See discussions, stats, and author profiles for this publication at: [Link]

net/publication/245389264

Model-in-the-Loop Testing

Article in Proceedings of the Institution of Mechanical Engineers Part I Journal of Systems and Control Engineering · May 2006
DOI: 10.1243/09596518JSCE207

CITATIONS READS

117 11,423

1 author:

Andrew R. Plummer
University of Bath
171 PUBLICATIONS 4,094 CITATIONS

SEE PROFILE

All content following this page was uploaded by Andrew R. Plummer on 16 September 2014.

The user has requested enhancement of the downloaded file.

 Proceedings of the Institution of Mechanical
Engineers, Part I: Journal of Systems and
Control Engineering
[Link]

Model-in-the-Loop Testing
A. R. Plummer
Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering 2006 220: 183
DOI: 10.1243/09596518JSCE207

The online version of this article can be found at:

[Link]

Published by:

[Link]

On behalf of:

Institution of Mechanical Engineers

Additional services and information for Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control
Engineering can be found at:

Email Alerts: [Link]

Subscriptions: [Link]

Reprints: [Link]

Permissions: [Link]

Citations: [Link]

>> Version of Record - May 1, 2006

What is This?

Downloaded from [Link] by guest on September 6, 2012

 183

Model-in-the-loop testing
A R Plummer
Instron Ltd, Coronation Road, High Wycombe, Buckinghamshire HP12 3SY, UK. email: andrew_plummer@[Link]

The manuscript was received on 7 October 2005 and was accepted after revision for publication on 20 December 2005.

DOI: 10.1243/09596518JSCE207

Abstract: Physical testing is used extensively to characterize mechanical systems. However,

in many cases, mathematical models are now available that adequately describe the behaviour
of part of the test specimen. Thus, test systems can be conceived which split the specimen
into a physical part, and a virtual part, i.e. a real-time computer simulation. This has the
potential to enhance convenience and reduce cost. The term ‘model-in-the-loop’ (MiL) has
been used in the automotive industry to describe this concept. In this paper, a general frame-
work for MiL system analysis is described. Two detailed examples are given. These concern
the integration of numerical aerodynamic and tyre models into car test rigs. In the first case,
the analytical results are validated by data from a real system. It is clear that particular demands
are placed on the actuation and sensing systems if the numerical and real parts of the specimen
are to interact correctly to give a realistic response for the complete system. Acceptable realism
is achieved in the case of the aerodynamic model example. However, the results are less
satisfactory in the tyre model example where both actuator response and sensor noise are
limiting factors.

Keywords: automotive testing, hardware-in-the-loop, electrohydraulic control, structural

testing, substructuring

1 INTRODUCTION mental conditions or aerodynamic forces, and so

the affected systems can be represented more
Mechanical systems are often tested in the laboratory accurately by a computer model.
to assess their dynamic characteristics. For example, 4. The characteristics of the simulated system can
new designs for road vehicles, aircraft, or building be varied to represent alternative configurations –
structures can be evaluated in this way. Such tests this is more convenient than changing physical
usually consist of exciting the mechanical system components.
using actuators to simulate its in-service loading
conditions, and measuring the system’s response in Hardware-in-the-loop (HiL) simulations have been
terms of motion, force, strain, etc. used successfully for a number of years. These have
Model-in-the-loop (MiL) testing is a new paradigm evolved from pure computer simulations of a con-
in which the test specimen is part real and part troller and plant, such that (usually) the controller
virtual. The virtual part is implemented via a real- simulation is run in real-time and used to control
time computer simulation. This arrangement has a the real plant, making use of the correct actuators
number of potential advantages, depending on the and sensors. The model-in-the-loop concept differs
application. in that the real-time simulation is part of the
mechanical system, not of a controller. Thus, the
1. Only the key component with unknown dynamics real-time simulation in a MiL system interfaces with
need be physically tested, reducing the cost and the rest of the system using actuators and sensors
complexity of the physical test apparatus. that would not otherwise be present; these actuators
2. A newly designed system can be tested even if and sensors have their own characteristics which
some parts have yet to be physically realized – detract from the realism of the complete system
these parts are simulated instead. response.
3. Sometimes it is difficult physically to replicate Figure 1 shows the principle of MiL testing: the
in-service conditions in the laboratory, e.g. environ- complete test specimen is emulated by a numerical

