND PIC Prog.
8 8 8
GND GND GND Feed12 VPP
7 7 7 5
GND GND GND GND Buttons VCC
6 6 6 1 5 VCC_12 4
WB(+)_1 WB(+)_2 WB(+)_3 INT2 INT4 GND
5 5 5 2 6 1 2 3
A GND GND GND VCC5_2 VCC5_2 PGD A
4 4 4 3 7 3 4 2
WB(-)_1 WB(-)_2 WB(-)_3 PGC
3 3 3 4 8 1
SENSE(+)_1 SENSE(+)_2 SENSE(+)_3 Header 2X2
2 2 2
SENSE(-)_1 SENSE(-)_2 SENSE(-)_3 Header 4X2A Header 5H
1 1 1 C1_Toggle C2_Toggle
INT2 INT4
RJ25 RJ25 RJ25 GND GND
Capacitor Capacitor
Con4 Con5 Con6 0.1uF 0.1uF
GND GND GND
8 8 8
GND GND GND
7 7 7
GND GND GND
6 6 6
WB(+)_4 WB(+)_5 WB(+)_6
5 5 5
GND GND GND Diode
4 4 4
WB(-)_4 WB(-)_5 WB(-)_6
3 3 3 VCC_12 VCC_diode ProgLED RDY_LED
SENSE(+)_4 SENSE(+)_5 SENSE(+)_6
2 2 2
SENSE(-)_4 SENSE(-)_5 SENSE(-)_6
1 1 1 SD350
Prog_LED RDY_LED
RJ25 GND GND
RJ25 RJ25 PowerON_LED
LED0 LED0
B B
LED_PowerON
GND Prog_R RDY_R
Prog_R Prog_LED RDY_R RDY_LED
LED0
Resistor Resistor
100R 100R
PowerON_R
VCC5_1 LED_PowerON
Resistor
100R

R_L6 R_L6 AN12

Resistor
100R
R_L5 R_L5 AN10
Resistor
Leds 100R
R_L6 R_L4 AN8
1 7 R_L4
R_L5
C 2 8 Resistor C
R_L4
3 9 100R
R_L3
4 10 R_L3 AN6
R_L2 R_L3
5 11
R_L1
6 12 Resistor
100R
Header 6X2A
GND R_L2 R_L2 AN4
Resistor
100R
R_L1 R_L1 AN2
Resistor
100R Title:
WTB_PDmodulev1.0
Size: Number: Designed by:

D
A4 1 Manuel Aguiar Ferreira D

Date: Sheet: 2 of
April 2015 2
File: Revision: *
1 2 3 4
 1 2 3 4

Bridge SPI
Sense2 (+) AIN4 VCC5
1 2 6
Sense2 (-) AIN3 DGND
3 4 5
Sense1 (-) AIN2 SS_con
5 6 4

MCLK2
MCLK1
Sense1 (+) AIN1 SCK_con
7 8 3
AVDD DVDD SDI_con

SDO
SCK
AINCOM

SDI
A 9 10 2 A

SS
SDO_con
1
Header 5X2H
DS_C1 DS_C2 AS_C1 AS_C2 Header 6H
AD7195 Capacitor Capacitor Capacitor Capacitor

MCLK2

MCLK1
CS AD7195 10uF 0.1uF 0.1uF 10uF
SCLK
MCLK2
MCLK1
DIN
DOUT/RDY
NC
SYNC
AGND AGND DGND DGND
ACX2 DVDD
_ACX2_ DVDD C_clk2 Cristal C_clk1
_ACX2_ AVDD Shunt
ACX2 AVDD DGND AGND
ACX1 DGND DGND DGND
_ACX1_ DGND Resistor
_ACX1_ AGND
ACX1 AGND Capacitor Cristal Capacitor
AVDD
AVDD BPDSW
BPDSW _ACX1_ R1 xx Cristal xx
0Ohm
AGND AGND
AGND NC Resistor
Sense1 (-)
NC REFIN(-) 1MOhm
AINCOM Sense1 (+) C2_Ref C1_Ref C3_Ref
AINCOM REFIN(+)
Sense1 (+) AGND Sense1 (+) Sense1 (-) Sense1 (-) AGND
_ACX2_ R2
AIN1
AIN2

AIN3
AIN4

AVDD
NC
NC
NC
NC

Capacitor Capacitor Capacitor

Resistor
0.01uF 1uF 0.01uF
B 1MOhm B
ACX1 R3 C1 C2 C3
AGND AIN1 AGND AIN1 AIN2 AIN2 AGND
Resistor
AIN1
AIN2

AIN3
AIN4

1MOhm
Capacitor Capacitor Capacitor
1nF 0.01uF 1nF
ACX2 R4 AVDD
Resistor C4 C5 C6
1MOhm AIN3 AGND AIN3 AIN4 AIN4 AGND

Capacitor Capacitor Capacitor

ADP3303-5V VCC5 1nF 0.01uF 1nF
Ferrite Bead
AVDD OUT IN
Ferrite Bead OUT IN MIC4427_1
BMB2A1000LN2 Cap_S5_3
NR ERR AGND AGND
Capacitor NC NC
GND SD _ACX1_ Sense1 (-)
0.1uF INA OUTA
AGND AVDD
ADP3303 GND Vs
SDI_con R_SDI SDI SCK_con R_CLK SCK _ACX2_ Sense1 (+)
C DGND INB OUTB C
Resistor Resistor MIC442x
Cap_S5_4 Cap_S5_1 Cap_S5_2 100Ohm 100Ohm
Capacitor Capacitor Capacitor
4.7uF 10uF 0.1uF SDO_con R_SDO SDO SS_con R_CS SS MIC4427_2
AGND AGND
Resistor Resistor NC NC
ACX2 Sense2 (-)
DGND DGND DGND 100Ohm 100Ohm INA OUTA
AGND AVDD
GND Vs
ACX1 Sense2 (+)
INB OUTB
MIC442x
DVDD ADP3303-3.3V VCC5
OUT IN
OUT IN Title:
Cap_S3_3
Capacitor NR
GND
ERR
SD
WTB_Ampv1.0
0.1uF
ADP3303
DGND
Size: Number: Designed by:

