Double pulse gas metal arc welding process parameter optimization and weld performance analysis based on response surface method

Double pulse gas metal arc welding (DP-GMAW) is a low-frequency modulated high-frequency pulse welding, which can fully stir the molten pool, and improve the forming and performance of the weld. In this study, a mathematical model was developed using the response surface method (RSM) with three main process parameters (welding current, peak-to-base ratio, welding speed) as input values with three key geometric characteristics parameters, and the mechanical properties of the resulting welds were investigated. The results showed that the parameter model developed in this study had less error and could optimize the process parameters better; the optimal process parameters (welding current, peak-to-base ratio, welding speed) were 160 A, 39.93%, and 83.11 cm min−1, respectively. The improved welding parameters result in better mechanical properties and better weld profile, and more stable welding process. As a result, this experiment provides a new perspective for process parameter optimization and mechanical properties research for weld manufacturing.


Introduction
Double pulse gas metal arc welding (DP-GMAW) is widely used for low-melting-point metal welding due to its advantages of low heat-input, reduce porosity, high efficiency, and good weld profile [1].The DP-GMAW process exhibits better gap-bridging ability and controlled metal droplet transfer mode than the conventional P-GMAW process, which allows the molten pool to be fully stirred during the welding process, effectively improving the solidification rate of the weld and reduce porosity [2].Among them, different process parameters have different effects on weld quality and morphology in the DP-GMAW process [3].In recent years, researchers have conducted optimization experiments by analyzing geometric features such as the height, penetration and width of the weld bead through different combinations of weld bead parameters [4][5][6].Sandra et al [7] conducted a parametric study of current, voltage, and welding speed during GMAW welding to derive the variation pattern of weld joint organization.Chen et al [8] used GMAW pulsed ultrasound method to improve the microstructure of joint tissue.In conclusion, it is feasible and effective to improve the welding performance through welding parameter optimization.Previously, researchers selected better process parameters through comparative experiments, but comparative experiments cost a lot of labor, material and time.Therefore, researchers began to try more effective methods to guide the key process parameters and weld performance.Nejad et al [9] obtained optimal welding parameters based on artificial neural network and multi-objective optimization algorithm for fatigue life assessment of friction stir welding of aluminum alloys.Yao et al [10] performed a grey correlation analysis of five main process parameters and three key characteristic parameters, and their results provided a new method for evaluating the effect of DP-GMAW process parameters on weld bead and morphology.However, although the above algorithm can successfully optimize the process parameters, the operational difficulty is greater, and there is a lack of research on welding performance.
It is very important to select a suitable algorithm for welding process parameter optimization.Researchers have studied several process parameter optimization methods, each of which has advantages and disadvantages.The orthogonal experimental design is simpler and less workload, but parameter optimization is limited.The polar difference method lacks the means to characterize the complex nonlinear relationship between multiple factors.The response surface method (RSM) combines the advantageous of statistics and mathematics, has high computational efficiency, and can apply functions to explicitly represent the complex functional relationships hidden between the input variables and the response target [11].
The RSM is suitable for solving problems related to nonlinear data processing.It is characterized by combining the advantages of orthogonal tests and regression analysis with simple primary or quadratic polynomial models to fit the factors and response values of the functional relationship of a complex location in a small range.It not only optimizes the process parameters but also are simple and easy to use [12][13][14].Sonar et al [15] used RSM methodology to optimize the welding parameters and solved the problem of reduced strength and corrosion resistance of IN-718 nickel-based alloy in the fusion zone due to segregation phenomena Pavan et al [16] used RSM to develop a mathematical model to optimize the process parameters of A-TIG welding, which improved the heart efficiency along with superior weld performance.Xue et al [17] used the RSM combined with the grey wolf algorithm to optimize the process parameters and improve the weld profile and performance.In summary, optimizing welding process parameters using RSM is already common, but there is still a lack of optimization for DP-GMAW process parameters [18][19][20].
In this study, the DP-GMAW process parameters were optimized using the RSM.In the Design Expert software, three main process parameters (welding current, peak-to-base ratio, and welding speed) that affect weld formation and mechanical properties were used as input values to design a three factor and three level experiment.Then, three key geometric characteristics parameters were used for mathematical modeling, Analysis of Variance (ANOVA) and model validation.Finally, tensile and electrochemical corrosion experiments were carried out.The results show that RSM can effectively reduce the number of tests, optimize weld properties, and provide a reference for DP-GMAW process improvement.It can also serve as a research basis for determining process parameters and managing weld formation.

