Modeling and optimization of height-related

geometrical parameters for thin wall structures

manufactured by metal additive manufacturing
Mostafa Rahmani Dehaghani
Yifan Tang
Suraj Panicker
Di Wu
Eric Coatanea
G. Gary Wang (  gary_wang@[Link] )
Simon Fraser University Faculty of Applied Sciences

Research Article

Keywords: cold metal transfer, wire arc additive manufacturing, geometrical deviation, multi-objective
optimization

Posted Date: August 3rd, 2023

DOI: [Link]

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Version of Record: A version of this preprint was published at The International Journal of Advanced
Manufacturing Technology on November 9th, 2023. See the published version at
[Link]
 Modeling and optimization of height-related geometrical parameters for thin wall

structures manufactured by metal additive manufacturing

M. Rahmani Dehaghani a, Yifan Tang a, Suraj Panicker b, Di Wu b, Eric Coataneab, G. Gary Wang a 1
a School of Mechatronic Systems Engineering, Simon Fraser University, Surrey, Canada
b Faculty of Engineering and Natural Sciences, Automation Technology and Mechanical Engineering, Tampere
University, Tampere, 33720, Finland

Abstract

Cold metal transfer wire arc additive manufacturing (CMT-WAAM) has attracted

attention in recent years due to its ability to print walls with higher dimensional accuracy

than regular WAAM. To print near net shape parts by CMT-WAAM, there is a need to define

a set of height-related geometrical parameters (HGPs) that can capture, quantify, and

compare the quality of the height of the produced parts. In the presenting study a set of HGPs,

namely, the average height error (AHE), maximum height variation (MHV), and average

absolute slope (AAS) are defined and assessed. Fifteen single-track multi-layer walls are

printed to check the effect of process parameters on the defined HGPs. It is found that the

stability and quality of the print cannot be guaranteed by checking the visual appearance of

the single beads and at least five-to-ten-layer walls should be printed. It is also found that

the travel speed (TS) and wire feed speed (WFS) have positive monotonic relationships with

AAS and MHV, respectively. Correlations between process parameters and HGPs are

modeled and optimized using multi-objective optimization and a validation test is performed

to check the validity of the developed models. Moreover, HGPs of walls printed using

unidirectional and bidirectional path strategies are calculated and compared. Defined HGPs

1 Corresponding Author, Email: gary_wang@[Link]

1
are able to quantify, capture, and compare the quality of height of a wall with only three

parameters. The HGPs can be used in further studies to report and compare the quality of

height of thin wall structures.

Keywords: cold metal transfer, wire arc additive manufacturing, geometrical deviation,

multi-objective optimization.

Acknowledgement

The authors gratefully acknowledge funding from the Natural Science and Engineering

Research Council (NSERC) of Canada [Grant numbers: RGPIN-2019-06601] and Business

Finland under Project #: 4819/31/2021 with affiliation to the Eureka! SMART project

(S0410) titled “TANDEM: Tools for Adaptive and Intelligent Control of Discrete

Manufacturing Processes.”

2
 1. Introduction

Wire arc additive manufacturing (WAAM) is one of the direct energy deposition (DED)

techniques that can produce large-scale metallic parts [1]. Compared to other metal additive

manufacturing (AM) techniques, WAAM provides higher material deposition rate which is

favorable as it reduces the production time, but it suffers from dimensional inaccuracy,

which limits its applications [2].

From the material point of view, WAAM is able to print materials with low weldability

such as aluminum and low steel alloys [3, 4] especially when it is combined with cold metal

transfer (CMT) mode developed by Fronius. In CMT, the material deposition during welding

happens in two steps. First, the welding torch generates an arc between the welding wire

and the surface where the weld needs to be deposited. Second, the welding system detects a

short-circuit in the arc and in turn retracts the wire, stopping the arc. Hence, CMT allows the

molten metal droplet to transfer from the wire tip to the melt pool at a much lower heat

energy input. Following the metal transfer the arc is turned back on by feeding the wire back

towards the weld surface. This process is repeated with a frequency (50-120 Hz) based on

predefined synergic lines. It is shown that using CMT increases the controllability over the

metal transfer and reduces the heat input to the part [5]. Although some variants of CMT-

WAAM have improved the dimensional accuracy in printing single track walls, the effect of

CMT-WAAM process parameters on the geometry of the final parts is yet to be investigated

and understood [6].

As for geometry, many industries such as automotive, aerospace, and medical

technologies use thin-walled features to make lightweight parts. However, most of the thin-

walled parts are machined from the bulk due to the tight mechanical tolerances, which is

3
time-consuming as up to 95% of the material volume is to be removed to shape the final part

[7]. Producing parts with near net shape and reduced surface waviness will reduce the

amount of post process machining that is required. Hereby, producing near-net-shape thin

wall structures by WAAM have been investigated by researchers. Some researchers assessed

the effect of process parameters on the geometry of the single beads which includes bead

width and height. Table 1 summarizes the effect of each process parameter on the bead

geometry for steel. The correlations between process parameters and bead height and width

in WAAM process are described by positive monotonic (), negative monotonic (), and

non-monotonic (NA). Studies that had different results were separated by “/”. In summary

and especially in CMT-WAAM process, for low alloy steel the most effective parameters are

the wire feed speed (WFS) and travel speed (TS) that determine the geometry and properties

of the final part [8]. In fact, the dimensionless ratio of WFS to TS plays an important role in

determining the geometry of the bead [9]. Nagarajan et al. [10] investigated the effect of

process parameters on the geometry of the bead and found the range for the ratio of WFS to

TS to be between 2.5 and 5 for mild steels in order to have a stable and continuous bead. As

for high strength low alloy (HSLA) steel, the ratio needs to be larger. Panicker et al. [11]

printed single track HSLA multi bead walls with the WFS/TS ratio between 5-10.

Table 1. Effect of process parameters on the bead geometry in WAAM processes made by steel.
(positive monotonic (), negative monotonic (), and non-monotonic (NA)).