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

184 A R Plummer

Fig. 1 Model-in-the-loop concept

model interacting with a physical system. External The concept of testing a specimen which is part
excitation of the specimen may originate from either numerical model and part physical structure has been
the numerical or physical parts. Figure 2 illustrates a studied for a number of years in the civil engineer-
more realistic situation: sensors and actuators have ing field. The terms ‘substructuring’ or ‘real-time
their own dynamic characteristics, they are affected hybrid testing’ have been used in this context. The
by noise and disturbances, and the actuator response difficulties associated with limited actuator response
is dependent on the behaviour of the attached have been a major concern. Horiuchi et al. [1] were
physical system (shown by the dashed line). These the first to propose a method of compensating for
practical realities must be taken into account in any the actuator response. A time delay was used as a
MiL design. simple actuator response model, which was com-
In this paper, a general modelling framework is pensated by a moving average filter, effectively a
described for MiL systems. This is then used to discretized lead term. This design was justified on the
present two examples of MiL testing from the motor basis of extrapolating a polynomial fit to the current
sports industry. In the first example, a numerical and previous actuator command samples. In [2],
aerodynamic model is used to add aerodynamic Darby et al. noted that the actuator delay varied
forces to a physical racing-car test rig. Practical with the stiffness of the physical structure, i.e. the
results are presented. Such test systems, often called actuator response is dependent on the dynamics of
seven-post test rigs, have been used for several years the physical system. The delay was estimated during
in Formula One car testing. An analysis of the realism a test, thus allowing adaptive delay compensation. An
of such systems is presented for the first time. The alternative approach is to force the actuator dynamics
second example is the inclusion of tyre models in a to follow a fixed reference model; then, cancelling
hub-coupled seven-post rig, i.e. where physical tyres the reference model dynamics is equivalent to delay
have been removed from the car. Based on measured compensation. To achieve this, Lim et al. [3] have
actuator response data and simulation results, the used a form of the adaptive minimal control
achievable realism accounting for actuator dynamics synthesis (MCS) algorithm for the actuator control
and measurement noise is presented. In both these loop, giving a consistent actuator response; low-
examples hydraulic actuation is used; this is the only frequency experimental results are presented. An
actuator technology suited to high-force dynamic alternative approach using MCS is to use the entire
testing. numerical model for the reference model [4].

Fig. 2 MiL with imperfect actuator and sensor behaviour

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 185

However, difficulties have been encountered when 2.2 Models for the partitioned system
the numerical model (differential equation) is of a
A general mechanical system with linear dynamics
higher order than one [5].
can be represented by a network of masses, springs,
The control of actuators so that they exhibit
and dampers. This system is to be partitioned into
a particular force/motion relationship has been
two parts: the numerical model and the physical
studied for some years for other applications; it is
system. The complete system is
often termed ‘impedance control’ [6]. Hydraulic
actuation systems have been used in the past to M 0
C +C −C
emulate a variety of mechanical systems. Feel-force
systems in flight simulators, which endow the pilot’s
AC 0 M
p
n n
−CT D C
np
nn
s2+
C +C
p
np
pp
s
D
controls with realistic force/motion characteristics, K +K −K x F
are traditionally hydraulic [7]. Hydraulic systems have
also been used to emulate the loads experienced by
+ n
C
−KT
np
nn
K +K
p
np
pp
x
p
DBC D C D
n = n
F
p
(1)

rotating machinery [8], and also the load experienced Here, M and M are the mass matrices (linear or
by a drive cylinder in an injection moulding n p
angular inertia) contained in the numerical and
machine [9]. A MiL approach has recently been used physical parts respectively. Similarly C and K are
in a suspension test rig by connecting a numerical n n
matrices of damping and spring constants respect-
mass-spring model to a physical magneto-rheological ively within the numerical model, and F is a vector
damper [10]. n
of external forces acting on the masses in the
numerical model. Likewise, C and K are matrices
p p
of damping and spring constants within the physical
2 A GENERAL MODELLING FRAMEWORK
system, and F is a vector of external forces acting
p
on the physical masses. Matrices C , C , and C ,
2.1 The purpose of linear modelling nn np pp
and K , K , and K , are damping and stiffness
nn np pp
In reality, the physical system will have uncertain matrices associated with the interface between the
dynamic characteristics, and exhibit non-linearities; numerical and physical systems; these are discussed
if it were easy to model, there would be no point in later.
testing it physically. Equation (1) can be partitioned into two parts, as
The numerical model may also be non-linear. How- represented in Fig. 3, giving the numerical model
ever, it is still valuable to represent both numerical
model and physical system by linear models for the A (s)x =F +(C s+K )x −(C s+K )x (2)
n n n np np p np nn n
purpose of analysis, with the intention that the main
dynamic characteristics can still be represented at where
least around one operating point. The thrust of the
A (s)=M s2+C s+K (3)
present study is to be able to compare the response n n n n
of the MiL system (which includes imperfect actuator and the physical system
and sensor behaviour) with the response of the
emulated system (the ideal response); finding the A (s)x =F +(CT s+KT )x −(C s+K )x (4)
p p p np np n pp pp p
difference between these two responses is more
important than precise modelling of the test where
specimen.
A (s)=M s2+C s+K (5)
For the modelling framework developed in the p p p p
following sections, it is assumed that the sensor In equations (2) and (4) all forces at the interface
dynamics can be neglected, as the sensor bandwidth between the numerical and physical systems are
is normally an order of magnitude higher than that included on the right-hand side. These interface
of the actuators. The actuator tracking and disturb- forces are also given by
ance rejection responses are included, as well as
measurement noise. The term ‘actuator’ is used to F =(C s+K )(x −x ) (6)
refer to the prime mover, including its individual i i i ni pi
closed-loop controller; this is often called the ‘transfer where x and x are the position variables adjacent
ni pi
system’ in the substructuring literature [1–5]. Also to the interface, and C and K are diagonal damping
i i
in this literature, the actuators are used as position and stiffness matrices for the interface. The elements
servosystems; this does not have to be the case – of these two matrices may originate from either the
force servosystems are equally valid. numerical or physical systems. Equation (6) can