D Cap_S3_4 Cap_S3_1 Cap_S3_2 A4 2 Manuel Aguiar Ferreira D

Capacitor Capacitor Capacitor
4.7uF 10uF 0.1uF Date: Sheet: 1 of 1
April 2015
DGND DGND DGND
File: Revision: *
1 2 3 4
 AD7105 Registers

/∗AD7195 R e g i s t e r s



























































∗/
#d e f i n e REG_COMM 0 /∗ C o m m u n i c a t i o n s R e g i s t e r (WO, 8
b i t ) ∗/
#d e f i n e REG_STAT 0 /∗ S t a t u s R e g i s t e r (RO, 8
b i t ) ∗/
#d e f i n e REG_MODE 1 /∗ Mode R e g i s t e r (RW, 24
b i t ∗/
#d e f i n e REG_CONF 2 /∗ C o n f i g u r a t i o n R e g i s t e r (RW, 24
b i t ) ∗/
#d e f i n e REG_DATA 3 /∗ Data R e g i s t e r (RO, 24/32
b i t ) ∗/
#d e f i n e REG_ID 4 /∗ ID R e g i s t e r (RO, 8
b i t ) ∗/
#d e f i n e REG_GPOCON 5 /∗ GPOCON R e g i s t e r (RO, 8
b i t ) ∗/
#d e f i n e REG_OFFSET 6 /∗ O f f s e t R e g i s t e r (RW, 16
b i t ∗/
#d e f i n e REG_FULLSALE 7 /∗ F u l l
S c a l e R e g i s t e r ∗/

/∗ C o m m u n i c a t i o n s R e g i s t e r B i t D e s i g n a t i o n s (AD7195_REG_COMM) ∗/
#d e f i n e COMM_WEN( x ) ( ( x ) << 7 ) /∗ W r i t e E n a b l e ∗/
#d e f i n e COMM_WRITE ( 0 << 6 ) /∗ W r i t e O p e r a t i o n ∗/
#d e f i n e COMM_READ ( 1 << 6 ) /∗ Read O p e r a t i o n ∗/
#d e f i n e COMM_ADDR( x ) ( ( ( x ) & 0 x7 ) << 3 ) /∗ R e g i s t e r A d d r e s s ∗/
#d e f i n e COMM_CREAD( x ) ( ( x ) << 2 ) /∗ C o n t i n u o u s Read o f Data R e g i s t e r ∗/

/∗ S t a t u s R e g i s t e r B i t D e s i g n a t i o n s (AD7195_REG_STAT) ∗/
#d e f i n e STAT_RDY ( 1 << 7 ) /∗ Ready ∗/
#d e f i n e STAT_ERR ( 1 << 6 ) /∗ E r r o r ( O v e r r a n g e , U n d e r r a n g e ) ∗/
#d e f i n e STAT_NOREF ( 1 << 5 ) /∗ E r r o r no e x t e r n a l r e f e r e n c e ∗/
#d e f i n e STAT_PARITY ( 1 << 4 ) /∗ P a r i t y ∗/
#d e f i n e STAT_CHD3 ( 1 << 2 ) /∗ C h a n n e l 3 ∗/
#d e f i n e STAT_CHD2 ( 1 << 1 ) /∗ C h a n n e l 2 ∗/
#d e f i n e STAT_CHD1 ( 1 << 0 ) /∗ C h a n n e l 1 ∗/

/∗ Mode R e g i s t e r B i t D e s i g n a t i o n s (AD7195_REG_MODE) ∗/
#d e f i n e MODE_SEL( x ) ( ( ( x ) & 0 x7 ) << 5 ) /∗ O p e r a t i o n Mode S e l e c t ∗/
#d e f i n e MODE_SEL_MASK ( 0 x7 << 5 ) /∗ O p e r a t i o n Mode S e l e c t Mask ∗/
#d e f i n e MODE_DAT_STA( x ) ( ( x ) << 4 ) /∗ S t a t u s R e g i s t e r t r a n s m i s s i o n ∗/
#d e f i n e MODE_CLKSRC( x ) ( ( ( x ) & 0 x3 ) << 2 ) /∗ C l o c k S o u r c e S e l e c t ∗/
#d e f i n e MODE_SINC3( x ) ( ( x ) << 1 5 ) /∗ SINC3 F i l t e r S e l e c t ∗/
#d e f i n e MODE_ENPAR( x ) ( ( x ) << 1 3 ) /∗ P a r i t y E n a b l e ∗/
#d e f i n e MODE_SCYCLE( x ) ( ( x ) << 1 1 ) /∗ S i n g l e c y c l e c o n v e r s i o n ∗/
#d e f i n e MODE_REJ60( x ) ( ( x ) << 1 0 ) /∗ 50/60 Hz n o t c h f i l t e r ∗/
#d e f i n e MODE_RATE( x ) ( ( x ) & 0 x3FF ) /∗ F i l t e r Update Rate S e l e c t ∗/

/∗ Mode R e g i s t e r : AD7192_MODE_SEL o p t i o n s ∗/
#d e f i n e MODE_CONT 0 /∗ C o n t i n u o u s C o n v e r s i o n Mode ∗/
#d e f i n e MODE_SINGLE 1 /∗ S i n g l e C o n v e r s i o n Mode ∗/
#d e f i n e MODE_IDLE 2 /∗ I d l e Mode ∗/
#d e f i n e MODE_PWRDN 3 /∗ Power
Down Mode ∗/
#d e f i n e MODE_CAL_INT_ZERO 4 /∗ I n t e r n a l Zero
S c a l e C a l i b r a t i o n ∗/
#d e f i n e MODE_CAL_INT_FULL 5 /∗ I n t e r n a l F u l l
S c a l e C a l i b r a t i o n ∗/
#d e f i n e MODE_CAL_SYS_ZERO 6 /∗ System Zero
S c a l e C a l i b r a t i o n ∗/
#d e f i n e MODE_CAL_SYS_FULL 7 /∗ System F u l l
S c a l e C a l i b r a t i o n ∗/

