*Corr. Author’s Address: Le Quy Don Technical University, 236 Hoang Quoc Viet, Ha Noi, VietNam, trungthanhnguyen@lqdtu.edu.vn 259 Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 Received for review: 2023-12-07 © 2024 The Authors. CC BY 4.0 Int. Licensee: SV-JME Received revised form: 2024-03-28 DOI:10.5545/sv-jme.2023.890 Original Scientific Paper Accepted for publication: 2024-04-11 Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality Van, A.-L. – Nguyen, T.-C. – H.-T. – Dang, X.-B. – Nguyen, T.-T. An-Le Van 1 – Thai-Chung Nguyen 2 – Huu-Toan Bui 2 – Xuan-Ba Dang 3 – Trung-Thanh Nguyen 2,* 1 Nguyen Tat Thanh University, Faculty of Engineering and Technology, Vietnam 2 Le Quy Don Technical University, Faculty of Mechanical Engineering, Vietnam 3 Ho Chi Minh City University of Technology and Education, Vietnam This work optimizes the current (I), voltage (V), flow rate (F), and arc gap (G) of the gas tungsten arc welding (GTAW) of the Ti40A titanium alloy to decrease the heat input (HI) and improve the ultimate tensile strength (TS) and micro-hardness (MH). The radial basis function network (RBFN) was utilized to present performance measures, while weighted principal component analysis (WPCA) and an adaptive non- dominated sorting genetic algorithm II (ANSGA-II) were applied to estimate the weights and generate optimal points. The evaluation via an area-based method of ranking (EAMR) was employed to determine the best solution. The results indicated that the optimal I, V, F, and G are 89 A, 23 V, 20 L/min, and 1.5 mm, respectively. The improvements in the TS and MH were 1.2 % and 19.8 %, respectively, while the HI was saved by 18.4 %. The RBFN models provided acceptable accuracy for prediction purposes. The ANSGA-II provides better optimality than the conventional NSGA-II. The HI, TS, and MH of the practical GTAW Ti40A could be enhanced using optimality. The optimization method could be utilized to deal with optimization problems for not only other GTAW operations but also other machining processes. Keywords: GTAW, Heat input, Ultimate tensile strength, Micro-hardness, radial basis function network Highlights • Optimizing GTAW operation considering the heat input, ultimate tensile strength, and micro-hardness. • A radial basis function network is utilized to propose the GTAW performance measures. • Process parameters, including the current, voltage, flow rate, and arc gap, were optimized. • An adaptive non-dominated sorting genetic algorithm II was proposed and applied. 0 INTRODUCTION Gas tungsten arc welding (GTAW) is a popular welding method that joins metals with a non- consumable electrode. It involves heating the metal to produce a weld pool by creating an arc between the electrode and the specimen. A shielding gas keeps the weld area safe from airborne contaminants. GTAW produces high-quality welds, exact control, excellent weld penetration, minimal spatter, and a variety of metals. Different GTAW operations have been considered and optimized to enhance the performances measured. For the GTAW Inconel 718 alloy, response surface method (RSM) models of the bead width (BW), depth of penetration (DP), heat-affected zone (HAZ), and area of the fusion zone were developed [1]. The authors stated that weld bead properties were mostly impacted by the welding current (I) and torch speed (S), respectively. The BW, Brinell hardness (BH), and micro-hardness (MH) were improved for the welded Inconel 625 using the grey relational analysis and TOPSIS [2]. The results presented that the optimal I, S, and arc gap (G) were 300 A, 90 mm/min, and 5 mm, respectively. In terms of the I, S, and G, the DP and BW models of the titanium joints were proposed [3]. The authors claimed that to create TWB joints free of defects, the ideal parameters of 135 A, 4.1 mm/s, and 3 mm may be used. An aspect ratio of 0.421 and the ideal hardness of 262 HB of the welded Inconel 625 could be generated at I = 300 A, S = 75 mm/ min, and G = 1 mm using the TOPSIS [4]. For the welded 5052 alloys, predictive models were provided in terms of the I, V, and S for the penetration shape factor, MH, and reinforcement form factor of the weld specimens [5]. The optimal I, V, and S reported were 140 A, 18 V, and 300 mm/min, respectively. The BW and DP models of the welded AISI316L were proposed in terms of the I and S, in which the PSO was used to find the best ANN model [6]. The small deviations (less than 4 %) indicated that the developed correlations were efficiently used in the GTAW operation. Wan et al. find that the tensile strength (TS) and elongation (EL) could be improved using the optimized weld shape [7]; the TS and EL of the welded 2219-T8 aluminium alloys were enhanced by 70 % and 4 %, respectively. Vijayakumar et al. [8] presented that the peak current of 50 A, inter-pulse current of 30 A, and inter-pulse frequency of 20 kHz could be used to improve the characteristics of the IP-TIG welded Ti6Al4V alloy. The RSM model of the joint strength was developed for the IN-718 weld by Sonar Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 260 Van, A.