Performance optimization and investigation of metal-cored filler wires for high-strength steel during gas metal arc welding

2.1 Selection of the base metal and consumables for experimentation

Having suitable properties and numerous applications in the fabrication of structural components in petroleum industries, S690QL steel plates have been chosen for the current experimental trials. This steel has a minimum tensile strength of 690 MPa, S represents the structural grade, and L denotes good low-temperature impact toughness. Tables 1 and 2 list the chemical and mechanical properties of the base metal selected for the current research, respectively. Welding bead-on plate trials were performed on an S690QL (EN10025-6) steel plate of 12 mm thickness using the GMAW process. Due to its numerous advantages such as extremely low diffusible hydrogen weld deposit, minimized risk of hydrogen-induced cracking, no re-drying, reduced clean-up time, improved productivity, high deposition rate and efficiencies, good reigniting characteristics, a metal-cored wire “MEGAFIL 742 M” (1.2 mm in diameter) was used for bead-on-plate trials [32,33].

A base metal plate of 1,200 mm × 900 mm × 12 mm dimensions was used and markings of 200 mm length were done for bead-on-plate trials. PipePro 300 was used for all the experimental trials in conventional GMAW mode. Based on the welding machine controls, the voltage (V), arc current (A), and GFR (L·min⁻¹) have been considered the intended input variables in this study. Table 3 shows input variables and their levels for current research.

2.2 Methodology of experiments

To minimize the number of experimental trials and to save time, the design of the experimental approach was adopted. A three-level design matrix for three welding variables was constructed with reference to the response surface methodology of BBD. The objective of the RSM method was to obtain an optimal response by following a particular sequence of designed experimental trials. It was intended to have a mathematical relationship between input welding variables and output responses using the RSM–BBD approach. Since the RSM–BBD is considered suitable for three-level factorial designs, it is used in the current research for developing relationships between input parameters such as the GFR, arc voltage, and current as well as the output responses like the weld BW, HAZ, BH, and DOP. All the experimental runs are given with their levels in Table 4. A preheat of 100–150°C was maintained prior to depositing the weld bead. As shown in Figure 1, a total of 15 beads of 200 mm in length were deposited using a Miller’s PipePro 300 GMAW welding machine keeping all other parameters unchanged as shown in Table 5. Upon cooling to room temperature, each bead on plate trials was cut into 200 mm × 60 mm × 12 mm with a band saw machine. Some of the bead-on-plate trials are shown in Figure 2. Further samples of 15 mm × 60 mm × 12 mm were cut and prepared for macrostructure analysis as shown in Figure 3. After cutting into small samples, each one was polished and etched with 2% Nital followed by macroscopic examination under a metallurgical microscope at 7× magnification. Output parameters such as reinforcement (R), BW (W), DOP, and HAZ were recorded, as shown in Table 6.

Figure 1

Bead on plate trials on S690QL.

Figure 2

Photograph of individual beads on plate trials.

Figure 3

Photograph of samples for macrostructural studies.

2.3 Optimization techniques

In the HTS algorithm (Figure 4), thermal equilibrium behaviour is replicated between surroundings and systems depicting a population-based technique [34]. The algorithm begins by setting the population size, termination criteria, and upper and lower boundaries for the design variable. Then, an initial population is generated randomly. The best solution from this random population is saved as the elite solution [35]. Next, each member of the population undergoes a heat-transfer phase (conduction, convection, or radiation) based on the probability parameter R. This phase updates the solution [36]. However, the updated solution is only accepted if it improves the functional value. The worst solutions in the population are replaced with elite solutions [37]. More details about the updating process for each phase are provided in the following subsection.