/∗ Mode R e g i s t e r : AD7192_MODE_CLKSRC o p t i o n s ∗/
#d e f i n e CLK_EXT_MCLK1_2 0 /∗ E x t e r n a l 4 . 9 2 MHz C l o c k c o n n e c t e d
∗ from MCLK1 t o MCLK2 ∗/

93
#d e f i n e CLK_EXT_MCLK2 1 /∗ E x t e r n a l C l o c k a p p l i e d t o MCLK2 ∗/
#d e f i n e CLK_INT 2 /∗ I n t e r n a l 4 . 9 2 MHz C l o c k n o t
∗ a v a i l a b l e a t t h e MCLK2 p i n ∗/
#d e f i n e CLK_INT_CO 3 /∗ I n t e r n a l 4 . 9 2 MHz C l o c k a v a i l a b l e
∗ a t t h e MCLK2 p i n ∗/
/∗ C o n f i g u r a t i o n R e g i s t e r B i t D e s i g n a t i o n s (AD7192_REG_CONF) ∗/
#d e f i n e CONF_CHOP( x ) ( ( x ) << 7 ) /∗ CHOP e n a b l e ∗/
#d e f i n e CONF_ACX( x ) ( ( x ) << 6 ) /∗ AC e x c i t a t i o n e n a b l e ( AD7195 o n l y ) ∗ /
#d e f i n e CONF_CHAN( x ) ( ( ( 1 << ( x ) ) & 0 xFF ) << 8 ) /∗ C h a n n e l s e l e c t ∗/
#d e f i n e CONF_CHAN_MASK ( 0 xFF << 8 ) /∗ C h a n n e l s e l e c t mask ∗/
#d e f i n e CONF_BURN( x ) ( ( x ) << 7 ) /∗ Bu rnou t c u r r e n t e n a b l e ∗/
#d e f i n e CONF_REFDET( x ) ( ( x ) << 6 ) /∗ R e f e r e n c e d e t e c t e n a b l e ∗/
#d e f i n e CONF_BUF( x ) ( ( x ) << 4 ) /∗ B u f f e r e d Mode E n a b l e ∗/
#d e f i n e CONF_UNIPOLAR( x ) ( ( x ) << 3 ) /∗ U n i p o l a r / B i p o l a r E n a b l e ∗/
#d e f i n e CONF_GAIN( x ) ( ( x ) & 0 x7 ) /∗ G a i n S e l e c t ∗/

#d e f i n e CH_AIN1P_AIN2M 0 /∗ AIN1 (+)
AIN2 (
) ∗/

#d e f i n e CH_AIN3P_AIN4M 1 /∗ AIN3 (+)
AIN4 (
) ∗/
#d e f i n e CH_TEMP 2 /∗ Temp S e n s o r ∗/
#d e f i n e CH_AIN2P_AIN2M 3 /∗ AIN2 (+)
AIN2 (
) ∗/
#d e f i n e AD7192_CH_AIN1 4 /∗ AIN1
AINCOM ∗/
#d e f i n e CH_AIN2 5 /∗ AIN2
AINCOM ∗/
#d e f i n e CH_AIN3 6 /∗ AIN3
AINCOM ∗/
#d e f i n e CH_AIN4 7 /∗ AIN4
AINCOM ∗/

#d e f i n e GAIN1 0
#d e f i n e GAIN8 3
#d e f i n e GAIN16 4
#d e f i n e GAIN32 5
#d e f i n e GAIN64 6
#d e f i n e GAIN128 7

/∗ GPOCON R e g i s t e r B i t D e s i g n a t i o n s (AD7192_REG_GPOCON) ∗/
#d e f i n e GPOCON_BPDSW ( 1 << 6 ) /∗ B r i d g e power
down s w i t c h e n a b l e ∗/
#d e f i n e GPOCON_GP32EN ( 1 << 5 ) /∗ D i g i t a l Output P3 and P2 e n a b l e ∗/
#d e f i n e GPOCON_GP10EN ( 1 << 4 ) /∗ D i g i t a l Output P1 and P0 e n a b l e ∗/
#d e f i n e GPOCON_P3DAT ( 1 << 3 ) /∗ P3 s t a t e ∗/
#d e f i n e GPOCON_P2DAT ( 1 << 2 ) /∗ P2 s t a t e ∗/
#d e f i n e GPOCON_P1DAT ( 1 << 1 ) /∗ P1 s t a t e ∗/
#d e f i n e GPOCON_P0DAT ( 1 << 0 ) /∗ P0 s t a t e ∗/

#d e f i n e INT_FREQ_MHz 4915200
/∗









































































∗/

94
 User Interface

Figure C.4: Wind tunnel testing GUI

Figure C.5: Wind tunnel static calibration GUI

Bonding of the Strain Gages

The quality of the installation greatly influences the accuracy of the measurement therefore, special
care has been taken in this step. Hoffmann (n.d.b) provides a series of practical hints to help in the installation
based on proven application means and methods which are described in detail. The same techniques have
been used in this project even though some of the equipment was not available.
To ease the alignment of the strain gages with the axial and circumferential directions, reference lines

95
were drawn on the surface of the elastic member by means of a scriber, as illustrated in figure C.6a. Figure
C.6b shows a close-up to an already bonded SG where it is possible to check the alignment.

(a) Drawing orthogonal lines to (b) Close-up to SG already bon-

use as a reference by means of a ded. It can be seen the reference
scriber lines.

(c) Protective rubber coating (d) All steps into manufacturing

applied on all sets of SG. a force transducer, from the left
to the rigth.