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

186 A R Plummer

Fig. 3 Partitioning into a numerical model A (s) and a physical system A (s)
n p
alternatively be written as where
F =B (s)(Q x −Q x ) (7) D (s)=B (s)Q A (s)−1 (20)
i i n n p p n i n n
where and similarly substituting equation (17) into (7)
B (s)=C s+K (8) B (s)x =D (s)F +(I+D (s)QT )F (21)
i i i i ni p p p p i
and where Q and Q select the variables adjacent to where
n p
the interface from the full position vectors (they are
D (s)=B (s)Q A (s)−1 (22)
matrices with all zero elements apart from a single p i p p
element of unity in each row). In summary, the model in equation (1) can be
So, comparing equations (2), (4) and, (7) partitioned into equations (14) and (15), which can
QT B (s)Q =C s+K (9) be alternatively written as (19) and (21) respectively.
n i p np np In the latter pair the numerical model is related to
QT B (s)Q =C s+K (10) the physical interface positions, and similarly the
n i n nn nn
QT B (s)Q =C s+K (11) physical system is related to the numerical interface
p i p pp pp positions. The way these equations are used depends
Substituting equations (7) and (9)–(11) into equations on the configuration of the test, and is described in
(2) and (4) the next section.
A (s)x =F −QT F (12)
n n n n i 2.3 Alternative configurations for MiL systems
A (s)x =F +QT F (13)
p p p p i
The numerical model can be realized using either
Or, in terms of the interface displacements
position or force-controlled actuators, and B (s)
i
x =Q A (s)−1(F −QT F ) (14) (which will be called the interface impedance) can
ni n n n n i
x =Q A (s)−1(F +QT F ) (15) be in either the physical or numerical model. This
pi p p p p i gives four distinct cases:
These equations describe the relationship between
interface forces and displacements for both the (a) physical system contains interface impedance,
numerical model and the physical system. However, position-controlled actuators;
another pair of equations is also required relating x (b) physical system contains interface impedance,
ni force-controlled actuators;
to F via the physical system, and x to F via the
i pi i (c) numerical model contains interface impedance,
numerical model.
From equations (12) and (13) position-controlled actuators;
(d) numerical model contains interface impedance,
x =A (s)−1(F −QT F ) (16) force-controlled actuators.
n n n n i
x =A (s)−1(F +QT F ) (17) In Fig. 4 the masses and, hence, displacements
p p p p i
Substituting equation (16) into (7) directly at the interface are contained in the
numerical model, and, hence, the interface impedance
F =B (s)(Q A (s)−1(F −QT F )−Q x ) (18)
i i n n n n i p p B (s) is in the physical system. Equations (14) and
i
B (s)x =D (s)F −(I+D (s)QT )F (19) (21) can be used in this case. The complete system
i pi n n n n i