Bead Bead
Process parameter References
width height
Wire feed speed (WFS) /NA  [12–14]
Travel speed (TS)   [12, 14, 15]
Wire diameter   [12]
Welding voltage  /NA [12, 14]

4
 Welding current   [15, 16]
Wire stick-out* /NA /NA [12–14]
* Distance between welding torch and top layer metal

Some researchers explore the relationships between the geometry of the thin wall

structures and process parameters. For instance, Kazanas et al. [17] investigated the effect

of WFS and TS on the effective wall thickness (which has direct relationship with the bead

width) and found that effective wall thickness is affected by both WFS and TS positive-

monotonically and negatively-monotonically, respectively. They also found that increasing

the wire diameter increases the surface waviness. Yang et al. [18] found that increasing the

inter-layer cooling time decreases the height variation in the thin walls, however, excessive

dwell time cannot decrease the heat accumulation efficiently and will make the deposition

process time-consuming. Maintaining a steady interpass temperature between two

consecutive layers resulted in homogeneous microstructure and mechanical properties for

HSLA steel walls [19].

As for modelling the part geometry with respect to process parameters, Bharat and

Anandakrishnan [20] fitted a linear regression model between process parameters and bead

geometries. The accuracy of the fitted models was assessed by printing single beads as

confirmation tests. Moreover, Gonzalez et al. [16] found the optimal process parameters by

printing single beads. Afterward, they used the optimal process parameters to print thin

walls. Ma et al. [21] developed a second order regression model that predicts the bead width

and height and used their model to develop a path planning methodology for thin walls with

variable thickness. Kumar et al. [22] also developed second order response surface models

5
to predict the bead geometries and used the model to produce near net shape thin wall

structures. Although there have been a lot of efforts to model and minimize the microscale

surface roughness of metallic parts produced by DED processes [23–25], there is a scarcity

of studies that model and minimize the macroscale geometrical deviations of the parts. In a

notable contribution by Lehmann et al. [26], a comprehensive model was developed to

investigate wall width deviations, with consideration given to process parameters such as

WFS and TS. Notably, this study deviated from the prevailing practice of using bead printing,

instead opting for wall printing to determine optimal process parameter ranges. Ten-layered

walls were printed to ascertain the most favorable conditions, with the wall width deviation

(or surface waviness) serving as the response factor.

Although the wall waviness and effective wall width are established as variables to

compare the quality of the width of the walls, it seems that there is not an established

variable to show the quality of the height of a wall. For example, Galeazzi showed the quality

of the printed walls in CMT-WAAM process by reporting the deviation of the wall height from

the mean value at different points [6]. One of the objectives of the study reported in our

present work is to define a set of geometrical parameters that can characterize the height

quality of a wall.

This study aims to improve the quality of the height of the HSLA steel walls

manufactured by CMT-WAAM under a fixed inter-pass temperature. As mentioned before,

most of the prior work that modeled the geometry of the thin wall structures trained their

models based on the single bead data whereas, in reality, the geometry of a wall is not just

related to the geometry of the beads. In other words, the training data to model the geometry

6
of the thin wall structures should be obtained from printed walls. Hence, the objectives of

this study are defined as below:

• Defining a set of geometrical parameters that specifies the quality of the height of any

thin wall (height-related geometrical parameters (HGPs)).

• Studying process parameters with printed thin walls based on a design of experiment

(DOE) approach.

• Developing different models to predict the HGPs of thin walls and comparing them.

• Multi-objective optimization of geometrical parameters and printing the optimal thin

walls for validation.

• Comparing the HGPs of thin walls printed unidirectionally and bidirectionally.

The remainder of the paper is organized as follows. Section 2 explains the methods and

materials used in this research. Section 3 presents the results of the experiments and models.

The results are further discussed in Section 4 and, finally, the conclusions are drawn in

Section 5.

2. Materials and methods

In the following subsections, the experimental setup, choice and range of process

parameters, definition of HGPs, employed modeling approach, and the multi-objective

optimization method are described in detail.

2.1. Material and experimental setup

For the CMT-WAAM experimental work, the AM70 alloy steel wire feedstock (1.2 mm

diameter) is used; its chemical composition is presented in Table 2. An S355 mild steel plate

(300 × 200 × 20 mm) is used as a substrate for the fabrication of a thin wall. The CMT-WAAM

7
setup (Figure 1) consists of a welding unit (Fronius CMT Advanced 4000) with an attached

wire feeder system, a 6-axis industrial robot with a 2-axis worktable (ABB 4600 40/2.55),

and an inert gas supply. The welding torch is mounted on the industrial robot arm such that

the travel direction is in the X axis, build direction is in the Z axis, and traverse direction is in

the Y axis. The computer aided design file of the printing geometry is generated in

SOLIDWORKS and imported to the robot path programming software, Robot Studio 6. A

contact-type digital thermal probe (Fluke 53 IIB) is used to monitor the interpass

temperature between consecutive weld passes to be between 150-200 degrees Celsius.

Table 2. AM70 wire feedstock chemical composition [27].

Element C Mn Si Cr Mo Ni Fe
Percentage (wt. %) 0.08% 1.70% 0.60% 0.20% 0.5% 1.50% 95.42%

Figure 1. CMT-WAAM equipment setup.

2.2. Process parameter selection and DOE

Process parameters, assumed to have the most significant effects on the final geometry

of the part, are selected to be modeled in this study. TS and WFS are selected as they are the

most effective parameters in the CMT-WAAM process [8]. Moreover, as shielding gas flow

8
affects the convection heat transfer during the wall printing, it is hypothesized that it affects

the quality of the height of the wall. Hence, the shielding gas flow rate (SGFR) is selected as

the third process parameter. The purpose of performing experiments is to find the

mathematical models between inputs (process parameters) and the outputs (HGPs). Since

the model may not be linear, three levels are selected for each process parameter. The range

for TS and WFS is selected in a way that the WFS/TS ratio varies from 3.33 to 12.5 as this

range includes the range used by Panicker et al. [11] for low alloy steels. Moreover, the range

of SGFR is chosen to include SGFR values used by other studies done by CMT-WAAM [28–

30]. Selected process parameters, levels, and their corresponding ranges are shown in Table

3In the CMT-WAAM process, the wire is continuously retracted and pushed at a high

frequency (50-120 Hz). Consequently, WFS value is not a static, fixed value. However, for the

purpose of this study, we focused on a specific aspect. Specifically, we utilized the WFS value

set by the operator prior to the deposition process for the modeling and optimization.