Figure C.6: Procedure to glue the SG

Figure C.7: Force tranducers quality control test setup

96
Appendix D

Calibration

Calibration Weights Uncertainty

Table D.1: Weight of the calibration weights
and their supports

Number Weights a [N ] Supports [N ]

1 0.9872 1.000
2 0.9809 0.999
3 0.9818 1.005
4 0.9874
5 0.9862
6 0.9859
7 0.9870
8 0.9808
9 0.9758
10 0.9863
11 0.9795
12 0.9800
13 0.9858
14 0.9844
15 0.9919
16 0.9852
17 0.9781
Upper bound a 0.9919 1.005
Lower bound a
0.9758 0.9990
a Relative to the standard gravity constant, g 
{
9.80665m s2

It was used a symmetric uniform distribution to describe the uncertainty associated with the weights
employed in the calibration. This method leads to the most conservative estimate since it gives the larger
standard deviation. Its calculation is based on the assumption that the end-points of the distribution are know
and that all values between a
and a are equally probable. The expected value of Xi is the midpoint of the
interval, xi  a
2 a , with associated variance

u2 pxi q 
pa
a q2 (D.1)
12

97
Figure D.1 shows the graphical representation of evaluating the standard uncertainty of an input quantity from
an a priori distribution (JCGM, 2008).

Figure D.1: Uniform uncertainty distribution

For the case of the weights, a
 0.9758 and a  0.9919 then, from equation D.5, uW  0.0046N .
Similarly for the supports a
 0.9990, a  0.1005, and uS  0.002N .

Calibration Data
Balance force transformation matrix, Ĉ
 
0.6653 0.4447 0.9799
0.2521 0.1237 0.1188

0.6376
0.3656 0.8856
0.0234
0.2726
0.1286



0.0185
0.7517 0.9869 0.1201 
Ĉ   
0.2210 0.1466
 0.0363
0.2287
0.1207 (D.2)
 0.7933 0.9657 0.1497
 

0.6740 0.3730 0.9155 0.0226
0.2678 0.1191 
0.7062
0.4496 1.0947 0.2660 0.1279
0.1137
Matrix of the uncertainty coefficients, Ĉu
 
14.75 4.47 8.66 1.01 1.07 0.87
14.70 
 4.44 7.48 1.30 1.29 1.08
 
14.81 1.26
Ĉu  
5.12 9.70 1.52 1.84   10
3
13.78 0.94
(D.3)
 2.99 6.60 1.01 0.85 
 
14.45 4.34 7.38 1.03 0.69 0.87
15.26 6.68 12.43 1.60 1.73 1.07

98
 Table D.2: Calibration loading plan

p FX r N s FY r N s FZ rN s MX [Link] MY [Link] MZ [Link]

1 0 0 0 0 0 0
2 1.9851 0 0 0 0 0
3 3.9527 0 0 0 0 0
4 5.9202 0 0 0 0 0
5 7.8878 0 0 0 0 0
6 9.8553 0 0 0 0 0
7 0 0 0 0 0 0
8 0 1.9851 0 0 0 0
9 0 3.9527 0 0 0 0 Table D.3: Uncertainty of the
10 0 5.9202 0 0 0 0 calibration loads
11 0 7.8878 0 0 0 0 Force uF Moment uM
12 0 9.8553 0 0 0 0 rN s [Link]
13 0 0 0 0 0 0 0 0 0 0
14 0 0 1.9851 0 0 0 1.9851 0.0066 0.0993 0.0020
15 0 0 3.9527 0 0 0 3.9527 0.0158 0.1976 0.0040
16 0 0 5.9202 0 0 0 5.9202 0.025 0.296 0.0061
17 0 0 7.8878 0 0 0 7.8878 0.0342 0.3944 0.0081
18 0 0 9.8553 0 0 0 9.8553 0.0434 0.4928 0.0101
19 0 0 0 0 0 0
20 0 0 1.9851 -0.0993 0 0
21 0 0 3.9527 -0.1976 0 0
22 0 0 5.9202 -0.2960 0 0
23 0 0 7.8878 -0.3944 0 0
24 0 0 9.8553 -0.4928 0 0
25 0 0 0 0 0 0
26 1.9851 0 0 0 0 -0.0993
27 3.9527 0 0 0 0 -0.1976
28 5.9202 0 0 0 0 -0.2960
29 7.8878 0 0 0 0 -0.3944
30 9.8553 0 0 0 0 -0.4928
31 0 0 0 0 0 0
32 0 0 1.9851 0 -0.0993 0
33 0 0 3.9527 0 -0.1976 0
34 0 0 5.9202 0 -0.2960 0
35 0 0 7.8878 0 -0.3944 0
36 0 0 9.8553 0 -0.4928 0

The uncertainty associated with the applied calibration force and moments are summarized in table
D.3. The latter also includes the error in the length of the lever arm ∆l, in addition to that of the applied
weights. The combined moment uncertainty is then given by

b
uM  ∆l2 u2F F 2 u2∆l (D.4)

where, ∆l  50mm and u∆l  1mm.

