Microstructure analyses and comparison of optimization results for GTAW process parameters through different advanced statistical tools

1. Introduction

Gas Tungsten Arc welding (GTAW) is also known as Helium Arc or tungsten inert gas welding. This process was initially developed in the 1930s BY Russel Meredith [1] to weld magnesium by using inert gas of helium and tungsten as an electrode. This process replaced riveting and helped to joined aircraft components of Al & Mg by welding through this process. Previously there are many refinements, and changes apply to this process. But the HeliArc welding process that was demonstrated by Meredith and developed in the aircraft industry has continued to this day with no change in its fundamentals. Direct current electrode negative with the Tungsten electrode initiate stable arc and can produce an excellent weld. Since from invention number of improvements was made in this process, including a constant current power source, water or gas-cooled torches were developed, and the tungsten electrode was alloyed with some active elements to improve and made stable emissivity [1, 2]. Gases and their blends were also introduced to improve the performance of this process [3].

In this process, the non-consumable electrode of tungsten is used to initiate an arc and to melt the metals. The electrode always held in a torch, which is connected to a gas source to provide gas for shielding. The contact tube is a watered cool copper tube and is used to connect an electrode with the power source. The contact tube also provides cooling to prevent over-heating, whereas the workpiece attaches to a power source with another cable. Shielding gas moves through Torch in a nozzle and directs towards the weld pool area. This kind of shielding is more useful as compared to the shielding in the SMAW process because of two reasons. One is the use of argon or helium (as inert gases), and the second is that gas nozzle directed towards the weld area. Sometimes non-inert gas also used in small quantities, so GTAW looks more appropriate name for this process. For thicker sections, when the filler rod uses, then it can be added manually or automatically into the arc [4].

2. Design methodology

For this work, the orthogonal array method of the Taguchi technique [6] selected with three levels of three welding variables (Current, gas flow rate, and travel speed) and two responses (Ultimate tensile strength and hardness). Then the S/N ratio has computed for each process level analysis. Here a higher S/N ratio of any process parameter level corresponds to the optimum level. In S/N ratio analyses for UTS “larger the better” and for hardness “smaller the better” was selected as a target for calculations. By design of experiment using central composite design (CCD) of response surface analyses used for optimization [7] along with desirability function [8]. ANOVA for variance analyses [9] also performed. Finally, confirmation weld welded to ascertain the optimization results.

3. Experiment set up and plan

The material used in the welding shop for each run of welding was a 6mm thickness stainless steel plate of material ASTM A240 TP 304L with a single bevel angle. This material is selected because this is a common stainless steel material that uses for the fabrication of pressure vessels, heat exchangers, and tanks in the oil and gas field. Chemical composition of this material is given below [10]:

Where response values are measured in actual, whereas mechanical tests performed in Mechanical testing Lab against each run of the experiment. Specimens for tensile testing prepared from weld pieces in a transverse direction to the welds in transverse and then GALDABINI SUN60-V630 machine (where UTS=Max. Load/Original cross-section area) used for tensile testing. Then hardness of weld and HAZ area were measured by using Wilson Hardness 432SVD machine. Average of hardness readings for each piece was used to present hardness value for weld metal. Weld Joints design, geometry, and typical layout for hardness testing locations given below in Figure 3 and GTAW welding equipment and machines used for testing in experimental work shown in Figure 4.

The experimental layout of this study was consist of an L9 orthogonal array. Three important variables with three levels each selected as factors are (A: Current, B: Gas flow Rate and C: Travel speed), and two responses are (Y1: UTS, Y2: hardness), as shown in Tables 1 & 2. Factors are continuous where continuous means those factors which can be assigned some numerical values. The level of process variables and relevant response values shown in Table 3. For analyses, each level of process variable S/N ratio computed where a higher S/N ratio meant better quality characteristics. Therefore the highest S/N ratio corresponds to an optimal value. S/N ratio for UTS and hardness and then based on the different ranking of variables calculated shown in Tables 4 & 5. Formulas used here for calculating responses as larger the better (UTS) and smaller the better (hardness) given in below Eqs. [eq1] & [eq2]. Here n is no. of variables and Y response values.

\[\label{eq1} S/N (\text{Largest the best}) = – 10 * log (\mathrm{\sum} (1/Y2) / n), \tag{1}\]

\[\label{eq2} S/N (\text{Smallest the best}) = – 10 * log (\mathrm{\sum}(Y2) / n). \tag{2}\]

4. S/N ratio analyses

Each factor effect on responses and then their ranking based on the S/N ratio given in Table 4 & 5. For factors ranking (a), the gas flow rate has a major effect along with traveling speed, whereas current has the least impact on Hardness. Because a low gas flow rate causes porosity, and a very high gas flow rate causes to increase brittleness. Then very high/low traveling speed causes a sudden increase and decrease in weld joint temperature, and in HAZ, this causes high hardness, which is not desirable.