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 187

In Fig. 7 the masses and, hence, displacements

directly at the interface are contained in the physical
system, and, hence, the interface impedance B (s)
i
is in the numerical model. Equations (15) and (19)
can be used in this case. The complete system with
position-controlled actuators is shown in Fig. 8
(case 3). The complete system with force-controlled
Fig. 4 Interface impedance is part of the physical actuators is shown in Fig. 9 (case 4).
system
2.4 An illustration of two special cases
with position (displacement)-controlled actuators Figure 10 is an example of two special cases. First,
is shown in Fig. 5 (case 1). The complete system there are no mass elements in one of the models, in
with force-controlled actuators is shown in Fig. 6 this case the numerical model. Second, there is an
(case 2). Here P(s) is the actuator closed-loop transfer external force which originates from one part of the
function matrix, and R(s) is the actuator sensitivity
transfer function matrix (i.e. the effect of disturb-
ances on the actuator output). Subscripts d and f
refer to displacement or force-controlled actuators
respectively (thus R (s) is the sensitivity of the con-
d
trolled displacement to a force disturbance, and R (s)
f
is the sensitivity of the controlled force to a displace-
ment disturbance). The signals F̂ and x̂ are the
i ni
interface forces and displacements in the numerical
model, i.e. they are measured or commanded values, Fig. 7 Interface impedance is part of a numerical
model
rather than the real physical signals F and x .
i ni

Fig. 5 Case 1: interface impedance in the physical Fig. 8 Case 3: interface impedance in the numerical
system, with position-controlled actuators model, with position-controlled actuators

Fig. 6 Case 2: interface impedance in the physical Fig. 9 Case 4: interface impedance in the numerical
system, with force-controlled actuators model, with force-controlled actuators

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

188 A R Plummer

Fig. 10 Special case, with the numerical model con-

taining no mass elements

system (here the numerical model) but is applied to Fig. 11 Representation of Fig. 10, using force-controlled
a mass in the other part (the physical system). actuators
If force-controlled actuators are used, then this is
an example of case 4 above. As there are no position example, equations (14) and (15) become
variables in the numerical model, equation (19)
(m s2+c s+k )x =(c s+k )z −F (26)
simplifies to n n n ni n n n i
(m s2+c s+k )x =(c s+k )z +F (27)
F =−B (s)x (23) p p p pi p p p i
i i pi So
The external forces acting on the masses in the
physical system are A (s)=m s2+c s+k (28)
n n n n
F =F +QT F (24) F =(c s+k )z (29)
p pp p np n n n n
Q =1 (30)
That is, the complete vector of forces is the sum of n
external forces F originating in the physical system, and
pp
and the forces F which act on the masses at the A (s)=m s2+c s+k (31)
np p p p p
interface and originate from the numerical model.
Figure 11 shows the complete system model. In F =(c s+k )z (32)
p p p p
this diagram a new variable is introduced Q =1 (33)
p
F∞ =−B (s)x +F (25) Also
i i pi np
B (s)=c s+k (34)
2.5 Example i i i
and from equation (22)
Figure 12 is an example two-degree-of-freedom mass-
spring-damper system taken from Lim et al. [3]. With c s+k
D (s)= i i (35)
position-controlled actuators, this is case 1. For this p m s2+c s+k
p p p

Fig. 12 Two-degree-of-freedom example

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 189

Fig. 13 Aerodynamic model-in-the-loop

So, with a knowledge of the actuator dynamics

(P (s) and R (s)), by inspection of Fig. 5 it is clear
d d
that the behaviour of the MiL system could be
analysed. Comparing this behaviour with that of the
emulated system (effectively Fig. 5 with P (s)=1,
d
R (s)=0, and m =0) allows the accuracy of the MiL
d f
emulation to be quantified.

3 AUTOMOTIVE TESTING EXAMPLE 1: Fig. 14 Representation of the physical system

AN AERODYNAMIC MODEL-IN-THE-LOOP

3.1 Modelling
impedance is in the numerical model, this is a case 1
In this example, a numerical aerodynamic model is configuration. In fact it exhibits the special features
used to add a vertical force and pitch moment to the described in section 2.4.
body of a physical racing car in a track simulation The physical system is given by
test rig. The basic ‘four poster’ test rig is widely used,
0 m z
and consists of a hydraulic actuator under each
wheel replicating the road height variation. However, ACb
0 i
b
s2+Cs+K
D h
b
b
BC D
for a racing car, aerodynamic down force is very
z F
significant; it can be twice the vehicle weight in the
case of a Formula One car. The down force depends
=(Cs+K ) r +
h
r
F
z
h
CD C D (36)

on vehicle forward speed, and body pitch and height,

where
and the extent to which it affects the front or rear of
0 c
C D
the car varies. As down force has a significant effect
C=LT f L (37)
of the chassis dynamics, it is very desirable to include
0 c
it in laboratory tests using a four-poster. A number r
k 0
C D
of test rigs have been built to achieve this purpose;
they have at least two extra actuators pulling down on K=LT f L (38)
0 k
the car body. This is shown in Fig. 13. An assessment r
1 −l
C D
of the realism of the approach is given below.
L= f (39)
A ‘bicycle model’ is appropriate for analysis. 1 l
Figure 14 shows the physical system; here the r
vertical force F and pitch moment F are the aero- and the road height and pitch is
z h
dynamic forces resulting from down force actuators
z r
at the front and rear of the car. The actuators have
their own force control loops, and as the interface CDh
r =L−1 f
r
r
r
CD (40)