99
 Table D.4: R matrix - average of all calibration runs

p R1 rmV s R2 rmV s R3 rmV s R4 rmV s R5 rmV s R6 rmV s

1 -0.0012 0.0029 0.0077 0.0022 0.0006 -0.0007
2 1.3759 -0.0877 -1.3859 -1.3148 -0.1219 1.3419
3 2.7564 -0.1734 -2.7646 -2.6182 -0.2444 2.6764
4 4.1610 -0.2639 -4.1650 -3.9395 -0.3669 4.0331
5 5.5530 -0.3502 -5.5649 -5.2568 -0.4898 5.3815
6 6.9798 -0.4459 -6.9977 -6.5987 -0.6184 6.7576
7 0.0034 0.0056 0.0060 0.0010 -0.0077 -0.0033
8 -0.6855 -1.5336 -0.9672 0.9003 1.5336 0.6603
9 -1.3778 -3.0721 -1.9305 1.8019 3.0662 1.3154
10 -2.0607 -4.5942 -2.8869 2.6959 4.5919 1.9634
11 -2.7622 -6.1605 -3.8701 3.6024 6.1423 2.6274
12 -3.4493 -7.7121 -4.8375 4.5022 7.6788 3.2851
13 0.0015 0.0034 0.0010 0.0000 -0.0037 0.0005
14 0.3310 0.3547 0.3656 0.3422 0.3232 0.3090
15 0.6623 0.7058 0.7259 0.6835 0.6570 0.6238
16 0.9987 1.0635 1.0873 1.0227 0.9927 0.9371
17 1.3317 1.4179 1.4421 1.3618 1.3260 1.2522
18 1.6694 1.7746 1.8041 1.7023 1.6622 1.5718
19 -0.0009 0.0050 0.0019 -0.0042 -0.0011 -0.0010
20 0.5199 0.5537 0.3818 0.3296 0.1419 0.1241
21 1.0447 1.0973 0.7748 0.6608 0.2792 0.2530
22 1.5705 1.6428 1.1561 0.9933 0.4181 0.3834
23 2.0956 2.1902 1.5407 1.3263 0.5576 0.5108
24 2.6200 2.7444 1.9222 1.6612 0.6966 0.6405
25 -0.0064 -0.0010 0.0032 0.0007 -0.0006 -0.0027
26 1.2293 0.0460 -1.5164 -1.1623 -0.2561 1.4653
27 2.4773 0.0978 -3.0386 -2.3268 -0.5173 2.9405
28 3.7287 0.1406 -4.5759 -3.4936 -0.7783 4.4165
29 4.9659 0.1870 -6.1105 -4.6562 -1.0373 5.8927
30 6.2437 0.2370 -7.6725 -5.8411 -1.2995 7.3916
31 -0.0023 0.0005 0.0011 -0.0022 0.0000 -0.0029
32 0.4468 0.4653 0.1386 0.1210 0.4475 0.4207
33 0.8952 0.9264 0.2671 0.2420 0.8878 0.8402
34 1.3503 1.3973 0.4042 0.3616 1.3297 1.2646
35 1.7944 1.8614 0.5273 0.4804 1.7720 1.6896
36 2.2443 2.3237 0.6615 0.6016 2.2132 2.1100

100
Table D.5: Relative error of the estimated calibration loading Ŷ [%]

p FX FY FZ MX MY MZ
1
2 0.31
3 0.06
4 0.77
5 0.92
6 1.53
7
8 0.64
9 0.23
10 0.31
11 0.16
12 0.26
13
14 1.07
15 0.45
16 0.07
17 0.05
18 0.19
19
20 1.23 0.54
21 0.55 1.18
22 0.40 0.51
23 0.30 0.17
24 0.18 0.14
25
26 0.17 0.43
27 0.76 0.43
28 1.22 0.69
29 1.28 1.37
30 1.75 1.61
31
32 0.58 0.14
33 0.64 0.45
34 0.16 0.40
35 0.32 0.07
36 0.29 0.25

pŶp,i
Yp,i q  100

 absabs
ep,i
pY q p,i
(D.5)

101
Appendix E

Experimental Methods

Rectangular Flat Plate Test Procedure

Figure E.1: Rectangular plate wind tunnel test setup

1. The test model is attached to the balance by means of a squared tubular beam which places the model
normal to the oncoming flow, 220mm upstream and
50mm vertically from the resolving center of the
balance, as illustrated in figure E.1. The rectangular plate is initially positioned relatively to the test
section by visual inspection and subsequently checked using a ruler and a set-squared.

2. The data acquisition systems of the balance and pressure transducer are connected to a computer and
then switched on. Both rely on stain gage technology so a period of 10 to 15min is necessary in order to
ensure all bridges have reached a stable working temperature, otherwise it might be induced a significant
drift in the measurements.
In the mean term the DRS is started and the test data introduced. In the example of figure E.2 it has
been provided the area of the test article, the room temperature that has been taken using a sensor

102
 external to the balance DAQ, the directory to which the output files should be written, the test ID and
acquisition period. Not as important to the experiment is the plot window and update frequency which
can also be modified.

Figure E.2: DRS test setup

3. 2 different approaches have been used to test this article, both hold data originated from 7 runs. On the
first approach the airflow velocity is gradually increased, 2m{s at a time, starting from 4m{s up to 20.
The second one is similar but the velocity decreases in turn. Each run includes 9 data points that are
obtained averaging the measured loads over an acquisition period of 10s, during which 1500 samples
are retrieved. The acquisition cycle of one run is described below:

(a) Tare the balance;

(b) Set the wind tunnel velocity at Uinitial  4 or 20m{s using the trimmer potentiometer (to do so
it is necessary to know à priori the correct position);
(c) Turn on the wind tunnel and let the flow stabilize. It takes just a few seconds.
(d) Start acquiring. After the acquisition period, the output files are automatically saved as .txt files
in the designated directory - see table E.1.
(e) Increment or decrement 2m{s to the airflow velocity, depending on the approach being carried out,
by adjusting the trimmer potentiometer. Wait a few seconds to let the flow stabilize and go back
to step (d) until the last run.
(f) Turn off the wind tunnel.
(g) As soon as the airflow stops, the load offsets at the end of all runs is acquired.

Table E.1: Output files

Document Description
[Link] Reports the average value of the aerodynamic load components and coefficients.
AeroLoad_vector.txt 6  N matrix of all samples of the aerodynamic load components.
[Link] Reports on the FIR filter setup employed.
Reports the conditions under which the run was carried out;
[Link]
attitude of the model, area of reference, airflow velocity and air density.
Time_xlabel.txt 1  N time vector.
U_vector.txt 1  N velocity vector.