The MOHTS algorithm, an enhanced version of the HTS algorithm, is designed to tackle problems that involve conflicting objectives [38]. It can find the optimal solution that balances these conflicting responses. At the start, the MOHTS algorithm creates a set of non-dominated solutions and stores them in an external archive. An e-dominance-based updating approach is employed to evaluate the solutions stored in the archive. By utilizing the non-dominated solutions from the external archive, the algorithm generates Pareto fronts, which represent trade-offs between different objectives. A Pareto optimal plot, also known as a Pareto front or Pareto set, is a graphical representation of the Pareto optimality concept in multi-objective optimization. It depicts the trade-off between different conflicting objectives in a problem, showing the optimal solutions that cannot be improved in one objective without sacrificing performance in another objective. In a multi-objective optimization problem, there are multiple objectives to be optimized simultaneously. These objectives are often conflicting, meaning that improving one objective comes at the expense of worsening another. The Pareto optimal solutions represent the best possible compromise between these conflicting objectives, where no solution can be improved in one objective without worsening another. A Pareto optimal plot displays these optimal solutions in a two- or three-dimensional graph. Each point on the plot represents a solution, with the coordinates corresponding to different objective values. The plot typically shows the non-dominated solutions, meaning that no other solution exists that is better for all objectives. The MOHTS algorithm utilizes a grid-based approach with a fixed archive size for the archiving process [37]. In each generation, the archive is updated using the e-dominance method in conjunction with the HTS algorithm. The number of objective functions in the optimization problem determines the dimensions of the space [39]. By dividing each dimension into boxes of size ɛ, solutions obtained during the optimization process are stored within these boxes. The boxes that are dominated by other solutions are removed, along with the solutions contained within them. If multiple solutions remain in a box, the dominant solutions among them are eliminated. Ultimately, the non-dominated solutions that survive within the boxes are retained in the archive [40].

• Heat transfer by the conduction mode:

  The solutions were restructured in the conduction mode as follows:

  (1) X j , i ′ = X k , i + ( − R 2 X k , i ) if f ( X j ) > f ( X k ) X j , i + ( − R 2 X j , i ) if f ( X j ) < f ( X k ) ; if g ≤ g max / CDF ,

  (2) X j , i ′ = X k , i + ( − r i X k , i ) if f ( X j ) > f ( X k ) X j , i + ( − r i X j , i ) if f ( X j ) < f ( X k ) ; if g > g max / CDF ,

  where X j , i ′ is the updated solution, j = 1,2, …, n; k is a randomly selected solution, j ≠ k and k ∈ (1,2, …, n); i is a randomly selected design variable, i ∈ (1,2, …, m); g _max is the maximum number of generation specified; CDF is the conduction factor; R is the probability variable, R ∈ {0, 0.3333}; and r _i ∈ {0, 1} is a uniformly distributed random number.

• Heat transfer by the convection phase:

  The solutions were updated in the convection phase as per the following equations:

  (3) TCF = abs ( R − r i ) , if g ≤ g max / COF round ( 1 + r i ) , if g > g max / COF,

  (4) X j , i ′ = X j , i + R × ( X s − X ms × TCF ) ,

  where X j , i ′ is the updated solution, j = 1,2, …, n and i = 1,2, …, m. COF is the convection factor; R is the probability variable, R ∈ {0.6666, 1}; r _i ∈ {0, 1} is a uniformly distributed random number; X _s is the temperature of the surrounding; X _ms is the mean temperature of the system; and TCF is the temperature change factor.

• Heat transfer by the radiation phase:

The solutions were updated in the radiation phase as follows:

where, X j , i ′ is the updated solution, j = 1, 2, …, n, i = 1, 2, …, m, j ≠ k, k ∈ (1, 2, …, n), and k is a randomly selected molecule; RDF is the radiation factor; R is the probability variable, R ∈ {0.3333, 0.6666}; and r _i ∈ {0, 1} is a uniformly distributed random number.

Figure 4

Process chart of the HTS algorithm.

3.1 Macrostructures and weld terminologies

Weld macrostructures were investigated at 7× magnification and weld bead geometrical features were measured, as shown in Figure 5. The measurements of the BW (W), reinforcement (R, BH), DOP, and HAZ were recorded for analysis. A macro-structural study reported no evidence of any weld defects and a clear view of weld metal, base metal, fusion line, and HAZ was observed.

Figure 5

Photograph of bead profile measurements.

3.2 Mathematical model

The values of measured output responses are summarized in Table 6 and further analysed using Minitab 17 software. The final regression model expressing the effect of input parameters on output responses like the BW (W), reinforcement (R, BH), DOP, and HAZ were derived as follows:

where x ₁ is the current (A), x ₂ is the voltage (V), and x ₃ is the GFR (L·min⁻¹).

3.3 ANOVA for DOP, reinforcement (R), BW (W), and HAZ of the MCAW process

To verify the accuracy and appropriateness of the regression equations obtained, ANOVA test results were utilized. Minitab 17 was employed to determine the significant and non-significant terms in the models. A confidence level of 5% was chosen to assess the significance. Therefore, a probability value <0.05 indicates a significant impact of the corresponding term on the response variables, namely DOP, BW (W), BH (R), and HAZ.