DP-GMAW process
The current waveform diagram of DP-GMAW process is shown in figure 1, where the current of the pulse set consists of alternating peak current (I P ) and base current (I b ).I ps and I bs are the peak and base currents in strong pulses, respectively, and I pw and I bw are the peak and base currents in weak pulses, respectively, while the peak- to-base ratio is the ratio of I P to I b during the welding process, and the ratio of the two directly affects the weld forming effect [3].I av represents the average current, which is the welding current.

Experimental methods
In this experiment, welding current, welding speed, and peak-to-base ratio were selected as input parameters, A three factor and three level experiment was designed using the Box-Behnken/design-response surface method (BBD-RSM) in the Design Expert software, as shown in tables 1 and 3.The BBD-RSM consists of incomplete group design and 2 k analytic design, which is generally used to fit the RSM of a three-level design, and is also suitable for the analysis of a large number of combination experiments and the study of nonlinear effects of factors.
The DP-GMAW platform consisted of a welding power supply (LORCH S5-RobotMIG), a wire feeder, a sixaxis robot (FANUC M-10iA), shielding gas systems (Ar,99.99%purity) and a control computer, as shown in figure 2(a).A 316 L plate (100 mm × 200 mm × 3 mm) was used as the substrate and 1.2 mm diameter filler wire of 316 L was used, the composition is shown in table 2. The shielding gas (>99.99%Ar) flow rate was 20 l min −1 .The duty ratio was 40% and the double pulse welding frequency was 4 Hz.Before the experiment, polish off the oxide on the surface of the base metal, and then wipe the surface with alcohol.
According to the specific sampling locations of analysis samples is shown in figure 2(b), the components were cut into multiple samples using a wire electric discharge machine (EDM).Then, the cross-sectional specimens were ground, polished, and etched with Keller's reagent.The height, penetration and width of the weld bead were measured using an industrial camera, as shown in figure 3(c), then the values were entered into the corresponding response value series to derive table 3. where the penetration was W , p the height was W h and the width was W .
w The microstructure, chemical composition and fracture morphology were observed with a scanning electron microscope, equipped with energy-dispersive spectroscopy (EDS).Tensile properties were determined using an electronic universal testing machine with a loading rate of 1.0 mm min −1 .The corrosion tests were analyzed by electrochemical workstation, to simulate the real marine environment, a 3.5% NaCl solution was configured at room temperature.

Mathematical model
The RSM uses a polynomial function to approximate the implicit limit state function by experiment and has a high approximation accuracy when the true limit state function is less nonlinear.Therefore, using the RSM for a continuous range of values allows the response variable to be effective in development, improvement, and optimization.A mathematical model can be developed through a correlation regression procedure to investigate the variation of weld width, penetration, and height with process parameters.Since the second-order model includes all the terms in the first-order model.For example, the response surface function of the width of the weld is expressed as follows.
Current is the average current in the welding process, denoted by the symbol A; Peak-base ratio is the ratio of peak current to base current in this experiment, denoted by the symbol B; Speed is the welding speed, denoted by the symbol C.
Response surfaces are used to optimize variables and predict response values by setting functional equations to fit the functional relationship between factors and response values [21].The analogous function can be written as follows, assuming that the experiment contains n independent variables and that each independent variable is continuous, controlled, and measurable.
where ( ) f x is the predicted response value, x , i x j are the examined factors, a , i a , j a , ii a jj are the regression coefficients, and ε is the noise or error term.