103
Results
Table E.2: Rectangular plate - CD data points

Re  10
5 Run 1 Run 2 Run 3 Run 4 Run 5 Run 6 Run 7

Up approach
0.388 1.079 1.074 1.043 1.036 1.085 1.075 1.064
0.498 1.107 1.126 1.109 1.107 1.151 1.118 1.124
0.669 1.121 1.138 1.129 1.127 1.150 1.153 1.139
0.862 1.143 1.139 1.144 1.134 1.146 1.146 1.140
1.019 1.151 1.144 1.157 1.154 1.152 1.156 1.151
1.196 1.152 1.155 1.161 1.147 1.153 1.156 1.154
1.350 1.149 1.150 1.148 1.151 1.142 1.148 1.146
1.522 1.140 1.145 1.145 1.141 1.140 1.143 1.147
1.704 1.143 1.140 1.143 1.146 1.135 1.135 1.139
Down approach
0.3882 0.999 0.982 0.910 1.194 0.967 1.206 0.965
0.4983 1.088 1.059 1.020 1.150 1.059 1.170 1.078
0.669 1.111 1.085 1.081 1.130 1.094 1.165 1.091
0.862 1.125 1.101 1.099 1.133 1.097 1.156 1.107
1.019 1.137 1.122 1.124 1.137 1.124 1.164 1.132
1.196 1.144 1.139 1.127 1.145 1.136 1.166 1.136
1.350 1.143 1.135 1.134 1.141 1.138 1.157 1.140
1.522 1.144 1.136 1.136 1.133 1.139 1.151 1.138
1.704 1.134 1.129 1.132 1.137 1.131 1.136 1.130

Table E.3: Rectangular plate - drag coefficient

Re  10
5 CDup uCDup CDdn uCDdn

0.388 1.081 0.0176 1.048 0.1111
0.498 1.115 0.0144 1.084 0.0487
0.669 1.132 0.0108 1.103 0.0276
0.862 1.146 0.0041 1.121 0.0202
1.019 1.144 0.0040 1.127 0.0132
1.196 1.141 0.0041 1.130 0.0110
1.350 1.140 0.0027 1.134 0.0070
1.522 1.137 0.0023 1.134 0.0054
1.704 1.137 0.0038 1.130 0.0028

104
 (a) Detail (b) Mounting setup

Figure E.3: 80I transverse cylinder

(a) Front (b) Back

Figure E.4: 120I disk model

(a) Plain cylinder (b) Cylinder + Cone

Figure E.5: 120I cylinder, axial to the oncomign flow

105
 Run 1
0.76
Run 2

Drag cofficient, CD
Run 3
0.74 Run 4
Run 5
Run 6
0.72
Run 7

0.7

0.68
0.4 0.6 0.8 1
Reynolds number, Re  10
5

Figure E.6: 80I transverse cylinder - Up approach scatter plot

Run 1
1.15 Run 2
Drag cofficient, CD

Run 3
Run 4
1.1 Run 5
Run 6
Run 7

1.05

0.4 0.6 0.8 1 1.2 1.4 1.6

Reynolds number, Re  10
5

Figure E.7: 120I disk - Up approach scatter plot

CCDDCyl
Cyl
0.76 1.14 CDDisk
Cylinder drag coefficient, CDCyl

Disk drag coefficient, CDDisk

1.12
0.74
1.1

1.08
0.72
1.06

0.7 1.04

1.02
0.68
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Reynolds number, Re  10
5

Figure E.8: Transverse cylinder and disk results - 95% confidence interval

106
Uncertainty Analysis

Table E.4: Rectangular plate - Up approach,

1st run values

U [m{s] A [m2 ] ρ [kg {m3 ] FX [N ]

4.8 0.015 1.204 0.224
6.1 0.015 1.204 0.380
8.2 0.015 1.204 0.695
10.6 0.015 1.204 1.168
12.6 0.015 1.204 1.629
14.7 0.015 1.204 2.241
16.6 0.015 1.204 2.853
18.8 0.015 1.204 3.613
21.0 0.015 1.204 4.530

Table E.5: Rectangular plate - CD partial derivatives and uncertainties

associated with its variables
BU B
CD
uA  104 BCD BCD
Bρ uρ partialA BFX uFX BU uU
0.898 0.0602 72.067 1.80 4.834 0.0073 0.452 0.211
0.926 0.0602 74.360 1.80 2.934 0.0172 0.363 0.166
0.940 0.0602 75.433 1.80 1.628 0.0308 0.274 0.124
0.951 0.0602 76.367 1.80 0.981 0.0483 0.216 0.097
0.950 0.0602 76.287 1.80 0.702 0.0696 0.182 0.082
0.948 0.0602 76.087 1.80 0.509 0.0944 0.155 0.070
0.947 0.0602 76.020 1.80 0.400 0.1215 0.137 0.062
0.944 0.0602 75.793 1.80 0.315 0.1538 0.121 0.055
0.944 0.0602 75.793 1.80 0.251 0.1898 0.108 0.049

Table E.6: Rectangular plate - variance of the CD variables

BCD 2
u2ρ BCD 2
u2A  104 BCD 2
u2FX B CD 2
u2U
Bρ BA BFX BU uCD [Drag counts]
0.00292 1.68 0.00125 0.01920 153.4
0.00311 1.79 0.00255 0.00753 115.6
0.00320 1.84 0.00252 0.00239 91.0
0.00328 1.89 0.00224 0.00089 81.2
0.00327 1.89 0.00239 0.00045 79.4
0.00326 1.88 0.00231 0.00024 77.4
0.00325 1.87 0.00236 0.00015 77.1
0.00323 1.86 0.00234 0.00009 76.5
0.00323 1.86 0.00227 0.00006 75.8

107
 Pressure Sensor Calibration

The dynamic pressure sensor was calibrated using a water column barometer, tilted at a 30 angle with
the horizontal to increase sensitivity.