Similarly, from Rank (b), we can see current and gas flow rates have a major impact, whereas travel speed has the least. Because current along with proper gas flow rate controls metal transfer rate, penetration, spatters, post-weld cleaning and therefore affects the quality of the weld by influencing metallurgical and mechanical properties. From these analyses, a high value of UTS i-e 564 MPa observed when a current was 150 A and gas flow rate value 12L/min. Similarly, the lower value of hardness 72.66 found when the Gas Flow rate was 12l/min with a travel speed of 95 mm/min.

5. Design of experiment by central composite design

Central Composite Design [11] is a response surface methodology used to design this experiment. This methodology not only fits the quadratic model with sequential experiments but also a useful and accurate model for confounded factor interactions with quadratic effects up to 4th order [9].

The quantitative relation between responses and factors in term of response surface methodology expressed as follow:

\[Y = f (Current,\ \ Gas\ \ flow\ \ rate,\ \ Travel\ \ speed),\] whereas \(Y\) is the responses that are to be optimized, and the other is a function of the controllable factors.

The system behavior obtained through the quadric model, which was developed through the least square method by considering the interaction of factors to maximize or minimize the response variables.

Eq. 3 that derived by Douglas C. Montgomery in statistical quality control and used for the development of a quadratic model.

\[\begin{aligned} \label{eq3} Y =& \beta_{0} + \beta_{1} \text{(Current)} + \beta_{2} \text{(Gas flow rate)} + \beta_{3} \text{(travel speed)} \notag\\ &+ \beta_{11}(Current^{2}) + \beta_{22} (\text{gas flow rate}^{2}) + \beta_{33} (\text{travel speed}^{2}) + \beta_{12} \text{(current*gas flow rate)}\notag\\& +\beta_{13} \text{(current*travel speed)} + \beta_{23} \text{(gas flow rate*travel speed)}. \end{aligned} \tag{3}\]

Here betas are linear, quadratic, and interaction coefficients of input factors. The \(\beta\)\(_{0}\) is the intercept term, whereas \(\beta\)\(_{1}\)/\(\beta\)\(_{2}\)/\(\beta\)\(_{3}\) and \(\beta\)\(_{11}\)/\(\beta\)\(_{22}\)/\(\beta\)\(_{33}\) are the linear terms and interaction between variables terms, respectively. To give flexibility in additional design levels can also be defined by introducing star/axial and center points.

Formula for calculating no of experiments is, N = 2\(^{n\ }\)+ 2*n + n\(_{c}\)

Here N represents total experiments, n No. of factors and n\(_{c\ }\)No. of star points as shown in Figure 5.

It is a rotatable composite design because it provides a constant prediction for all points which are midway between the design center placed in last. The star points in the rotatable center composite design are the distance of each axial point. With three factors and n\(_{c\ }\)=4 star points, 18 experiment runs calculated by using the above formula, and with one replica total of 36 experiment runs are obtained. The DOE summarizes in Table 6, and the detail of the experiment data is given in Table 7 [12].

6. Variance analyses by anova

Analyses of variance (ANOVA) also performed to investigate the influence of controllable factors on measured responses through p-value by using Stat Graphics software [13]. In Hypothesis testing, when the null hypothesis is true P-Value shows probability for a given statistical model to find the significance of results. Before experimenting, first choose a model and threshold value of p, which is called “significance level,” normally 5% or 1%, and denoted as \(\alpha\) (here 5% has selected). Normally 5% or 1% and denoted as \(\alpha\) (here 5% has selected). Very small p-value i-e less than 0.05 shows strong indication against the null hypothesis, whereas a large p-value greater than 0.05 shows weak indication against the null hypothesis. But when P values are close to the cutoff i-e 0.05, then it could be marginal going either way. Degree of freedom tells about independent no of information that went into the calculation of estimate & depends upon the exact design of your test. If Df is one, then it all about means.

Variance analyses for weld hardness and ultimate tensile strength are calculated and given in Tables 8 and 9. For Standard Pareto charts, see Figure 6.

ANOVA Table 8 [13] shows hardness variability for each factor separately and their interactions as well. A P-Value, less than 0.05 for any factor and their interaction, shows significance in these analyses. Therefore based on P values from this table, we can say gas flow rate and traveling speed have a direct momentous effect on hardness. Although P values for the interaction of travel speed and current (AC) are more than 0.05 difference is very less. R-squared value (which shows how much model fitted in variability) is 77.93% for hardness.