Table 3. The range for process parameters and value for each level.

Level 1 Level 2 Level 3

Process parameter Factor Range
(-1) (0) (1)
TS (mm/s) A 8-15 8 11 15
WFS (m/min) B 3-6 3 4.5 6
SGFR (L/min) C 12-20 12 16 20

Full factorial design of experiments (DOE) yields 27 (33 ) experiments to be performed,

which is time consuming and cost inefficient. Therefore, the central composite face DOE is

used with 15 experiments for three levels and three factors. Table 4 shows all the 15

experiments and the corresponding level of each of the process parameters.

9
 Table 4. Central composite face DOE for three factors and three levels.

Exp. Number Factor A Factor B Factor C

1 -1 -1 -1
2 -1 1 1
3 -1 1 -1
4 -1 -1 1
5 1 -1 1
6 1 -1 -1
7 1 1 -1
8 1 1 1
9 0 0 0
10 -1 0 0
11 1 0 0
12 0 -1 0
13 0 1 0
14 0 0 -1
15 0 0 1

For each set of process parameters, first, a three-layer wall is printed to check if the

correspondent set of process parameters would lead to a valid geometry. Then, a single-track

multi-layer wall is printed for each DOE point with the length of 160 mm. All of the walls are

printed unidirectionally in this stage of the experiments. Unidirectional and bidirectional

printing are the two wall scanning strategies that are defined in Figure 2.

10
 Figure 2. a) Bidirectional and b) unidirectional printing directions.

2.3. Pilot experiments

To understand the feasibility of the selected set of process parameters, 15 three-layer

walls (one for each DOE point) are printed and the height and width of the walls are

measured with a caliper (Least count = 0.05 mm) from three different locations (every 40

mm after beginning of the bead). The visual appearance of the walls is checked to ensure the

feasibility of the print of selected DOE points. Another purpose of pilot experiments is to find

the appropriate step height for the next set of experiments. It is crucial to understand the

distinction between step height and layer height. The step height is a predetermined

parameter, while the layer height is the actual height of each layer determined during the

deposition process. To have high-quality prints, it is important to choose the step height as

close as possible to the average layer height. Hence, the average height of the walls in the

11
pilot experiments is chosen as the step height of the robot for the actual wall printing

process.

2.4. HGPs definition and geometry scanning

As mentioned earlier, a set of geometrical parameters that can specify the quality of the

height of the wall is not established yet. The objective of this section is to define a set of

parameters that characterizes the quality of the height of thin walls. This set of parameters

are called height-related geometrical parameters, HGPs. Application of HGPs is not limited

to thin walls manufactured by CMT-WAAM and one can utilize HGPs to characterize,

compare, and improve the quality of the height of walls manufactured by other metal

additive manufacturing techniques.

First, the average height of a wall should be as close as possible to the preset height.

Hence, the first parameter of the HGP is the average height error (AHE). AHE is defined as

the absolute difference between the average height of the wall per layer and the step height.

AHE can be represented in percentage to get a better understanding of the value of the error.

Mathematically, AHE can be represented as:

𝐻𝐻𝐴𝐴𝐴𝐴𝐴𝐴
− 𝑆𝑆𝑆𝑆
𝐴𝐴𝐴𝐴𝐴𝐴 = � 𝑛𝑛
𝑆𝑆𝑆𝑆
� (1)

where, 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 is the average height of the wall over its length, 𝑛𝑛 is the number of layers, and

𝑆𝑆𝐴𝐴 is the step height preset to print each wall.

The second height-related feature of a desired wall is the difference between maximum

and minimum heights of the wall, which should be kept minimum. Hence, the second

parameter of HGPs is the maximum height variation (MHV). MHV can be written as:

12
 𝑆𝑆𝑚𝑚𝑚𝑚𝑚𝑚 −𝑆𝑆𝑚𝑚𝑚𝑚𝑛𝑛
𝑀𝑀𝐴𝐴𝑀𝑀 = � � (2)
𝑛𝑛

where, 𝐴𝐴𝑚𝑚𝑚𝑚𝑚𝑚 is the maximum height of the wall and 𝐴𝐴𝑚𝑚𝑚𝑚𝑛𝑛 is the minimum height of the wall.

To reduce the effect of number of layers, the difference is divided by 𝑛𝑛.

Finally, the third characteristic of a flat wall is that it should have the lowest average

slope through its length. Therefore, the last parameter of HGPs is the average absolute slope

(AAS) of the wall. AAS can be formulated as:

𝛿𝛿𝛿𝛿
𝑚𝑚𝐴𝐴𝑚𝑚𝑛𝑛�� ��
𝛿𝛿𝑚𝑚
𝐴𝐴𝐴𝐴𝑆𝑆 = (3)
𝑛𝑛

where, 𝑥𝑥 and 𝑧𝑧 axes align with travel and build directions, respectively, as shown in Figure

𝛿𝛿𝛿𝛿
3. The term 𝛿𝛿𝑚𝑚 is the slope of the wall at each point along the 𝑥𝑥 axis on the top layer and

𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛(⋅) outputs the mean value of the slopes. A set of HGPs is a vector that describes the

quality of the height of a wall. Equation 4 defines a set of HGPs for any wall. The lower is each

one of the parameters, the better is the quality of the wall height. In other words, the wall

with the perfect height can be described as a zero vector (𝐴𝐴𝐻𝐻𝐻𝐻𝐻𝐻 = (0, 0, 0)).

𝐴𝐴𝐻𝐻𝐻𝐻𝐻𝐻 = (𝐴𝐴𝐴𝐴𝐴𝐴, 𝑀𝑀𝐴𝐴𝑀𝑀, 𝐴𝐴𝐴𝐴𝑆𝑆) (4)

The ability of the HGPs to quantify and capture the quality of the height of thin walls can

be further explored by looking at Figure 3. Since all of the walls have the same average height,

they have the same AHE value. However, AAS value of the wall in Figure 3(c) is lower than

those in Figures 3(a) and 3(b). On the other hand, the wall in Figure 3(b) has a lower MHV

value than those in Figures 3(a) and 3(c). One can understand and compare the behavior of

heights of different walls by taking a look at the HGP vector of the walls. It is important to

highlight that in this work HGPs are applied to thin wall structures made by single-track

13
printing. However, the application of HGPs is not limited to single track structures. One can

use HGPs to quantify and compare the quality of height of multiple track structures as long

as the thickness of structure is negligible in relation to other dimensions.