π
Pdyn  ρw g∆Hsin 6
(E.1)

where the water density ρw is equal to 1000kg {m3 , and ∆H is the variation of the water column height.

Table E.7: Pressure sensor calibration

Sensor output [mV ] mmH20 [mm] Pdyn [P a]

0 1 0
38.7 4 14.7
87.0 6 24.5
154.7 10 44.1
241.7 16 73.5
348.0 22 102.9
473.7 30 142.1
618.7 38 181.3
783.0 48 230.3
966.7 60 289.1

300
Calibration data points
 0.2963x 0.2661, R2  0.9996
Analog dynamic pressure [P a]

200

100

0
0 200 400 600 800 1,000
Sensor output [mV ]

Figure E.9: Pressure sensor calibration

108
Bibliography

2000 (December). IEEE Standard for Terminology and Test Methods for Analog-to-Digital Converters. IEEE-
SA Standards Board IEEE Std 1241-2000. IEEE Instrumentation and Measurement Society.

Aerolab. May 2015. Sting Balance. Website.

URL: [Link]

AGARD. 1996 (October). Rotary-Balance Testing for Aircraft Dynamics. Advisory Report AR-265. Advisory
Group for Aerospace Research & Development (AGARD).

Analog Devices. May 2015. Analog Devices, AD7195 datasheet. Website.

URL: [Link]

Barlow, Jewel B, Rae, William H, & Pope, Alen. 1999. Low-Speed Wind Tunnel Testing. 3th edn. Wiley-
Interscience.

Baturin, V V. 1972. Fundamentals of industrial ventilation. 3rd edn. Pergamon Press.

Bergmann, Andreas. 2007. Hermann Schlichting - 100 Years - Scientific Colloquium Celebrating the Anniversary
of His Birthday. In: Radespiel, Rolf, Rossow, Cord-Christian, & Brinkmann, Benjamin Winfried (eds),
"Modern Wind Tunnel Techniques for Unsteady Testing - Development of Dynamic Test Rigs. Springer.
Braunschweig, Germany.

Bergmann, Andreas, Huebner, Andreas, & Loeser, Thomas. 2008. Experimental and numerical research on
the aerodynamics of unsteady moving aircraft. Progress in Aerospace Sciences, 44(2), 121–137.

Bevington, Philip R, & Robinson, D Keith. 2003. Data Reduction and Error Analysis for the Physical Sciences.
3rd edn. McGraw-Hill Higher Education.

Chipkin. May 2015. Types of Strain Gauges. Website.

URL: [Link]

Dai, JS, Sodhi, C, & Kerr, DR. 1994. Design and analysis of a new six-component force transducer for robotic
grasping. Pages 809–17 of: Proceedings of the Second Biennial European Joint Conference on Engineering
Systems Design and Analysis, ASME PD, vol. 64.

Dasgupta, B, Reddy, S, & Mruthyunjaya, TS. 1994. Synthesis of a force-torque sensor based on the Stewart
platform mechanism. Pages 14–23 of: Proceedings of the National Convention of Industrial Problems in
Machines and Mechanisms, IPROMM’94, vol. 64. Bangalore, India.

Dwarakanath, TA, Dasgupta, Bhaskar, & Mruthyunjaya, TS. 2001. Design and development of a Stewart
platform based force-torque sensor. Mechatronics, 11(7), 793–809.

Edwards, Alex T. 2000. Comparison of Strain Gage and Fiber Optic Sensors on a Sting Balance in a Supersonic
Wind Tunnel. Master Thesis, Virginia Polytechnic Institute and State University.

109
Erm, Lincoln P. 2006. Development and Use of a Dynamic-Testing Capability for the DSTO Water Tunnel.
Tech Report. Defence Science and Technology Organisation (DSTO).

Eyre, R C W, Fail, R, & Lawford, J A. 1959. Low-Speed Experiments on the Wake Characteristics of Flat
Plates normal to an Air Stream. Technical Report 19,958. Aeronautical Research Council, London.

Ferreira, Manuel. 2015. Calibration Procedure of a 6-6 In-parallel Wind Tunnel Balance. IST Report. IST - U
Lisbon.

Ferreira, Ricardo. 2013. Analysis of Different Types of Chassis for Formula Student. Master Thesis, IST - U
Lisbon.

FST. May 2015. Projecto FST Novabase. Website.

URL: [Link]

Gaillet, A, & Reboulet, C. 1983. An isostatic six-component force and torque sensor. Proceedings of the 13th
Int. Symposium on Industrial Robotics, 102–111.

Geyling, FT, & Forst, JJ. 1960. Semiconductor Strain Transducers. The bell system technical Journal, 39(3),
705–731.

Greenwell, DI. 1996 (December). Analysis of a six degree-of-freedom dynamic wind tunnel test rig mechanism
based on the Stewart platform. Tech Report AS/HWA/TR96102/1. Defence Evaluation Research Agency
(DERA), Farnborough.

Harris, Cyril M, & Piersol, Allan G (eds). 2002. Shock and Vibration Handbook. 5th edn. McGraw-Hill.

HBM. HBM Strain Gage Catalog - Foil strain gages Series Y.

HBM. May 2015a. Determining Strain Gauge Excitation Voltage. Website.

URL: [Link]
strain-gauge-excitation-voltage/.

HBM. May 2015b. Force Measurement Tips & Tricks. Website.

URL: [Link]
spoilt-for-choice-piezoelectric-or-strain-gauge-based-force-transducers/.

Hoe, Garrison, Owens, D Bruce, & Denham, Casey. Forced Oscillation Wind Tunnel Testing for FAZER Flight
Research Aircraft. Tech Report. AIAA.

Hoffmann, Karl. Applying the Wheatstone Bridge Circuit. HBM.

Hoffmann, Karl. Practical hints for the installation of strain gages. HBM.

Holub, Jan, & Vedral, Josef. 2004 (April 18-21). Measurements of Effective Resolution of ADC in Microcon-
vertor ADUC 834. In: 7th IEEE International Workshop on DDECS. Prage, Czech Republic.