Similarly, standard error, which quantifies the precision associated with statistical analyses, is 5.15. Mean absolute error, which is an average of absolute errors, and it quantifies how close the forecast or predictions are. Less value means less error, which is, in this case, is 2.76. Below, the Pareto chart drawn for hardness in Figure 6 shows the same.

R-sq. = 96.4706 %, Std. An error of Est. = 10.3591, Mean absolute error = 5.20509

ANOVA Table 9 [13] shows the ultimate tensile strength variability for each factor separately and their interactions as well. All P values of less than 0.05 can see in Table 9. From this table, we can say current has a significant effect on tensile strength along with combination (AB) with a gas flow rate. Because current also governs arc initiation, and its instability and a big change in current can also affect weld strength and weld quality by metallurgical effects as well. R-squared value (as explained in hardness analyses under Table 8) shows the model is fitted 96.4706% of variability for tensile strength. Similarly, the standard error is 10.3591, and the mean absolute error is 5.20509.

Based on the above variance analyses, standard Pareto charts draw for hardness and UTS [13]. Absolute values of the standardized effects can see from Pareto charts. The reference line depends upon the significant level and indicates the effects of significance.

The X-Axis is unitless in the graph and shows a standardized effect of affecting factors. Cut off p-value has selected 2.5 (\(\alpha\)=5%) on a scale, and bars drawn for major affecting factors and their combinations along Y-Axis. Two colors with sign +/- shows factor’s increasing/decreasing effects.

From charts, we see the gas flow rate along with travel speed has the main effect, and similarly, current, and Gas flow rates have the main effects on UTS.

7. Desirability function

For optimization, desirability and loss functions are two approaches that normally use. Desirability function takes precedence over loss function due to its flexibility and applications. A desirability function approach is very useful for multi-response processes optimization. Here the idea is that quality of process having multiple characteristics with any one of them is outside the desired limit is unacceptable [8]. This method helps to find those operating conditions which provide desirable response values. Depending upon whether a maximum, minimum, or target value assigned to a particular response, different desirability functions are available to be used. For this work, larger the best (LTB) selected for UTS, and smaller the best (STB) for hardness, and desirability for each response calculated by using Equations given below. Where d\(_{i}\)\(_{\ }\)is desirability, y\(_{\ }\)is response value, Y\(_{max\ }\)is upper/maximum limit, and Y\(_{min\ }\)is the lower/minimum limit. Desirability ranges from zero to one, and the highest value will be optimal. Generally, 0.7 and above is considered satisfactory.

Individual desirability for each response obtained using equations 4 and 5 developed by Derringer and Suich for larger the best and smaller the best. \[\label{GrindEQ__4_} d_i\mathrm{=}\left\{ \begin{array}{cc} 0, & \mathop{y}^{\mathrm{\textrm{ˆ}}}\mathrm{\le }Y_{min} \\ \left(\frac{\mathop{y}^{\mathrm{\textrm{ˆ}}}\mathrm{-}{\mathrm{Y}}_{min}}{{\mathrm{Y}}_{max}\mathrm{-}{\mathrm{Y}}_{min}}\right), & {\mathrm{Y}}_{min}\mathrm{\le }\mathop{Y}^{\mathrm{\textrm{ˆ}}}\mathrm{\le }{\mathrm{Y}}_{max},r\mathrm{\ge }\mathrm{0} \\ \mathrm{1,} & \mathop{y}^{\mathrm{\textrm{ˆ}}}\mathrm{\ge }{\mathrm{Y}}_{min} \end{array} \right. \tag{4}\] \[\label{GrindEQ__5_} d_i\mathrm{=}\left\{ \begin{array}{cc} \mathrm{1,} & \mathop{y}^{\mathrm{\textrm{ˆ}}}\mathrm{\le }Y_{min} \\ \left(\frac{\mathop{y}^{\mathrm{\textrm{ˆ}}}\mathrm{-}{\mathrm{Y}}_{max}}{{\mathrm{Y}}_{min}\mathrm{-}{\mathrm{Y}}_{max}}\right), & {\mathrm{Y}}_{min}\mathrm{\le }\mathop{Y}^{\mathrm{\textrm{ˆ}}}\mathrm{\le }{\mathrm{Y}}_{max},r\mathrm{\ge }\mathrm{0} \\ 0, & \mathop{y}^{\mathrm{\textrm{ˆ}}}\mathrm{\ge }{\mathrm{Y}}_{min} \end{array} \right. \tag{5}\]

By using these equations, individual desirability of each response was calculated for each run and given in Table 7.