Figure 3. Schematic of thin walls with different height variations and HGP values.
After performing the experiments, each one of the walls is scanned using a Hexagon

Metrology laser scanner (HP-L-20.8 7520SE) with the scanning accuracy of 53 microns [31].

The scanned points are detected by PolyWorks 2017, and the cloud point file is saved as an

“.stl” file and imported to SOLIDWORKS. The height of the wall at each point is obtained by

fitting a spline to the top of the wall. The beginning and end of the wall (beginning and end

of the spline) are selected at points at which the slope of the wall is greater than a large

𝛿𝛿𝛿𝛿
number (e.g., 100) (�𝛿𝛿𝑚𝑚� > 100) to avoid possible ambiguity caused by edge defect. The

equally spaced points of the spline are output to a MATLAB script where the HGPs are

determined for each wall.

2.5. Statistical analysis, modeling, and optimization

After scanning the walls and preparing responses (HGPs) for all of the walls, single

variate Analysis of Variances (ANOVA) is performed to understand the significance of

process parameters on HGPs of the wall. From the result of ANOVA, the responses (HGPs)

that are affected significantly by process parameters are modelled. K-fold cross validation is

used as only limited data is available. Four different metamodeling approaches including

14
gaussian process regression (GPR), kriging (KRG), gaussian radial basis function (GRBF), and

second order response surface model (RSM) are used. The difference between GPR and KRG

is that GPR has a random noise added to the base KRG model. The metamodeling approach

that leads to the highest R-squared (𝑅𝑅 2 ) value is selected as the final model.

The process parameters that lead to the lowest HGPs are selected by optimizing the

modeled HGPs using multi-objective optimization (MOO). MOO is performed using OASIS

software tool and one point is selected from the pareto frontier as the optimal point. OASIS

is an optimization software tool that is designed to solve low to high dimensional

single/multi objective optimization problems [32]. The optimal set of process parameters is

used to print two walls to verify the quality improvement and the model’s accuracy.

2.6. Comparing unidirectional and bidirectional printing

Using the optimal process parameters found in the previous step, a single-track multi-

layer wall is printed with the bidirectional printing strategy. The HGPs of the wall are

compared with the wall printed unidirectionally.

3. Results

3.1. Pilot experiments

A three-layer single-track wall is printed for each set of process parameters. The visual

appearance of some of the DOE points is shown in Figure 4. From the visual appearance of

the beads, it is concluded that all the DOE points lead to feasible prints, and the process is

stable for all cases. Moreover, the average layer height, width, and WFS/TS ratio for each

experiment are shown in Table 5. The step height for the experiments described in Section

3.2 is specified as the average layer height of the pilot experiments. It is also worth noting

that the WFS/TS ratio is highly correlated with the layer height and width, and it can be seen

15
in Table 5. The relationship between WFS/TS, wire and bead cross sectional areas can be

explained by Equations 5 and 6. Using the conservation of mass, it can be concluded that:

𝑊𝑊𝑊𝑊𝑆𝑆 × 𝐴𝐴𝑤𝑤𝑚𝑚𝑤𝑤𝐴𝐴 = 𝑇𝑇𝑆𝑆 × 𝐴𝐴𝑏𝑏𝐴𝐴𝑚𝑚𝑏𝑏 (5)

where, 𝐴𝐴𝑏𝑏𝐴𝐴𝑚𝑚𝑏𝑏 and 𝐴𝐴𝑤𝑤𝑚𝑚𝑤𝑤𝐴𝐴 are the bead and wire cross sectional areas, respectively. By dividing

two sides of Eq. 5 by 𝑇𝑇𝑆𝑆 × 𝐴𝐴𝑤𝑤𝑚𝑚𝑤𝑤𝐴𝐴 it is derived that:

𝑊𝑊𝑊𝑊𝑆𝑆 𝐴𝐴𝑏𝑏𝐴𝐴𝑚𝑚𝑏𝑏
𝑇𝑇𝑆𝑆
= (6)
𝐴𝐴𝑤𝑤𝑚𝑚𝑤𝑤𝐴𝐴

where the WFS/TS ratio is equal to the ratio of the cross-sectional area of the printed bead

to the wire cross sectional area. As the wire cross section is constant during the process, the

higher the WFS/TS ratio, the higher the printed bead cross-sectional area, which is the

double integral of the layer height and width.

Table 5. Average layer height for each of the DOE points calculated from three-layer beads.

Average layer Average layer WFS/TS ratio

Exp. number
height (mm) width (mm)
1 1.90 5.77 6.25
2 1.82 8.55 12.5
3 1.89 8.77 12.5
4 1.70 5.40 6.25
5 1.34 4.13 3.33
6 1.30 4.03 3.33
7 1.59 6.70 6.67
8 1.43 6.70 6.67
9 1.64 5.52 6.82
10 1.79 7.27 9.38
11 1.87 6.88 5.00
12 1.36 4.65 4.55
13 1.69 7.97 9.09
14 1.62 7.03 6.82
15 1.75 6.13 6.82

16
 Figure 4. Bead visual appearance of beads printed with some of the DOE points from the
pilot experiments.

3.2. Printing multi-layer thin walls and performing ANOVA

After printing the multi-layer thin walls, three of the walls (DOE points 5, 6, and 12) are

rejected as they have sawtooth waves on the surface and calculating the HGPs is neither

feasible nor reasonable for these DOE points. The sawtooth waves can be seen for DOE points

6 and 12 in Figure 5. Although the appearance and geometry of the three-layer wall with the

same process parameters are checked and approved in the pilot experiment, the final

geometry after the print is not desirable.