Honeywell. May 2015. Ultra Low Pressure Silicon Sensors. Website.

URL: [Link]
pdf.

Hou, Yulei, Zeng, Daxing, Yao, Jiantao, Kang, Kaijia, Lu, Ling, & Zhao, Yongsheng. 2009. Optimal design
of a hyperstatic Stewart platform-based force/torque sensor with genetic algorithms. Mechatronics, 19(2),
199–204.

JCGM. 2008 (September). Evaluation of measurement data - Guide to the expression of uncertainty in
measurement. Tech. rept. Joint Committee for Guides in Metrology. First edition.

110
Kleckers, Thomas. 2009 (September 6-11). Fiber Bragg sensors compared with electrical strain gauges for
use in force measurement - Prospects and Potentials. In: XIX IMEKO World Congress Fundamental and
Applied Metrology. Lisbon, Portugal.

Kreuzer, Manfred. Strain Measurement with Fiber Bragg Grating Sensors. Tech Report. HBM.

L Hannah, Richard, & E Reed, Stuart. 1994. Strain Gage Users’ Handbook. London: Chapman and Hall.

Lee, Jaehoon. 2000. Investigation of Quality Indices of In-parallel Platform Manipulators and Development of
Web Based Analysis Tool. Ph.D. thesis, University of Florida.

Lerner, Jorge Colman, & Boldes, Ulfilas (eds). 2011. Wind Tunnels and Experimental Fluid Dynamics Research.
InTech. Chap. 7.

Leung, Sunny Yin Fat, & Link, Yoel Y. 1999 (August). Comparison and Analysis of Strain Gauge Balance
Calibration Matrix Mathematical Models.

Lienhard, John H. 1966. Synopsis of Lift, Drag, and Vortex Frequency Data for Rigid Circular Cylinders.
Bulletin 300. Washington State University, Washington.

Merlet, J P. 2006. Parallel Robots. 2nd edn. Springer. Chap. 6.

Micrel Semiconductors. May 2015. Micrel Semiconductors, MIC4427 datasheet. Website.

URL: [Link]

Microchip. May 2015. Microchip Technology Inc., dsPIC30F6012A datasheet. Website.

URL: [Link]

O’Grady, Albert. 2000. Transducer/Sensor Excitation and Measurement Techniques. Analog Dialogue, 34(5).

Owens, D Bruce, Brandon, Jay, Croom, Mark A, Fremaux, C Michael, Heim, Eugene H, & Vicroy, Dan D.
Overview of Dynamic Test Techniques for Flight Dynamics Research at NASA LaRC (Invited). Tech Report.
AIAA.

Ramaswamy, MA, Srinivas, T, & Holla, VS. 1987. A simple method for wind tunnel balance calibration
including non-linear interation terms. Pages 263–267 of: Proceedings of the ICIASF 1987 RECORD.

Ranganath, R, Nair, PS, Mruthyunjaya, TS, & Ghosal, A. 2003. Studies on a Stewart Platform based Force-
Torque Sensor in a Near-Singular Configuration. In: Huang, Tian (ed), Proceedings of the 11th World
Congress in Mechanism and Machine Science. China Machinery Press. Tianjin, China.

Reis, MLCC, Castro, RM, & Mello, OAF. 2013. Calibration uncertainty estimation of a strain-gage external
balance. Measurement, 46(1), 24–33.

Romiti, A, & Sorli, M. 1992. Force and moment measurement on a robotic assembly hand. Sensors and
Actuators A: Physical, 32(1-3), 531–8.

Sandu, Marin, Sandu, Adriana, & Ionascu, Georgeta. 2011. Design of a compact six-component force and
moment sensor for aerodynamic testing. Incas Bulletin, 3(1), 95–100.

Sarmento, António. 2011 (February). Escoamentos Exteriores em torno de Corpos Não-fuselados. Fluid
Mechanics. U Lisbon - IST.

Sauber. May 2015. Tech Bites: Short. Simple. Informative. Website.

URL: [Link]
PL75DTRycWkHZUjKPiP94k5terjLybfX0y.

111
Sheingol, Dan, & Kester, Walt. 2005. Data Conversion Handbook. Norwood,MA,USA: Analog Devices Inc.
Engineeri.

Sorli, M, & Pastorelli, S. 1995. Six-axis reticulated structure force/torque sensor with adaptable performances.
Mechatronics, 5(5), 585–601.

Sorli, M, & Zhmud, N. 1993. Investigation of force and moment measurement system for a rotating assembly
hand. Sensors and Actuators A: Physical, 37(38), 651–7.

Stewart, D. 1965. A Platform with Six Degrees of Freedom. Proceeding of the Institution of Mechanical
Engineers, 180(1), 371–386.

Stoughton, RS, & Arai, T. 1993. A modified Stewart platform manipulator with improved dexterity. Robotics
and Automation, IEEE Transactions, 9(2), 166–173.

Svinin, MM, & Uchiyama, M. 1995. Optimal geometric structure of force/torque sensors. Int J Robot Res,
14(6), 560–73.

Tropea, Cameron, Yarin, Alexander L, & Foss, John F (eds). 2007. Springer Handbook of Experimental Fluid
Mechanics. Springer.

Tsai, LW. 1999. Robot Analysis: The Mechanics of Serial and Parallel Manipulators. New York, USA: John
Wiley and Sons.

Yao, Jiantao, Hou, Yulei, Wang, Hang, Zhou, Tieling, & Zhao, Yongsheng. 2011. Spatially Isotropic Config-
uration of Stewart Platform-based Force Sensor. Mechanism and Machine Theory, 46(2), 142–55.

Zhang, Bo, & Crane, C. 2004. Special Singularity Analysis for a 6-DOF Parallel Platform. In: 10th Int.
Conference on Robotics and Remote Systems for Hazardous Evironments. Florida, USA.

112