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

190 A R Plummer

Comparing with equation (13) In equations (44)–(46) the notation which was
introduced in section 2.4 and shown in Fig. 11 is
m 0
A (s)=
p C D b
0 i
b
s2+Cs+K (41) used. Thus, the external force which acts on the
physical car body but originates from the numerical
z model is denoted F , and when this is added to the
C D
np
x = b (42) force from the interface impedance F , then the new
p h i
b term is denoted F∞ .
i
1 0 As stated, force-controlled actuators are used. This
Q =
p C D
0 1
(43) is the only option for this application, as in some
circumstances the aerodynamic stiffness K can be
i
F
C D
very small. With position-controlled actuators, the
F∞ = z (44)
i loop equivalent to that in Fig. 15 has a gain which
F
h tends to infinity as K tends to zero, and the system
i
The physical external forces are generated by the would become unstable.
road excitation
z
CD
3.2 Parameter values
F =(Cs+K ) r (45)
pp h
r The aerodynamic down force is applied by two
The numerical model is vertical hydraulic actuators, one under the front axle
and the other under the rear. The individual actuator
F∞ =K x +F (46) frequency response is shown in Fig. 16; this has been
i i pi np
where K is a linearization of the ride-height-related estimated using sine sweep data from the real force-
i controlled actuator. If this individual response is
aerodynamic force component, and F is the force
np denoted p (s), then
with zero body displacement. Both these parameters f
are dependent on vehicle forward velocity. Figure 15
1 0
is the simplified form of Fig. 11, which applies in
this case.
P (s)=
f 0 1 f C D
p (s) (47)

Similarly, if the actuator disturbance response esti-

mated from real data as shown in Fig. 17 is denoted
r (s), then
f
R (s)=LTLr (s) (48)
f f
Note that the force controller used for the actuators
has been designed especially for this and similar
applications, and gives very good performance by
industry norms. The controller is described by
Plummer [11, 12].
The parameters for a Formula One racing car are
Fig. 15 Aero model example: a simplified block diagram contained in Table 1. Note that the aerodynamic

Table 1 Parameters for the aero model example

Vehicle model Body mass (m ) 550 kg
b
Body pitch moment of inertia (i ) 350 kg m2
b
Total front suspension stiffness (k ) 300 N/mm
f
Total front suspension damping (c ) 2 Ns/mm
f
Total rear suspension stiffness (k ) 240 N/mm
r
Total rear suspension damping (c ) 4 Ns/mm
r
Centroid to front axle (l ) 1.8 m
f
Centroid to rear axle (l ) 1.4 m
r
−83.9
C D
45.3
Aerodynamic model (for one vehicle speed) Aerodynamic stiffness (K ) kN and kNm per m and rad
i −51.9 231.6
−8.28
Nominal vertical/pitch force (F )
n C −1.49 D kN and kN m

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 191

Fig. 16 Measured actuator tracking response, p (s)

f

Fig. 17 Measured actuator disturbance response, r (s) (N/mm)

f

model is for one example forward car velocity only; be assessed. The vertical response of the body to
a range of velocities would have to be considered for vertical road excitation is shown in Fig. 18, and the
a complete analysis. pitch response of the body to pitch excitation is
shown on Fig. 19. There is a good correlation between
the MiL system and the ideal vertical response. The
3.3 Results
match between the pitch responses is not so good,
Using the models developed above, the response of but the existence of a significant pitch resonance at
the MiL system will be compared to that of the around 8 Hz is still captured in the MiL system.
emulated system. The emulated system is the same Figure 20 shows the vertical response from the real
as the MiL system of Fig. 15 but with P (s)=1 and car, which matches the responses of Fig. 18 well. It
f
R (s)=0. Any signals of interest can be compared; in is not possible to acquire real data for the emulated
f
the present study the position response of the body system under the same laboratory conditions without
to a road position input is used. This is a two-input a wind tunnel (and perhaps a moving road surface,
two-output system, but only the diagonal terms will as this affects the aerodynamic forces). The utility of