17
 Figure 5. Sawtooth waves of the multi-layer walls printed with DOE points 6 (top) and 12
(bottom).

HGPs vectors for all the 12 remaining DOE points are calculated and ANOVA is

performed for each of the responses including the average height error (AHE), maximum

height variation (MHV), and average absolute slope (AAS). Table 6 summarizes the ANOVA

results for AHE, MHV, and AAS. It can be seen that the most significant impacts are 1) the

effect of wire feed speed on the maximum height variation (P-Value = 0.041) and 2) the effect

of travel speed on the average absolute slope (P-Value = 0.0865). Based on the P-values

presented in Table 6, it can be concluded that the WFS and TS are the most significant factors

for MHV and AAS, respectively. On the other hand, none of the parameters seems to have

significant effect on AHE.

Table 6. Summary of the ANOVA for all three responses (AHE, MHV, and AAS).

P-Value
Process parameter
AHE MHV AAS
Travel speed (TS) 0.1971 0.1357 0.0865
Wire feed speed (WFS) 0.9297 0.041 0.3717
Shielding gas flow rate (SGFR) 0.2834 0.5583 0.4067

18
3.3. Modeling of height-related geometrical parameters
To further investigate the effect of process parameters on MHV and AAS, these two

parameters are modeled with different metamodeling approaches. The relationship between

the process parameters and AHE is not modelled as it seems that none of the selected process

parameters have significant impact on AHE. As the number of DOE points is limited (12 data

points), 4-fold cross validation is used. In 4-fold cross validation, four metamodels are

developed and the final predicted value is the average value of all four models. The first

metamodel is trained using data points 4 to 12 (9 points). The second one is trained using

data points 1-3 and 7-12 (9 points in total), and so on. The final model is the average value

of these four models. Tables 7 and 8 show different prediction and experimental values of all

the 12 experiments for AAS and MHV, respectively. The last row compares the R-squared

value calculated on all of the data points for different metamodels. In Table 7, GRBF model

leads to the highest R-squared value (𝑅𝑅 2 = 0.9499), hence GRBF is chosen as the model to

predict AAS, the average absolute change of the height per unit length per layer of the wall.

By comparing the experimental AAS values (last column) under different process

parameters, it can be understood that the AAS and TS have positive monotonic relationship.

It means that increasing the travel speed leads to more variation on the slope of the thin

walls.

Table 7. Experimental and model values of AAS of walls printed with different DOE points.
Process parameters AAS model predictions
Exp. Experimental
number TS WFS SGFR AAS value
GPR GRBF* KRG RSM
(mm/s) (m/min) (L/s)
1 8 3 12 0.2905 0.2801 0.2956 0.2819 0.2727
2 8 6 20 0.2327 0.2304 0.2379 0.2488 0.2297

19
 3 8 6 12 0.2009 0.2053 0.2075 0.2036 0.1878
4 8 3 20 0.2200 0.2202 0.2243 0.2926 0.2128
7 15 6 12 0.2970 0.3109 0.3050 0.3690 0.3159
8 15 6 20 0.2536 0.2657 0.2614 0.2516 0.2581
9 11 4.5 16 0.2747 0.2681 0.2655 0.2719 0.2612
10 8 4.5 16 0.2327 0.2346 0.2332 0.2505 0.2301
11 15 4.5 16 0.2855 0.3024 0.2957 0.2998 0.3007
13 11 6 16 0.2844 0.2675 0.2675 0.2859 0.2755
14 11 4.5 12 0.2699 0.2802 0.2774 0.2693 0.2796
15 11 4.5 20 0.2536 0.2532 0.2535 0.2276 0.2579
R-squared value 0.9111 0.9499 0.9413 0.2818
* The bold column shows the selected model.

For MHV, as shown in Table 8, KRG outperforms other metamodeling approaches and is

selected to be exploited in further analyses (𝑅𝑅 2 = 0.8900). It is worth mentioning that MHV

values in Table 8 represent the difference between the highest and lowest points of the wall

per layer. It can be understood from the experimental data in Table 8 that the WFS and MHV

have a positive monotonic relationship. In other words, lower WFS leads to lower height

difference.

Table 8. Experimental and model values of MHV of walls printed with different DOE points.
Process parameters MHV model prediction Experimental
Exp.
TS WFS SGFR MHV value
number GPR GRBF KRG* RSM
(mm/s) (m/min) (L/s) (mm)
1 8 3 12 0.6364 0.7771 0.6887 0.8851 0.6551
2 8 6 20 1.2486 1.2106 1.2711 1.0311 1.1265
3 8 6 12 1.1520 1.2543 1.2016 1.2403 1.2347
4 8 3 20 0.7322 0.7965 0.7745 0.6522 0.8367
7 15 6 12 0.8716 0.9806 0.9274 0.7122 0.8955
8 15 6 20 0.8712 0.8418 0.9129 0.8562 0.8949
9 11 4.5 16 1.0438 0.9692 1.0249 0.9904 1.0235
10 8 4.5 16 0.9198 0.9485 0.9849 0.8292 0.9514
11 15 4.5 16 0.8965 0.8625 0.9346 1.0705 0.9286
13 11 6 16 0.9915 1.0663 1.0433 0.9242 1.0564

20
 14 11 4.5 12 1.0183 1.0696 1.0886 1.1798 1.1251
15 11 4.5 20 0.7865 0.8388 0.8567 1.0703 0.7809
R-squared value 0.8768 0.8404 0.8900 0.2943
*The bold column shows the selected model.

3.4. Multi-objective optimization and confirmation tests

To find the optimal process parameters that lead to the lowest AAS and MHV values,

MOO is used. The multi-objective optimization is formulated using Equation 7.

min �𝐴𝐴𝐴𝐴𝑆𝑆(𝑥𝑥), 𝑀𝑀𝐴𝐴𝑀𝑀(𝑥𝑥)� (7)

Subject to:

𝑥𝑥𝑚𝑚𝑚𝑚𝑚𝑚 ≤ 𝑥𝑥 ≤ 𝑥𝑥𝑚𝑚𝑚𝑚𝑛𝑛

Where, 𝑥𝑥 is the vector of process parameters containing TS, WFS, and SGFR as the design

variables. 𝐴𝐴𝐴𝐴𝑆𝑆(𝑥𝑥) and 𝑀𝑀𝐴𝐴𝑀𝑀(𝑥𝑥) are the mathematical models of average absolute slope and

maximum height variations, respectively, developed in the previous step. 𝑥𝑥𝑚𝑚𝑚𝑚𝑛𝑛 and 𝑥𝑥𝑚𝑚𝑚𝑚𝑚𝑚 are

the bounds of the design space that can be found in Table 3 for each design variable (process

parameter)

After performing MOO and analyzing the pareto frontier, it is found that all of the optimal

points have the same TS and WFS at their lowest level (TS = 8 mm/s, WFS = 3 m/min) with

variations only in SGFR. MHV and AAS values of different Pareto frontier points can be seen

in Figure 6. The red square point is selected as the optimal point. The process parameters of

the selected point are (TS = 8 mm/s, WFS = 3 m/min, SGFR = 17 L/min). The reason to choose

this point is to have a compromise between both objectives (MHV and AAS). This

compromise is explained more in Section 4.3.