The ANOVA of DOP, presented in Table 7, utilized a stepwise approach to identify the significant terms contributing to the response. The quadratic model for DOP was found to be statistically significant, encompassing regression, linear, square, and interaction terms. Based on a 95% confidence level, the linear terms (A, V, and GFR), square terms (V × V and GFR × GFR), and interaction terms (A × V and A × GFR) were determined to be statistically significant factors. The lack of fit had a large probability value and was found to be non-significant, indicating that the model for DOP is acceptable and well-fitted. Higher F-values indicated that the current (A) had the greatest impact on the DOP response, followed by V and GFR. An R ² value of 96.81% indicated that the regressions accurately predicted the response values. The model summary in Table 7 demonstrated minimal differences between the R ² values, all of which were close to 1. Consequently, the developed regression model for DOP exhibited accuracy and suitability, as validated by the ANOVA test results.

By following the same approach ANOVA tests for reinforcement (R), BW (W), and HAZ were also carried out (Tables 8–10). Similar results can be drawn from their respective ANOVA test tables, which confirmed that the developed regression models for R, W, and HAZ corresponding to the MCAW process also showed accuracy and suitability over the authentication results of the ANOVA test.

3.4 Main effects plots for DOP, reinforcement (R), BW (W), and HAZ of the MCAW process

The influence of welding variables on the DOP is shown in Figure 6 using the main effects plots. The main effects plots display the average response for each level of the factors, connected by a line. On the plot’s Y-axis, the mean of the means is depicted, representing the range of DOP values obtained for different levels of the input welding parameters. It can be observed from the plot that with increasing current, the DOP increases considerably. This is due to the higher heat input associated with increased current, which results in stronger arc impingement downwards resulting in increased penetration. Higher current results in an increase in heat input into the weld pool. The increased heat causes the temperature of the weld pool to increase, leading to greater melting and fluidity of the base metal. This increased fluidity allows the molten metal to penetrate deeper into the joint, resulting in a greater DOP [16,41]. An increase in voltage increased the DOP. With higher voltage, the size of the weld pool tends to increase. A larger weld pool provides more molten metal available for penetration into the joint. Consequently, the increased volume of molten metal facilitates a greater DOP [42]. An increase in the GFR decreases the DOP because, at higher flow rates, the gas jet may impinge on the weld pool with greater force, altering the dynamics of the molten metal flow. This can result in a shallower penetration as the forceful gas jet tends to push the molten metal away from the weld joint rather than allowing it to penetrate deeper [16].

Figure 6

Main effects plots for DOP corresponding to the MCAW process.

The impact of welding variables on the HAZ is shown in Figure 7 using the main effects plots. It is evident from the graph that an increase in current results in increased HAZ. This is because an increased current results in higher heat input in the base metal. This leads to a larger volume of metal being heated to high temperatures, thereby enlarging the HAZ. With a higher current, more heat is introduced into the base metal, and it takes longer for the metal to cool down after the welding process. Slower cooling rates result in a broader region experiencing elevated temperatures, contributing to a larger HAZ [16]. It was observed that an increase in voltage increases the HAZ. The effect of voltage on the HAZ is more indirect and dependent on the overall welding parameters, such as heat input. Voltage influences heat input and affects the HAZ similar to that of current [16]. A slight decrease in HAZ was observed with increasing GFR. The GFR can influence the cooling rate of the weld and, consequently, the HAZ. Higher GFRs can enhance the cooling of the weld area, which may result in faster solidification and reduced heat transfer to the surrounding base metal. This can lead to a narrower HAZ with a reduced risk of heat-related issues like excessive grain growth, softening, or loss of mechanical properties.

Figure 7

Main effects plots for HAZ corresponding to the MCAW process.

Figure 8 shows the main effects plots depicting the effect of welding variables on the BW (W). Higher current leads to a larger and more fluid weld pool. As the electrode is moved along the joint, the larger weld pool spreads out, resulting in a wider bead formation. With a higher current, more of the electrode material is consumed per unit time, resulting in a greater volume of molten metal being deposited. This contributes to the increased width of the weld bead [43]. Negligible variation in the BW was found with increasing voltage. Voltage is closely related to the length of the welding arc. Higher voltage tends to increase the arc length, which can affect the stability and control of the welding process. However, in terms of BW, the direct influence of voltage is relatively low compared to other factors [44]. It is evident from the plot that an increase in the GFR resulted in a negligible effect on the weld BW.

Figure 8

Main effects plots for the BW W corresponding to the MCAW process.