Weld bead profile
The weld bead profile is an important measure of the weld quality [22].As shown in table 4, the weld bead profile is different with different parameters.The weld bead L6 and L9 have undercuts, but the overall profile is straighter, more uniform, and the weld has a better profile compared to the other welds obtained in this study.L3 and L4 are not uniform and straight enough; L13 and L14 have more spatter; L5 and L17 have broken welds.Therefore, this study will apply the RSM to optimize the welding process parameters to address the above issues.

Analysis of variance
ANOVA is a statistical model applied to data analysis to analyze and compare the various factors that cause variance changes, to explore the main reasons for the differences in process parameters on the width, penetration, and height of the weld bead, and to determine whether the differences are significant.
The width of the weld is first analyzed by ANOVA, and after inputting the response values into the corresponding experimental design, table 5 is plotted based on the results.Among many models, Linear, 2FI, and Cubic models are selected for ANOVA in Design-Expert software, and the results shows distortion and can not better simulate the real situation of the response surface.While Quadratic model prediction fit coefficient R 2 pred = 0.771 and modified fit coefficient R 2 adj = 0.9304 are closer, and the function match and both values are the maxima of the corresponding series in the recommended model.Therefore, the Quadratic model is used as a function of the width of the weld bead for this experiment.The model is chosen to have a quadratic relationship between the objective function and the independent variables, and the response surface is quadratic.Compared with the simple linear model, its structure is more complex, but its flexibility and accuracy are higher than those of the linear model, and it is more conducive to function significance testing.
The model data for the weld width are derived by performing ANOVA on the functional model and plotted in table 6.The model F value for the weld width is 11.25 and the corresponding value of Prob>F in the model is 0.0002 much less than 0.05, indicating that probability of the model's error due to noise is small, so the model is significant.And the variance of the model measurements R 2 = 0.9696 is close to 1.The closer the R 2 is to 1, the more significant the model is.The signal-to-noise ratio Ap is 17.767 is much greater than 4, indicating that the model fits.The predicted fit coefficient R 2 pred = 0.771 is closer to the modified fit coefficient R 2 adj = 0.9304, the function fits and the model is significant.
Next, ANOVA is performed on the weld penetration, and the above steps are repeated.Among the models for each penetration, the functions match because the predicted fitting coefficient R 2 pred = 0.2475 of the Quadratic model is closer to the modified fitting coefficient R 2 adj = 0.7385, and both are the maximum of the corresponding series in the recommended model.Therefore, the Quadratic model used for penetration in this experiment is more conducive to function significance testing.Similarly, the corresponding value of Prob>F in this model is 0.0136 which is much less than 0.05, indicating that probability of the model's error due to noise is small, so the model is significant.Additionally, the model is noteworthy because the model measurement variance R 2 = 0.8856 is close to 1.The function essentially fits, and the model is important because the anticipated fit coefficients are closer to the adjusted fit coefficients.
Finally, the ANOVA is performed on the height of the weld bead, and in each model, the function matched because the predicted fitting coefficient R 2 pred = 0.6447 of the Quadratic model is closer to the modified fitting coefficient R 2 adj = 0.6659, and both values are the maximum of the corresponding series in the recommended model.Therefore, the Quadratic model is also used for the weld height.The corresponding value of Prob>F in this model is 0.0293 less than 0.05, and the variance of the measured values R 2 = 0.8538 is close to 1, indicating that the model is significant.And the predicted fitting coefficients are closer to the modified fitting coefficients and the functions match, so the model is significant.
In summary, the regression model can be a good simulation of the real surface, and the fit is good.Combined with the arithmetic results of the three Model terms is significant, the Lack of Fit term is not significant, so the welding current, peak-to-base ratio, and welding speed on the weld width, penetration, and height of the impact are significant.Calculated through the Design-Expert software, the final three mathematical models can be expressed a.s.

Model validation Figures 3(a)-(c)
shows the relationship between the actual and predicted values of the width, penetration and height of the weld bead, respectively.The residuals are distributed as points on both sides of the diagonal line, and the closer they are, the closer the actual values are to the predicted values and the smaller the error is, so the model is appropriate.