21
 0.78
0.76

MHV (mm)
0.74
0.72
0.7
0.68
0.21 0.23 0.25 0.27 0.29 0.31
AAS

Figure 6. Pareto frontier of multi-objective optimization of MHV and AAS.

To check the accuracy of the models and find process variations, two multi-layer single

track walls are printed using the optimal process parameters. The geometry of both printed

walls can be seen in Figures 7(a) and 7(b). Visually, the geometry of the walls are similar and

both of them suffer from edge defect which is a result of unidirectional printing [33]. HGPs

of both walls can be compared in the first two rows of Table 9. The reason for printing two

walls with the same process parameters is to investigate the repeatability of the process and

the effect of uncertainties on the final geometry of the wall. The differences between the HGP

values of two walls are 6.3%, 4.5%, and 2.2% for AHE, MHV, and AAS, respectively.

The prediction values for MHV and AAS can be seen in the third row of Table 9. To check

the validity of the developed models, one method is to find the confidence interval from the

experiments and test if the estimated value lies in the confidence interval. The experimental

confidence interval is the experimental mean ± experimental standard deviation. This

experimental confidence interval is equal to the interval between the experimental data

22
points if the number of experiments is two. Hence, the developed models for each one of the

responses is valid, if the estimated value is between the two experimental values. As for the

MHV, the predicted value (MHV=0.7618) is in the interval [0.7309, 0.7638] hence the

maximum height variation model is valid. However, the AAS model is not validated as the

AAS predicted value ( 𝐴𝐴𝐴𝐴𝑆𝑆 = 0.2395 ) is out of the confidence interval, which will be

discussed in Section 4.2.

Table 9. Height-related geometrical parameters of the printed walls with optimal process
parameters.

MHV
AHE AAS
(mm)
Unidirectional printing 1 0.0358 0.7638 0.2715
Unidirectional printing 2 0.0382 0.7309 0.2654
Models’ predictions - 0.7618 0.2395
Bidirectional printing 0.0144 0.1866 0.2431

Figure 7. Side view of the walls printed with optimal process parameters. a) and b) were
printed unidirectionally and c) was printed bidirectionally.

3.5. Unidirectional vs. bidirectional

One of the factors that can reduce the edge defect in thin walls is the scanning strategy.

Printing the walls bidirectionally would decrease the edge defect as described in the

literature [10, 33]. Hence, one multi-layer single track wall is printed bidirectionally. From

the visual appearance, it is found that the geometry is improved as shown in Figure 7(c),

23
compared with the walls printed unidirectionally. In terms of quantitative parameters, the

improvement can be seen by comparing the HGP values of unidirectional and bidirectional

prints in Table 9. The values of AHE and MHV are significantly reduced when the part is

printed bidirectionally. The amount of decrease of the AAS value, however, is not as

significant as the other two HGPs (AHE and MHV).

4. Discussion

4.1. Process stability and WFS/TS ratio

To check if the process is stable for all of the DOE points, three-layer walls are printed.

All of the beads are continuous, and the process seems to be stable. However, in some of the

real experiments (DOE points 5, 6, 12) and after printing 5-10 layers the process become

unstable, and a sawtooth wave pattern appears and grows layer by layer. This observation

confirms our hypothesis that to optimize the WAAM process parameters, the models should

not be trained on data gathered from bead printings as the final geometry is determined after

printing 5-10 layers. Moreover, by checking the average layer height of the DOE points 5, 6,

and 12 in Table 5, one can understand that these three DOE points have the least average

layer height (all below 1.4 mm). Therefore, if one wants to check the stability of the print, the

average layer height of the beads should be checked rather than checking the visual

appearance of the beads. Our observation shows that for any specific machine and material,

there is a layer height threshold that if the process parameters lead to a lower layer height

than the threshold, it will print an unstable wall. This threshold is found to be 1.4 mm for the

used machine and material (AM70) in this paper. This can also be found by checking the wire

feed speed to travel speed ratio as the cross section of the bead is the result of multiplication

of this ratio to wire cross sectional area (last column of Table 5). Based on the experimental

24
results, we find that the WFS/TS threshold is around 4.9 as the ratio for the failed

experiments (DOE points 5, 6, and 12) is below 4.9.

4.2. Model validity and future considerations to improve height quality

The results of confirmation tests show that the model to predict MHV is valid. However,

the AAS model is not validated as the AAS predicted value is not in the confidence interval

found from the confirmation tests. One way to validate the AAS model is to train the model

with more experiments and applying the calibration framework defined in Ref. [34] where a

bias between the predicted model and experimental values are defined and the bias is

calibrated in a loop. Although the later model is not valid, the aforementioned relationships

between process parameters and AAS still exist. For example, it is found that the travel speed

has the most significant impact on the AAS value (Table 6) and the relationship between AAS

and TS is positive monotonic. It means that by increasing the travel speed, the height of the

wall will have more variations. The P-values for AHE are all above 0.197 and thus this

parameter cannot be controlled by the process parameters defined in this study. Hence, AHE

is not modeled as a function of process parameters. The effect of other parameters such as

wire stick out, interpass temperature, arc current and voltage on AHE should be further

investigated in future studies.