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

192 A R Plummer

Fig. 18 Aero example: the vertical response, road to body

Fig. 19 Aero example: the pitch response, road to body

the real result is to verify that the linear models poster, the vertical road excitation is imparted to
capture the main dynamic features sufficiently well the vehicle through its tyres, by placing the vehicle
to give confidence that a comparison using those on four wheel pans driven vertically by hydraulic
models is valid. actuators. In a hub-coupled rig the wheels and
tyres are removed, and the hydraulic actuators are
attached directly to the hubs. However, particularly
4 AUTOMOTIVE TESTING EXAMPLE 2: TYRE in racing cars, the tyres contribute a significant
MODEL-IN-THE-LOOP part of the corner compliance, and so the chassis
dynamics are radically altered. This is overcome if the
4.1 Modelling tyres are included as a numerical model, an approach
which has a number of potential advantages. As the
The feasibility of using a numerical tyre model for a characteristics of a hot rolling tyre on the road are
physical hub-coupled vehicle test rig is explored in not the same as those of a cold non-rolling tyre on a
this example. In a conventional tyre-coupled four- conventional tyre-coupled four-poster, then a virtual

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 193

Fig. 20 Aero example: the actual vertical response measured using the test rig

tyre has the potential to provide enhanced realism. Comparing with equation (13)
This is particularly true for racing car tyres which
m 0 c −c
exhibit very significant variation in temperature
and angular velocity. The technique also allows the
A (s)=
p C
D C 0
s
m
u
s2+ c
−c
c
c
c
D
c s

vertical dynamics of the vehicle with different tyre k −k

characteristics to be investigated quickly. In addition,
hub-coupled rigs are more appropriate where lateral
C+
D c
−k
c
k
c
c (50)

and/or longitudinal forces are included, as exerting x

these through the physical tyre may cause tyre slip.
A quarter car model is used for analysis, as shown
x = s
p x
u
C D (51)

in Fig. 21. The physical system is given by F =0 (52)

p
m 0 c −c k −k Q =[0 1] (53)
AC s
0 m D C
u
s2+
−c
c
c
c
c
D C
c s+
−k
c
c
k
c
c
DB p
Although not essential, it is assumed that the
actuators are force-controlled; this has the advantage
x 0
C D CD
× s =
x
u
F
1 i
(49) that wheel lift-off conditions could be simulated,
i.e. where the tyre stiffness is effectively zero. As the

Fig. 21 Tyre model-in-the-loop

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

194 A R Plummer

interface impedance is in the numerical model, this Table 2 Parameters for the tyre model example
example is also a case 1 configuration, and again Quarter-car parameters Sprung mass (m ) 150 kg
exhibits the special features described in section 2.4. s
Suspension stiffness (k ) 120 N/mm
c
The numerical model has no masses of its own Suspension damping (c ) 2.5 Ns/mm
c
Unsprung mass (m ) 20 kg
and can be represented by u
Tyre model Tyre stiffness (k ) 250 N/mm
t
Tyre damping (c ) 0.5 Ns/mm
F∞ =−B (s)x +F (54) t
i i pi np
where
B (s)=c s+k (55) response for this has been estimated from a test with
i t t
pink noise excitation; the tracking response is shown
F =B (s)z (56)
np i in Fig. 23.
x =Q x =x (57)
pi p p u
4.2 Assessment of MiL accuracy
Figure 22 is a block diagram of the system.
The parameters for the numerical model and To assess how realistic the MiL system is, its response
physical system are shown in Table 2. These will be compared with the ideal response, i.e. that of
are typical for a Formula One racing car. The the emulated system. The measure chosen is the
force-controlled actuator tracking and disturbance response of the tyre force (F ) to track excitation (z).
i
responses are as shown previously (Figs 16 and 17). This is shown in Fig. 24 for both the model-in-the-
However, for part of the analysis below, a wider loop (actual) and ideal systems. The bounce mode
frequency range is required, and the frequency of the car body at approximately 4 Hz is displayed
reasonably well in the MiL system response, but there
is very significant deviation above 10 Hz: the actuator
dynamics have introduced a very lightly damped
resonance at 17 Hz.
Breaking the loop at (say) the summing junction
in the numerical model in Fig. 22, and considering
the open-loop transfer function, gives the frequency
response shown in Fig. 25. As the emulated system
is passive, it is unconditionally stable, and the phase
lag is always less than 180°. However the phase lag
for the MiL system does increase beyond 180°, and
so the presence of the actuator dynamics has
introduced the possibility of instability. The gain
Fig. 22 Tyre model example: a simplified block diagram margin in this case is 1.4.