Figure 9 shows the main effects plots depicting the impact of welding variables on reinforcement (R). With increasing current, reinforcement increases considerably. When the welding current is increased, heat input increases, resulting in a greater volume of the molten metal. The higher energy input allows for greater melting and deposition of the filler material, leading to an increased volume of the weld bead and, consequently, increased reinforcement. As the electrode is moved along the joint, the larger weld pool allows for more filler material to be deposited, resulting in increased reinforcement [45]. An increase in voltage slightly reduces reinforcement. Increased voltage can cause the welding arc to widen. A wider arc distributes the heat over a larger area, resulting in a broader weld bead. The increased width of the arc can lead to a flatter weld bead profile [42]. An increase in GFR slightly increases the reinforcement as can be seen from the main effects plots. The primary role of the shielding gas is to protect the weld area from atmospheric contamination. Increasing the GFR ensures better coverage of the weld zone, which helps prevent the ingress of atmospheric air. Adequate shielding gas coverage helps maintain a stable and controlled arc, resulting in improved weld quality and increased reinforcement [42].

Figure 9

Main effects plots for the reinforcement R corresponding to the MCAW process.

3.5 Residual plots for DOP, reinforcement (R), BW (W), and HAZ of the MCAW process

Residual plots effectively illustrate the robustness and validity of ANOVA test results. The validity and suitability of the ANOVA for the selected model rely on meeting the criteria of the residual plots. In Figure 10, the residual plot for DOP is presented, consisting of four plots. The normal probability plot shows a linear pattern and the absence of residual clustering, indicating a normal distribution of errors. This suggests that the model is appropriate. The residual versus fit plot demonstrates that the fitted values are randomly scattered, indicating a good fit. The histogram plot displays a bell-shaped curve, providing further evidence for a satisfactory ANOVA. The absence of any discernible pattern in the plot of residuals versus observation order confirms the reliability of the ANOVA statistics. Consequently, all four plots validate the ANOVA statistics, enhancing the ability to predict future results. Similar observations were made for the reinforcement (R), BW (W), and HAZ with the MCAW process (Figures 11–13). This indicates that all four plots, for all output responses, successfully validate the ANOVA, providing reliable predictions for future results of these outputs [40].

Figure 10

Residual plots for DOP corresponding to the MCAW process.

Figure 11

Residual plots for HAZ corresponding to the MCAW process.

Figure 12

Residual plots for the reinforcement (R) corresponding to the MCAW process.

Figure 13

Residual plots for the width (W) corresponding to the MCAW process.

3.6 Parametric optimization for the MCAW process

The HTS algorithm was implemented to obtain the desired values of responses by assessing the imperative levels of welding variables. HTS was performed by considering DOP, width (W), and reinforcement (R) as maximization, and HAZ as minimization. The levels of MCAW parameters used during the employment of algorithms are as follows:

Current: 200 A ≤ A ≤ 240 A

Voltage: 23 V ≤ V ≤ 27 V

GFR: 18 lpm ≤ GFR ≤ 22 lpm

3.6.3 Generating Pareto optimal points

In the current study, single-objective optimization was executed exclusively using the HTS algorithm and the results are summarized in Table 11. It can be noted here that when the solitary objective was at its optimum value, the remaining three objectives were outlying from their desired values. For example, when minimization of HAZ optimization was performed, values obtained for DOP, R, and W were 3.4494, 4.2333 mm, and 9.3349 mm, respectively. These values of DOP, R, and W were not at their optimum level. However, they can be approached at their optimum values by increasing the current but with the cost of increasing HAZ. This performance points towards the conflicting effect of input parameters on the output responses, which can be efficiently resolved by finding Pareto optimal points, which are basically trade-offs between the output responses [40].

The MOHTS algorithm can be used for two or more contradictory objective functions simultaneously. It uses a grid-based mechanism to store non-dominated solutions in an external library. Furthermore, the data collection was restructured at every generation during the implementation of the HTS algorithm using the e-dominance method and the process ends with remaining non-dominated solutions only [35].