4.3. ANOVA and optimization results

The result of ANOVA test shows that AAS and MHV are mostly controlled by TS and WFS,

respectively (Table 6). Moreover, the results represented on Tables 7 and 8 show that TS and

WFS have positive monotonic relationships with AAS and MHV, respectively. These results

are aligned with the optimal points on Pareto frontier, as all the optimal points have the same

TS and WFS values. The selected optimal point for validation tests is chosen in a way that can

25
be explained by Figure 8. Figure 8 shows the MHV and AAS values of the DOE experimental

points (12 data points) as well as the average values of MHV and AAS from two optimally

printed walls (red square point). The optimal wall has a lower MHV value than 11 out of the

12 walls. Similarly, the optimal wall has lower AAS value than 5 out of the 12 walls.

1.3
1.2
MHV (mm)

1.1
1
0.9
0.8
0.7
0.6
0.18 0.21 0.24 0.27 0.3 0.33
0.2727points
DOE 0.26845
Optimal point AAS

Figure 8. MHV and AAS values of DOE points and optimal unidirectionally-printed wall.

4.4. Comparison of printing strategies and HGP capabilities

By comparing the visual appearance of unidirectionally printed walls (Figures 7(a) and

7(b)) and the bidirectionally printed wall, it can be understood that the bidirectional printing

leads to less height variation and a more even wall surface. This can be confirmed by

comparing HGP values shown in Table 9. The AHE and MHV values of the bidirectionally

printed wall are significantly lower than those of the unidirectionally printed walls,

respectively. It confirms the hypothesis that the printing strategy can improve the quality of

the height of the wall and prevents edge defect in the walls [33, 35]. Moreover, it shows that

the height-related geometrical parameters defined in this study are able to represent the

26
quality of the height of the wall. By comparing the AAS value of the walls printed with

different printing strategies (the last column of Table 9) it can be understood that although

bidirectional printing leads to less AAS value (0.2431 < 0.2654), the difference is not as

significant as the difference in the other two HGPs (MHV and AHE). The reason is that AAS

captures how sharp the height changes and how many times the slope sign changes.

Although the quality of the height of the wall improves by printing bidirectionally, there are

some small waves on the top surface of the bidirectionally printed wall (Figure 7c) that are

captured by AAS. As a result, the value of the AAS does not drop significantly compared to

the walls printed unidirectionally. The small waves on the top surface of the bidirectionally

printed wall could be missed if the HGP values does not include AAS.

4.5. Model generalization and future studies

It is worth mentioning that if the developed models are intended to be applied to other

alloys or geometries, it is essential to verify the generalization of the models by conducting

similar experiments on different materials and geometries. Future studies should focus on

checking the generalization of the developed models. However, the Height-Related

Geometrical Parameters defined in this study offer a broader applicability, as they can be

utilized to assess the height quality of thin structures produced using metal Additive

Manufacturing technologies, regardless of the specific alloy, geometry, or process.

In Figure 7, it can be noticed that although the quality of the height of the walls improves

by optimizing height-related geometrical parameters, there are some bumps in an inclined

pattern and the quality of the width of the walls still needs to be improved. This can be a

future investigation where both the width and height of the single-track multi-layer walls

could be modeled and optimized.

27
 5. Conclusion

A set of height-related geometrical parameters (HGPs) are defined to represent the

quality of the height of single-track walls printed by CMT-WAAM. The central composite

faced design of experiment is used to print 15 single-track walls with three variable process

parameters, namely, travel speed (TS), wire feed speed (WFS), and shielding gas flow rate

(SGFR). It is found that the stability of the print cannot be guaranteed by printing single

beads and a wall of at least 5-10 ten layers should be printed to check if the selected

parameters lead to stable prints. A minimum threshold of 4.9 for the WFS/TS ratio is

introduced for AM70 to print single-track stable walls by CMT-WAAM. The HGPs of printed

walls are modelled and it is found that TS and WFS significantly affect the average absolute

slope and maximum height variation of the walls, respectively. The quality of the height of

the wall is optimized by performing MOO and a validation test is performed to validate the

models. It is found that the MHV model is valid, but more experiments should be performed

to validate the AAS model. With the same process parameters, a single-track wall is printed

bidirectionally to be compared with the walls printed unidirectionally. Defined HGPs are not

only able to quantify the differences in visual appearance of the height of the walls but they

could capture the quality of the height of the wall with only three parameters. The future

work would be to apply the calibration and validation framework defined in Ref. [34] to

validate the AAS model and also calibrate the uncertainties in the process. Moreover, another

future work is to model and optimize the width and height of the wall simultaneously.

References
1. Baufeld B, der Biest O Van, Gault R (2010) Additive manufacturing of Ti–6Al–4V
components by shaped metal deposition: microstructure and mechanical properties.
Mater Des 31:S106–S111.
[Link]

28
2. Ngo TD, Kashani A, Imbalzano G, et al. (2018) Additive manufacturing (3D printing): a
review of materials, methods, applications and challenges. Compos Part B Eng
143:172–196. [Link]
3. Pickin CG, Young K (2006) Evaluation of cold metal transfer (CMT) process for welding
aluminium alloy. Sci Technol Weld Join 11:583–585.
[Link]
4. Feng J, Zhang H, He P (2009) The CMT short-circuiting metal transfer process and its
use in thin aluminium sheets welding. Mater Des 30:1850–1852.
[Link]
5. Zhang HT, Feng JC, He P, et al. (2009) The arc characteristics and metal transfer
behaviour of cold metal transfer and its use in joining aluminium to zinc-coated steel.
Mater Sci Eng A 499:111–113.
[Link]
6. Galeazzi D, Silva RHG e., Viviani AB, et al. (2022) Evaluation of thermal and geometric
properties of martensitic stainless steel thin walls built by additive manufacturing cold
metal transfer (CMT) processes. Int J Adv Manuf Technol 120:2151–2165.
[Link]
7. Oliveira D, Calleja A, de Lacalle LN, et al. (2012) Flank milling of complex surface. Mach
Complex Sculpt Surfaces; Davim, JP, Ed; Springer London, UK
8. Yildiz AS, Davut K, Koc B, Yilmaz O (2020) Wire arc additive manufacturing of high-
strength low alloy steels: study of process parameters and their influence on the bead
geometry and mechanical characteristics. Int J Adv Manuf Technol 108:3391–3404.
[Link]
9. Näkki J (2018) Properties of alloy 625 claddings made with laser and CMT methods
10. Nagarajan HPN, Panicker S, Mokhtarian H, et al. (2019) Graph-based metamodeling
for characterizing cold metal transfer process performance. Smart Sustain Manuf Syst
3:169–189. [Link]
11. Panicker S, Nagarajan HPN, Tuominen J, et al. (2022) Investigation of thermal
influence on weld microstructure and mechanical properties in wire and arc additive
manufacturing of steels. Mater Sci Eng A 853:143690.
[Link]
12. Dickens PM, Pridham MS, Cobb RC, et al. (1992) Rapid prototyping using 3-D welding.
Solid Free Fabr Proc 280–290.
13. Song Y-A, Park S, Chae S-W (2005) 3D welding and milling: part II—optimization of
the 3D welding process using an experimental design approach. Int J Mach Tools
Manuf 45:1063–1069.
[Link]
14. Lu X, Zhou YF, Xing XL, et al. (2017) Open-source wire and arc additive manufacturing
system: formability, microstructures, and mechanical properties. Int J Adv Manuf
Technol 93:2145–2154. [Link]
15. Prado-Cerqueira JL, Diéguez JL, Camacho AM (2017) Preliminary development of a
wire and arc additive manufacturing system (WAAM). Procedia Manuf 13:895–902.
[Link]
16. González J, Rodríguez I, Prado-Cerqueira J-L, et al. (2017) Additive manufacturing with
GMAW welding and CMT technology. Procedia Manuf 13:840–847.
[Link]