D is overall desirability and calculated by the geometric mean of all individual desirabilities.

\[\label{GrindEQ__6_} D = \{ d_1(Y_1).d_2(Y_2).d_3(Y_3).\cdots. d_k(Y_k) \}^{\frac{1}{k}}. \tag{6}\]

After performing this, the optimal values obtained using the above equation for each response are given in Table 10.

Here, D is overall desirability of each response, and D* shows overall process desirability and geometric means of both overall desirabilities D.

Overall desirability, D*= 0.82. Optimized desirability and response values are the base that used to get the optimized settings for factors and given in Table 11, whose graphical presentation shown in Figure 7 [13].

The graphical presentation of desirability plots for factors shown above showing optimal values of current, gas flow rate and travel speed, as mentioned in Table 10. Desirability (D) for both responses and overall process (D*) found about 0.8, as mentioned in Table 9 and shown as yellow/orange area in the top right portion as per the color contrast range of Figure 7.

8. Validation of gtaw optimization results

After performing optimization to get optimal values as an output, then the next stage was to validate the results. Optimal factors values obtained were used to run the final weld. The same material used for welding under the same circumstances and welding conditions. Final results, including UTS and hardness values, micro and macrostructure, were found very close to the optimal responses obtained through the optimization process, as given in Table 12. Final confirmation welding and samples which have taken for testing shown in Figure 8.

Instron-model#3367 UTS machine was used to perform tensile testing. The actual response value of UTS found 561.78 MPa as compared to an optimal value obtained from desirability analyses was 562.891 MPa. Tensile test results, TS machine, broken sample, and graph shown in Figure 9.

Vicker hardness tester of Buehler USA, model Micromet-3 advance, was used for hardness testing. Actual hardness values observed 72\(\mathrm{\sim}\)74HRB, whereas the optimal value was 71.365 HRB. The hardness tester model and sample used shown in Figure 10.

Three samples were taken for the GTAW process to find base metal, HAZ, and weld metal hardness. In final results, only weld metal hardness included where the average value of weld metal hardness readings was considered i-e 73.048 HRB.

Ametek brand spectrometer was used to find the chemical composition of weld metal and found as follows, as shown in Figure 12.

The microstructure of final weld, HAZ, and base metal observed by using an optical microscope of brand Zeiss, model-Axioplan 2 Imaging. The microscope used and microstructures observed are shown in Figure 13.

Because of little or no heat effect on base metal other than of weld metal and HAZ area, a fine austenitic grain crystalline structure observed for solution annealed stainless steel base material. In HAZ, irregular and broken austenitic grains were observed due to temperature changes during welding and then suddenly cooling in air. In the weld metal area, dendritic austenitic grains observed due to the melting of base metal and the addition of filler with subsequent cooling to form a new structure that changed the weld metal composition and its microstructure.

To check the quality of the Macro weld examination also performed. Here for examination, one cross-section of the weldment was polished and then etched by using the etchant of Aqua Rega (HNO3 + HCL in the ratio of 1:3, respectively). Macro test performed to find any possible imperfections like porosity, lack of penetration, and sidewall fusion and weld profile, mainly hot weld cracks or sensitization due to the formation of carbides precipitates. Macrograph has shown in Figure 14, where examination performed at less than 20x magnification.

Now final optimal results obtained from Taguchi analyses, desirability analyses, and obtained from actual testing are summarized in Table 12. From this table, we can say results are very close to each other by using both techniques. This minor difference in results (optimal factors values & and then the subsequent effect on response values) is obviously due to the different approaches and levels deployed for experiments by each statistical technique.

9. Conclusion

Conclusion of this research work can be summarized as follow:

1. The study highlights the important role of standards and welding best practices deployed in the contemporary welding world. And describes the GTAW process and various controlling variables as well as Standards compliant tolerance zone on these variables acceptable window of Welding Process parameters selection, which give good quality welds.

2. The study provides a framework of exploring the best combination of conditions that provide the Highest Quality (or Enhanced Quality), Weld, by optimizing within this standard based acceptable zone of welding conditions. This constrained optimization is essentially multi-criteria optimization. The statistical tools used for this purpose in Studying the GTAW process are; the Design of Experiments (DOE) through Signal over Noise i-e S/N ratio and Desirability Function (DF) analyses to further narrowing down the optimum region for process parameters settings within welding standard-based recommended search space.

3. The above framework of the analyses can be applied for any such welding case. It can be combined in a single software package to find optimal weld conditions leading to enhanced weld quality.