The MOHTS algorithm was executed to find the non-dominant Pareto optimal points for simultaneous optimization of R, W, DOP, and HAZ. Figure 14 shows the plot of Pareto optimal solutions in three dimensions where the x-axis of the 3D plot represents R, the y-axis represents W, and the z-axis represents DOP. In Figure 15, the x-axis of the 3D plot represents R, the y-axis represents W, and the z-axis represents HAZ. After 10,000 evolution functions, Pareto optimal solutions were obtained. Figure 14 shows a total of 50 feasible Pareto optimal solutions, and Table 13 shows the results of Pareto fronts using MOHTS. It should be eminent that each point on the Pareto curve is a matchless optimal solution with the corresponding input process parameters. The operator has the flexibility to choose any Pareto point along with its corresponding input process parameters according to specific requirements. While various methods exist for generating Pareto optimal points, in the MOHTS algorithm, all the resulting Pareto points are non-dominant solutions, and they can be obtained in a single step, eliminating the need for additional filtering to identify non-dominant points [36].

Figure 14

3D Pareto graph showing R vs W vs DOP for MCAW.

Figure 15

3D Pareto graph showing R vs W vs HAZ for MCAW.

Table 13

Pareto optimal solutions for the MCAW process

Sl. No.	Current (A)	Voltage (V)	GFR (L·min−1)	DOP (mm)	R (mm)	W (mm)	HAZ (mm)
1	240	24	22	2.1752	4.6256	12.258	2.2884
2	240	27	18	3.4494	4.8233	11.8544	2.6134
3	240	27	22	2.8622	5.3873	11.838	2.5974
4	210	23	22	2.2139	4.2333	9.3349	1.5674
5	240	27	20	2.9366	5.0537	11.8462	2.709
6	217	24	21	2.015	4.278644	9.611437	1.9469
7	217	23	18	2.46421	3.653844	9.121237	1.7276
8	210	23	18	2.3667	3.6693	8.7545	1.5834
9	240	26	22	2.5076	5.1336	12.0568	2.4944
10	213	23	18	2.40849	3.659124	8.888477	1.6452
11	237	27	19	3.06971	4.880624	11.40318	2.6517
12	230	27	18	3.1849	4.6941	10.4993	2.4954
13	235	27	18	3.31715	4.7513	11.12853	2.5544
14	240	27	21	2.8446	5.2076	11.8421	2.6791
15	215	23	21	2.10305	4.0396	9.432425	1.7521
16	228	27	22	2.71856	5.239364	10.25827	2.4558
17	233	27	18	3.26425	4.726644	10.86524	2.5308
18	218	26	22	2.31312	4.940704	9.585892	2.1864
19	214	25	22	2.13014	4.698116	9.477668	1.9702
20	235	24	22	2.1623	4.5626	11.53213	2.1964
21	226	24	19	2.26304	4.024796	10.14571	2.1205
22	214	24	18	2.31884	3.895616	9.107668	1.826
23	229	25	18	2.64371	4.190036	10.44545	2.2292
24	233	23	18	2.68709	3.727444	10.67084	2.0572
25	221	24	21	2.0398	4.285236	9.906053	2.0205
26	235	27	19	3.02405	4.8536	11.12443	2.6281
27	240	25	18	2.8658	4.3157	11.9148	2.4074
28	219	23	19	2.25689	3.757056	9.405913	1.8425
29	231	26	18	2.88492	4.456056	10.68701	2.3844
30	226	23	18	2.58958	3.676596	9.871108	1.913
31	227	26	18	2.79164	4.420984	10.23776	2.3284
32	234	27	18	3.2907	4.738676	10.99495	2.5426
33	223	26	22	2.35732	4.959384	9.983157	2.2564
34	224	24	18	2.48944	3.912096	9.844208	2.01
35	222	24	19	2.20928	4.006364	9.773772	2.0469
36	232	26	18	2.90824	4.466304	10.80899	2.3984
37	239	24	19	2.43776	4.150116	11.78169	2.3597
38	217	24	18	2.37002	3.894344	9.288037	1.8812
39	222	27	18	2.9733	4.633364	9.693572	2.401
40	228	26	22	2.40152	4.992864	10.47707	2.3264
41	229	23	18	2.63137	3.694836	10.19065	1.9748
42	236	23	20	2.24504	3.988516	11.36177	2.2146
43	238	26	18	3.04816	4.540224	11.62205	2.4824
44	234	24	19	2.37056	4.090076	11.07515	2.2677
45	233	27	22	2.77841	5.290644	10.84884	2.5148
46	226	23	19	2.32906	3.778896	10.01621	1.9867
47	235	25	19	2.47175	4.352	11.25943	2.4001
48	210	26	18	2.3952	4.3776	9.0185	2.0904
49	214	25	18	2.34086	4.134116	9.195668	1.9862
50	239	26	20	2.56592	4.785016	11.83749	2.592