29
17. Kazanas P, Deherkar P, Almeida P, et al. (2012) Fabrication of geometrical features
using wire and arc additive manufacture. Proc Inst Mech Eng Part B J Eng Manuf
226:1042–1051
18. Yang D, Wang G, Zhang G (2017) Thermal analysis for single-pass multi-layer GMAW
based additive manufacturing using infrared thermography. J Mater Process Technol
244:215–224. [Link]
19. Zhai W, Wu N, Zhou W (2022) Effect of interpass temperature on wire arc additive
manufacturing using high-strength metal-cored wire. Metals (Basel) 12.
[Link]
20. Bharat Kumar CH, Anandakrishnan V (2020) Experimental investigations on the effect
of wire arc additive manufacturing process parameters on the layer geometry of
Inconel 825. Mater Today Proc 21:622–627.
[Link]
21. Ma G, Zhao G, Li Z, Xiao W (2019) A path planning method for robotic wire and arc
additive manufacturing of thin-walled structures with varying thickness. IOP Conf Ser
Mater Sci Eng 470:12018. [Link]
22. Kumar A, Maji K (2020) Selection of process parameters for near-net shape deposition
in wire arc additive manufacturing by genetic algorithm. J Mater Eng Perform
29:3334–3352. [Link]
23. Omar SMT, Plucknett KP (2023) Influence of layer thickness upon the dimensional
accuracy and surface roughness of AISI D2 tool steel manufactured using directed
energy deposition BT - Proceedings of the 61st Conference of Metallurgists, COM
2022. Springer International Publishing, Cham, pp 39–42.
24. Amiri M, Crawford GA, Earthman JC (2021) Quantitative percussion diagnostics for
evaluating porosity and surface roughness of cold sprayed and laser deposited
materials. J Mater Res Technol 14:312–323.
[Link]
25. Pereira A, Carou D, Fenollera M, et al. (2022) Experimental study on the manufacturing
of steel inclined walls by directed energy deposition based on dimensional and 3D
surface roughness measurements. Materials (Basel). 15.
26. Lehmann T, Jain A, Jain Y, et al. (2020) Concurrent geometry- and material-based
process identification and optimization for robotic CMT-based wire arc additive
manufacturing. Mater Des 194:108841.
[Link]
27. Bohler 3D PrintAM70, low-alloyed, high strength steel, Material Safety Data Sheet.
28. Rodriguez N, Vázquez L, Huarte I, et al. (2018) Wire and arc additive manufacturing: a
comparison between CMT and TopTIG processes applied to stainless steel. Weld
World 62:1083–1096. [Link]
29. Rodrigues TA, Duarte V, Avila JA, et al. (2019) Wire and arc additive manufacturing of
HSLA steel: Effect of thermal cycles on microstructure and mechanical properties.
Addit Manuf 27:440–450. [Link]
30. Le VT, Mai DS, Doan TK, Paris H (2021) Wire and arc additive manufacturing of 308L
stainless steel components: optimization of processing parameters and material
properties. Eng Sci Technol an Int J 24:1015–1026.
[Link]
31. Hexagon Meterology (2016) HP-L-20.8 Laser scanner for ROMER absolute arm

30
32. Empower operations. [Link] lass accessed on July 27, 2023.
33. Miedzinski M (2017) Materials for additive manufacturing by direct energy
deposition. Chalmers University of Technology.
34. Rahmani Dehaghani M, Tang Y, Wang GG (2022) Iterative uncertainty calibration for
modeling metal additive manufacturing processes using statistical moment-based
metric. J Mech Des 145. [Link]
35. Liu M, Kumar A, Bukkapatnam S, Kuttolamadom M (2021) A review of the anomalies
in directed energy deposition (DED) Processes & Potential Solutions - Part Quality &
Defects. Procedia Manuf 53:507–518.
[Link]

Statements and Declarations

Funding
The authors gratefully acknowledge funding from the Natural Science and Engineering Research

Council (NSERC) of Canada [Grant numbers: RGPIN-2019-06601] and Business Finland under

Project #: 4819/31/2021 with affiliation to the Eureka! SMART project (S0410) titled “TANDEM:

Tools for Adaptive and Intelligent Control of Discrete Manufacturing Processes.”

Competing Interests
The authors have no relevant financial or non-financial interests to disclose.

Author Contributions
All authors contributed to perform and complete this research study. Specifically, definition of

the height-related geometrical parameters and designing and performing the experiments (partially)

were performed by Mostafa Rahmani Dehaghani, Suraj Panicker, and Yifan Tang. The data collection

is performed by Di Wu. G. Gary Wang and Eric Coatanea contributed in development of the height-

related geometrical parameters, modeling, and optimization of the parameters. The first draft of the

manuscript was written by Mostafa Rahmani Dehaghani and all authors commented on previous

versions of the manuscript. All authors read and approved the final manuscript.

31