Effects of process parameters on the weld bead 4.1. Effects of the weld width
It is crucial to study the influence of process factors on the weld width due to the welding process parameters dictate the weld width, which directly affects the properties [23].The interaction of welding current and speed on the width is shown in figure 4(a), where the significance of the center point is 8.07 mm for a current of 140 A, a peak-to-base ratio of 30%, and a welding speed of 100 cm min −1 .Figure 4(b) shows the interactive effect of welding speed and peak-to-base ratio on the weld width when the welding current is 140 A. As shown in illustration, increasing the peak-to-base ratio and reducing the welding speed are conducive to expanding the melt width.When the welding current is 140 A, the welding speed is 80 cm min −1 , the peak-to-base ratio of 40%, and the width reaches a maximum of 10.71 mm.

Effects of the weld penetration
The interactive effect of welding speed and current on weld penetration is shown in figure 5(a), with the central point meaning that the depth of melt is 0.5 mm when the welding current is 140 A, the peak-to-base ratio is 30%, and the welding speed is 100 cm min −1 .Therefore, when the welding current exceeds 140 A, the value of penetration is generally greater than 0.5 mm, and reducing the welding speed and increasing the welding current is beneficial for increasing penetration.Figure 5(b) shows the interaction of welding speed and peak-to-base ratio on penetration, where reducing the welding speed is beneficial to increase the weld penetration.When the welding speed is 100 cm min −1 , the welding current is 160 A, and the peak-to-base ratio is 40%, the maximum weld penetration reaches 1.43 mm.

Effects of the weld height
The height of the weld bead also has a significant influence on weld properties [24].Figure 6(a) shows the interactive effect of welding speed and welding current on the height of the weld bead, the significance of the central point is the height of the weld bead of 1.46 mm at a welding current of 140 A, a peak-to-base ratio of 30% and a welding speed of 100 cm min −1 .When the welding current is between 120-140 A, increasing the welding speed will reduce the weld height.Figure 6(b) shows the interaction between welding speed and peak-to-base ratio on the weld height, increasing the peak-to-base ratio is conducive to reducing the weld height; when the welding speed is 100 cm min −1 , the welding current is 160 A, and the peak-to-base ratio is 30%, the height of the weld achieves a maximum value of 1.78 mm.In summary, the width and penetration of the weld bead increase with the increase in welding current and peak-to-base ratio and decrease with the welding speed increase.The height of the weld bead varies approximately parabolic as the welding current increase, and decreases as the welding speed increases with the peak-to-base ratio.This unveils the law of change of process parameters on the geometric characteristics of the weld.

Optimization of process parameters
In this experiment, the average value of penetration is 0.86 mm, which is less than 50% of the base metal and the average value is small.Related studies have shown that the deeper the weld penetration, the better mechanical properties of the weld [25,26].Therefore, when the weld penetration is small, selecting the mean value of weld width and height and the maximum value of weld penetration for optimization will be more beneficial to enhance the weld properties [27].Other parameters are also selected as mean values to achieve parameter optimization, and the specific guidelines are shown in table 7.  The recommended process parameters after optimization by Design-Expert are shown in table 8.The optimum welding current is 160 A, the peak-to-base ratio is 39.93%, and the welding speed is 83.11 cm min −1 .and the predicted values are 12.73 mm for width, 1.556 mm for penetration, and 1.182 mm for height.
The welds generated by two validation experiments used the improved optimum parameters are A1 and A2.As shown in figure 7, the weld beads are is straight and uniform, with less spatter and a better forming effect, which also indicates that the welding process is more stable.The actual values of weld width are 12.56 mm and 12.43 mm; weld penetration is 1.46 mm and 1.52 mm; and weld height is 1.12 mm and 1.17 mm.The average prediction error of weld width is calculated to be about 1.85%, weld penetration is 4.24%, and weld height is 3.13%; the error is small and the model prediction was relatively accurate.