Fig. 23 The measured actuator tracking response up to 140 Hz

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 195

Fig. 24 The tyre force response to road input (N/mm), comparing the actual (MiL) response
with the ideal (emulated system) response

Fig. 25 Tyre example: the open-loop frequency response

4.3 The improvement of MiL accuracy dynamics. The frequency response of the filter is
shown in Fig. 26. The combination of this filter
To improve the realism of the MiL system, a filter with P (s) is shown in Fig. 27. The phase lag in the
can be used to shape the actuator command before f
0 to 30 Hz range has been reduced by 60 per cent.
being used to drive the actuator, i.e. before P (s) in The resulting MiL system response is now closer to
f
the block diagram. The intention is to cancel the the ideal, as seen in Fig. 28. However, the control
response lag of the actuator. In reality, noise will limit sensitivity, defined in this case as the sensitivity of
what can be achieved at high frequency, and so a the actuator command signal to measurement noise,
filter which rolls off above 100 Hz is included. A is now much greater (Fig. 29). Peaking at 12 N/mm of
least-squares fit in the frequency domain, using the measurement noise, this is about 30 times larger than
frequency band 0–30 Hz, is used to determine the before at around 100 Hz. In reality, measurement
best command filter to cancel the actuator tracking noise is likely to be a few microns, and the noise

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

196 A R Plummer

Fig. 26 The actuator command filter for improving tyre MiL accuracy

Fig. 27 The actuator tracking response including a command filter

amplification may be unacceptable. In addition, it Sensors, actuators, and a real-time computing plat-
is much more likely that a vigorous vertical road form are required to implement the numerical model
input signal will cause saturation of the hydraulic in the MiL system. All three of these elements will
servovalve. exhibit non-ideal behaviour and, hence, introduce
errors into the characteristics of the whole system.
Generally, the actuator dynamics and sensor noise
5 CONCLUSIONS will be the most significant sources of error.
Two automotive testing examples, both parti-
A general framework for modelling MiL systems has cularly relevant to the motor sports sector, have been
been presented. This has included the case where presented. The analysis of a real-time aerodynamic
local force control loops are closed around the model in combination with a Formula One car test
actuators, as well as the case where local position rig indicated an acceptable degree of realism. Experi-
control loops are closed around the actuators. mental results validated the modelling assumptions.

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 197

Fig. 28 The tyre force response to road input (N/mm), showing the MiL system accuracy with
the actuator command filter

Fig. 29 The sensitivity of the actuator command signal to measurement noise

However, an analysis of a real-time tyre model for a damped resonance in the emulated system which
similar rig indicated very significant error. Partially involves interaction between (say) stiffness in the
cancelling the actuator tracking dynamics with an numerical model with mass in the physical test
additional filter much improved the realism, but also system, then actuator phase lag of just a few degrees
greatly increased the sensitivity to measurement at the resonant frequency could be a problem.
noise. Actuator tracking response compensation is a
It is clear from these examples that, although the sensible way of improving accuracy. The delay com-
MiL paradigm is a powerful one, such systems must pensation technique popular in the substructuring
be implemented with care. Within the automotive literature is based on a very simple actuator model
testing industry, there is evidence of a lack of under- (a delay), whereas in this study the actuator is
standing of this fact. To assess MiL systems it is characterized by its frequency response. Also, the
essential to consider the effect of the actuator increasing magnitude with frequency of the delay
dynamics within the loop; purely having a ‘fast’ compensation filter is not generally recognized. The
actuator may not be enough. If there is a lightly design of the actuator command filter is a balance

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

198 A R Plummer

between minimizing the error between the MiL 12 Plummer, A. R. A servohydraulic control system for
and emulated systems, without an unacceptable implementing virtual components in mechanical
increase in measurement noise sensitivity. As such, systems. In Proceedings of Workshop on Power
transmission and motion control 2003, Bath,
the problem is amenable to optimization, i.e. the
September 2003, pp. 131–146.
minimization of a norm measuring a combination of
these criteria. This will be the subject of future work.
APPENDIX

REFERENCES Notation
A (s) characteristic matrix polynomial
1 Horiuchi, T., Inoue, M., Konno, T., and Namita, Y. n
for numerical model
Real-time hybrid experimental system with real-time A (s) characteristic matrix polynomial
actuator delay compensation and its application to p
for physical model
a piping system with energy absorber. Earthquake
Engng Struct. Dynamics, 1999, 28, 1121–1141. B (s) interface displacement to force
i
2 Darby, A. P., Williams, M. S., and Blakeborough, A. transfer function (termed interface
Stability and delay compensation for real-time sub- impedance)
structure testing. J. Engng Mech., December 2002, c damping element in numerical
1276–1284. n
model (scalar)
3 Lim, C. N., Neild, S. A., Stoten, D. P., and Drury, D. c damping element in physical
Real-time dynamic substructure testing via an p
model (scalar)
adaptive control strategy. First International Con-
ference on Advances in experimental structural C damping matrix for numerical–
i
engineering, Nagoya, Japan, July 2005, pp. 393–400. physical interface
4 Wagg, D. J. and Stoten, D. P. Substructuring of C damping matrix in numerical
n
dynamical systems via the adaptive minimal con- model
trol synthesis algorithm. Earthquake Engng Struct. C damping matrix in physical model
Dynamics, 2001, 30, 865–877. p
C ,C ,C damping matrices giving
5 Neild, S. A., Drury, D., and Stoten, D. P. An improved nn np pp
interaction between numerical and
substructuring control strategy based on the adaptive
minimal control synthesis algorithm. Proc. IMechE, physical systems
Part I: J. Systems and Control Engineering, 2005, 219, D (s) reformulation of numerical model
n
305–317. as shown in equation (20)
6 Hogan, N. Impedance control: an approach to D (s) reformulation of physical model as
p
manipulation. Parts I–III. ASME J. Dynamic Syst. shown in equation (22)
Measmt Control, 1985, 107, 1–24, F interface force
7 Lipscombe, J. M. and Lewis, D. J. G. Feel-force i
F∞ interface force vector, including
system with an inertia reduction capability. In i
numerical external force acting on
Proceedings of the AGARD Conference on Flight
simulation, The Hague, 1975. physical mass
8 Zhang, R. and Alleyne, A. G. Dynamic emulation F external force vector acting on
n
using resistive control input with an electro- numerical model masses
hydraulic example. In Proceedings of the Bath F external force vector acting on
Workshop on Power transmission and motion con- p
physical masses
trol, September 2002, 245–260. F physical force vector acting on
9 Guerrier, P. K. and Edge, K. A. Hydraulic emulation pp
physical masses
in injection moulding. ASME international mech-
anical engineering congress and exposition, New F numerical force vector acting on
np
York, USA, November 2001, FPST-25011. interface physical masses
10 Batterbee, D. C., Sims, N. D., and Plummer, A. R. k stiffness element in numerical
n
Hardware-in-the-loop simulation of a vibration model (scalar)
isolator incorporating magnetorheological fluid k stiffness element in physical model
damping. II ECCOMAS Thematic Conference on p
(scalar)
Smart structures and materials. Lisbon, Portugal,
K stiffness matrix for numerical–
July 2005. i
11 Plummer, A. R. Integrating virtual components into physical interface
mechanical systems: some actuator control issues. K stiffness matrix in numerical
n
IMechE Seminar on Recent advances in control tech- model
niques for actuators, London, December 2002. K stiffness matrix in physical model
p

Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering JSCE207 © IMechE 2006

Downloaded from [Link] by guest on September 6, 2012

 Model-in-the-loop testing 199

K ,K ,K stiffness matrices giving interaction Aerodynamic model example

nn np pp
between numerical and physical
c front axle damping
systems f
c rear axle damping
m position measurement noise r
d C suspension damping matrix
m force measurement noise
f F vertical aerodynamic force
m mass element in numerical model z
n F pitch aerodynamic moment
(scalar) h
i pitch moment of inertia
m mass element in physical model b
p k front axle stiffness
(scalar) f
k rear axle stiffness
M numerical mass matrix r
n K suspension stiffness matrix
M physical mass matrix
p l centroid to front axle distance
P (s) actuator position tracking response f
d l centroid to rear axle distance
transfer function matrix r
L axle to modal coordinate trans-
P (s) actuator force tracking response
f formation
transfer function matrix
m body mass
Q numerical interface variable b
n p (s) actuator transfer function, force
selector matrix f
command to force output
Q physical interface variable selector
p r front road vertical displacement
matrix f
r (s) actuator transfer function, position
R (s) disturbance rejection response f
d disturbance to force output
transfer function matrix in position
r rear road vertical displacement
control r
z vertical body displacement (under
R (s) disturbance rejection response b
f centroid)
transfer function matrix in force
z vertical road displacement (under
control r
centroid)
s differential operator
x numerical position vector
n h body pitch angle
x numerical interface position vector b
ni h road pitch angle
x physical position vector r
p
x physical interface position vector
pi
z displacement in numerical model Tyre model example
n
(scalar)
z displacement in physical model c quarter car suspension damping
p c
(scalar) c tyre damping
t
k quarter car suspension stiffness
c
k tyre stiffness
Diacritical mark t
m sprung mass
s
ˆ indication of measured position m unsprung mass
u
or force, or commanded position or x sprung mass displacement
s
force, rather than the actual value x unsprung mass displacment
u
achieved z road displacement

JSCE207 © IMechE 2006 Proc. IMechE Vol. 220 Part I: J. Systems and Control Engineering

Downloaded from [Link] by guest on September 6, 2012

View publication stats