Microstructure
Figure 8 shows the optical microstructure of the weld bead of L6 and L7 before parameter optimization consists mainly of δ-skeletal-ferrite and γ-austenite.The appearance of L6 is formed better, mainly by δ-skeletal-ferrite, a large number of intra-grain austenite (IGA) organization, and a small number of α cluster crystals.The reason for its formation may be due to more appropriate welding parameters, the welding process is more stable, α cluster-like crystals are the first to precipitate from γ-austenite, thus forming the organizational form of L7, with the heat input continues to accumulate superimposed, the grain organization precipitation, growth, followed by the interconnection of grains into δ-skeletal-ferrite, the formation of L6 internal metallographic morphology   [28].The use of the RSM to optimize the optimal process parameters for welding is more stable, and the resulting weld is straight and uniform; therefore, the ferrite organization of A1 and A2 is transformed from the initial short dendritic to the final reticular structure, and the morphological tissue grains are conducive to enhancing the strength and toughness of the weld [29].
Therefore, the use of the RSM to optimize the process parameters is conducive to enhancing the stability of the weld, so that the internal metallurgical organization of the weld is fully grown to obtain IGA, IGA connected to form the final mesh organization to enhance the strength and toughness of the weld bead.

Tensile test
The mechanical properties of the weld bead are an important criterion to measure whether the process parameters have practical value.L7 and L11 represent a poor forming effect in the experiment, L6 and L9 have better forming results.Therefore, L6, L9, A1, and A2 were selected for the tensile tests as shown in figure 9 and table 9.
The minimum tensile strength of L7 is 562 Mpa and the maximum tensile strength of A2 is 661 Mpa.The tensile strength of L7 is 99 MPa higher than that of A2, indicating that the optimized parameters not only improve the weld formation but also have better mechanical properties.The strength requirement of 316 L material used in the industry is 450 MPa [30], and the tensile strength of the weld bead in this experiment is much higher than this value.The reason for this phenomenon is that the L7 weld is poorly formed, uneven and  uneven shape, and shallow depth of fusion.This structure will cause stress to concentrate at a certain point during the tensile test of the L7 specimen, which is more likely to cause weld fracture.The A2 obtained by parameter optimization is straighter and more uniform, which is conducive to the uniform diffusion of force to the periphery during stretching.The internal metallographic organization of the weld is net-like, which will make the obstruction of dislocation movement increase, so the tensile strength and plasticity of the weld are enhanced.
The tensile fracture morphology is shown in figure 10.The samples of L7 and L11 have a lot of cleavage facets, micro-voids, and tearing ridges, and it is very strong or flexible.Most of the L6 and L9 components are dimples of various sizes and a few micro-voids, indicating that the weld is ductile fracture.Additionally, the tough dimples' weld A1 and A2 fractures are more isometrically uniforms, which improves their strength and plasticity.The reasons for this result are as follows the weld first separates internally to form tiny holes called micro-voids during the tensile process.And then, the tiny holes then grow and collect with other holes due to slip, and when the plastic deformation reaches a certain point, a tough dimple is formed [31].The more uniform the size of the tough dimples, the greater the tensile strength of the weld bead.

Corrosion
E corr and I corr , commonly used in electrochemistry to express the corrosion current density and self-corrosion potential, can be used to gauge how well a material resists corrosion.As can be seen from figure 11(a) and table 10, with the potential negative shift but not see a significant passivation zone in the curve, indicating that the corrosion rate of the weld material is very large, in the anodic polarization phase to generate passivation film dissolution rate is much greater than the repair rate, resulting in the material substrate can be rapidly corroded.From the electrochemical thermodynamic point of view, the smaller the value of E corr , the more easily the weld loses electrons in the electrochemical reaction and suffers from a greater tendency to corrode [32]; The E corr of welds L6, L9, L7, and L11 is larger than A1 and A2, indicating that the parameter-optimized welds are more resistant to corrosion.By the Tafel extrapolation method to calculate the value of I corr , the weld L6, L9, L7, and L11 is also larger than A1and A2, the larger the I corr is the faster the corrosion reaction, so after the optimization of the parameters of the weld A1, A2 corrosion performance is better.
Figure 11(b) shows the impedance spectra are all capacitive arc semicircles, which show capacitive resistance characteristics.After optimization of the parameters, the radius of the capacitance-resistant arcs of welds A1 and A2 became significantly larger, with a consequent increase in charge transfer resistance, causing an increase in    resistance to electrochemical reactions, indicating that the stability of the passivation film of the optimized welds increased [32].Therefore, the welds obtained after optimization of the parameters of this study are more resistant to corrosion.Figure 12 shows the element distribution of the weld corrosion parts.After electrochemical corrosion, the weld surface passivation film has been destroyed, the L6 cross-section appears with more small holes.The L7 corrosion cross-section exists in a larger size, a larger number of holes, indicating a higher degree of corrosion, the internal structure is not stable.However, the corroded cross-section of A1 is smoother and flatter, and the number of holes is smaller and smaller in size.The reason is that the optimization of the process parameters using the RSM in this experiment can make the welding process more stable, which is conducive to the uniform propagation of heat input in the weld and enables the ferrite to be transformed from a short dendritic crystal to a reticulated structure.The more stable reticulated grains are conducive to the generation of a dense passivation film during corrosion, reducing oxidation reactions and enhancing the corrosion performance of the material [33].As can be seen from table 11, the Fe and O contents in welds L6, L9, L7, and L11 show a large increase compared to A1 and A2, while the Ni and Cr contents decrease.The reason is that in the electrochemical experiments Ni, Cr will generate a dense passivation film with O to protect the material from corrosion, and when the Ni, Cr content is less to protect the material is limited, O will occur more oxidation reaction with Fe to generate Fe 2 O 3 or Fe 3 O 4 so that the corrosion surface Fe and O content increased.Therefore, A1 and A2 corrosion performance is better, this result is consistent with the analysis of the polarization curve and Nyquist diagram.In conclusion, employing the RSM to optimize welding parameters is advantageous for improving the resistance to corrosion.

Conclusion
In this experiment, the RSM was used to optimize process parameters the DP-GMAW technology.The experiment takes welding current, peak to base ratio, and welding speed as input values, and takes width,  penetration, and height of the weld bead as response values to establish a mathematical model.In addition, tensile and electrochemical corrosion experiments were conducted on the weld seam, and the following conclusions were drawn.
1.The DP-GMAW process parameter model based on the RSM can better predict the forming a pattern and optimize process parameters of 316 L welds.It has the advantages of small error and the accurate model.
2. The width and penetration of the weld bead increased with the increased in welding current and peak-to-base ratio and decreased with the welding speed increased.The height of weld bead changed parabolically as the welding current increased, and decreased as the welding speed and peak-to-base ratio increase.
3. The optimal process parameters after experimental optimization are: welding current is 160 A, peak-to-base ratio is 39.93%, and welding speed is 83.11 cm min −1 .
4. The optimized optimal parameters could make the welding process more stable, and the resulting weld was uniform and straight.The tensile strength was increased by 99Mpa and the corrosion performance was better.

Figure 3 .
Figure 3. Actual versus predicted values (a) width of the weld bead, (b) penetration of the weld bead and (c) height of the weld bead.

Figure 4 .
Figure 4. Response surface of weld width, (a) effect of welding speed and current, (b) effect of welding speed and peak-to-base ratio.

Figure 5 .
Figure 5. Response surface of weld penetration, (a) effect of welding speed and current, (b) effect of welding speed and peak-to-base ratio.

Figure 6 .
Figure 6.Response surface of weld height, (a) effect of welding speed and current, (b) effect of welding speed and peak-to-base ratio.

Figure 9 .
Figure 9. Stress-strain curve of the weld.

Figure 10 .
Figure 10.Tensile fracture morphology of the weld.

Figure 12 .
Figure 12.Corrosion surface morphology and element distribution.

Table 1 .
Experimental parameters of DP-GMAW process.

Table 2 .
Chemical composition of the substrate and filler wires (wt%).

Table 3 .
Experimental design and responses.

Table 4 .
Weld appearance and cross-section.

Table 5 .
Model recommendations for weld width.

Table 6 .
Model ANOVA for weld width.

Table 7 .
Process parameters optimization rules.

Table 9 .
Mechanical properties of 316 L weld.

Table 11 .
Percentage of elements in the corrosion section (wt%).