-L. – Nguyen, T.-C. – H.-T. – Dang, X.-B. – Nguyen, T.-T. et al. [9], who found that the developed weld exhibited 32 % higher strength and superior corrosion resistance than TIG ones. The RSM models of the maximum yield strength and EL were developed for the IP-TIG welded Alloy 718 joints [10]. The authors stated the yield strength and EL of the IP-TIG joint were 94.5 % and 82.9 % of base metal, respectively. The ANOVA is used to determine the optimal values of the GTAW mild steel plates [11]. The maximum impact strength was obtained at the I of 158.605, a notch angle of 59°, and a single pass, respectively. A convolutional neural network was used to train the BW and DP models [12]. The R 2 value of 0.998 indicated that the developed models could be used to control the quality in real time. The Taguchi and RSM were used to enhance the TS of the welded SS316L stainless steel pipes [13]; the optimal working cycle and peak current were 66.5 % and 114.7 A, respectively. Deep learning was proposed to predict the DP value [14]. The low error level showed that the developed models can be utilized in the GTAW process. Pandya et al. [15] indicated that oxide flux increased the weld penetration and mechanical strength of welded 2205 duplex stainless steel. Moreover, the DP of 6.23 mm, TS of 775 MPa, and MH of 322 HV were obtained using the RSM. Similarly, Baskoro et al. presented that the DP and TS of the welded 304 stainless steel could be increased by 89.9 % and 17.2 %, respectively, with the SiO 2 flux [16]. However, the limitations of related works can be expressed as follows. The mechanical and shape characteristics are frequently taken into account, while the heat input has not been discussed. Reducing the heat input will help the GTAW operate more energy-efficiently. The RSM is widely used to propose performance models, while the application of the ANN has been rare. Moreover, the HI, TS, and MH models have not been developed for the GTAW Ti40A plates. The impacts of GTAW process parameters on the HI, TS, and MH have not been analysed. The optimal GTAW parameters have not been selected to improve the HI, TS, and MH simultaneously. The Taguchi and RSM methods are highly likely to find local outcomes. Therefore, an efficient approach to finding global data is necessary. 1 OPTIMIZING FRAMEWORK The HI is defined as a ratio of energy consumption per length and computed as: HI IV S ii    , (1) where I i , V i , S, and η are the instant current, instant voltage, torch speed, and thermal efficiency, respectively. The TS is computed as: TS TS n i i n    1 , (2) where TS i and n are the tensile strength of the i th sample and the number of samples, respectively. The MH is computed as: MH MH n i i n    1 , (3) where MH i and n are the micro-hardness of the i th location and the number of positions, respectively. Table 1. Process parameters of the GTAW operation Parameters Levels Current, I [A] 70, 90, 110, 130 Voltage, V [V] 22, 23.5, 25, 26 Flow rate, F [L/min] 12, 15, 17, 20 Arc gap, G [mm] 1.5, 2.5, 3.5, 4.5 Fig. 1. The scheme of the GTAW process Fig. 1 illustrates the GTAW process. Table 1 displays the process parameters with their respective levels. The machine’s manufacturer’s recommendations are used to calculate the values of I and V. The F is chosen in accordance with the attributes of the air supplier, whilst the G is cited from relevant sources. The optimization issue is expressed as: Finding X = [I, V, F , and G]. Maximizing TS and MH; Minimizing HI. Constraints: 70 A ≤ I ≤ 130 A; 22 V ≤ V ≤ 26 V; 12 L/min ≤ F ≤ 16 L/min; 1.5 mm ≤ G ≤ 4.5 mm. Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 261 Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality Fig. 2. Optimization approach for the GTAW Fig. 2 shows the optimization framework for the GTAW operation. Step 1: Use the L 32 orthogonal array to conduct GTAW experiments [17]. Step 2: The RBFN models of the outputs are developed regarding GTAW parameters [18]. The RBFN with Gaussian function is used to present the correlations between the inputs and outputs. The input parameters are combined into the hidden one, while the output for the given input (s) and vector (c i ) is expressed as: out sc ii         exp, 1 2 2 2  (4) where || s i – c i || is the Euclidean distance between s and c i . The Gaussian function is expressed as: () exp, rr    2 (5) where γ is a parameter that is found using the cross- validation stage. The RBFN model for a given input s is expressed as: out ww sc i i m i           0 1 2 2 1 2 exp,  (6) where w 0 and w m are the bias and weight, respectively. Step 3: The WPCA is used to compute the weights. The normalized response (n r ) is computed as: n Rm Rm r i i n   () () . 1 (7) The correlative data (S jl ) is computed as: S CovR jRl j jl ii Ri Ri            () ,( ) () (l) ,  (8) where Cov(R i (j) and Ri(l)) denote the covariance of sequences I i (j) and I i (l), respectively. Eigenvalues ( λ k ) and eigenvectors (V ik ) are computed as: () . SJ V km ik   0 (9) The major principal coefficient is computed as follows: PC IiV mm i n ik    1 () . (10) Step 4: The ANSGA-II is used to find solutions. The adaptive crossover probability and adaptive mutation probability are used in the ANSGA-II to identify the best solutions. Fig. 3 illustrates how the ANSGA-II operates. • Producing initial population X(0) and computing the function value f(x) for each individual. • The middle generation X’(t) is created by performing the adaptive crossover and mutation operators. • When an individual’s fitness exceeds the average, they can be passed on to the next generation, which lowers the crossover probability (p c ) and mutation probability (p m ). However, if an individual’s fitness level is lower than the average, they may be removed, which would result in greater p c and p m levels. The crossover and mutation operations’ equations are written as follows: p p pp ff ff p c c cc avg avg c           1 12 1 max ' () , (11) p p pp ff ff p m m mm avg avg m           1 12 1 max ' () . (12) • Creating a new non-dominate set P’(t) by joining the non-dominate solutions. • To create a new population X(t +1), randomly generate new individuals and join them to the non-dominated solutions. Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 262 Van, A.-L. – Nguyen, T.-C. – H.-T. – Dang, X.-B. – Nguyen, T.-T. Step 4: The EAMR is used to select the best optimality. The positive solution is computed as: Svvv v ii ii im    123 ... . (13) The negative solution is computed as: Svvv v ii ii im    123 ... . (14) The performance indicator (I i ) is computed as: I S S i i i    . (15) The best solution is selected with the highest I i . Fig. 3. The operating principle of the ANSGA-II 2 EXPERIMENTAL FACILITIES Using the WEDM method, the specimens are sliced from the Ti40A plates and then cleaned using carbide sheets. The measurements of each specimen are 3 mm in thickness, 64 mm in width, and 80 mm in length. The anti-corrosion cover of the marine ship is mostly made of welded Ti40A plates. Each filler stick has the following measurements: 8 mm for length, 2 mm for breadth, and 3 mm for thickness. The chemical compositions of the Ti40A are shown in Table 2. All testing is conducted using a ZX7-200 welding machine and a designed fixture (Fig. 4). During the processing stage, two welded layers are produced to enhance the junction. Tensile and micro-hardness tests were conducted using the Exceed E45 and Wilson hardness machines, respectively. The instantaneous voltage and current are measured using the clamp meter known as PAC22. Fig. 4. Experiments of the GTAW operation Table 2. Chemical compositions of the Ti40A Element Fe O C N H Other Ti [%] 0.30 0.25 0.08 0.03 0.015 0.05 Balance Figs. 5 and 6 display the representative values obtained at experimental No. 3 and 31, respectively. As shown in Fig. 7, the microstructure of the various welding locations, such as the base material (BM), heat-affected zone (HAZ), and welded zone (WZ), are examined. Fig. 7a displays the BM’s microstructure with large particles and irregular dimensions in comparison to the preceding sections. In the welded zone, the grain size is approximately 170 µm. The uniformly fine grain (about 15 µm) without any faults, such as flash, fractures, voids, and porosity, is produced in the welded region (Fig. 7a). In Fig. 7b, the coarser grain and uneven size of the heat- affected zone (HAZ) are detected without any flaws, including voids and pits. Because of the heat from the welding, the grain size in the HAZ is about 105 µm. 3 RESULTS AND DISCUSSIONS 3.1 Impacts of Process Parameters on Responses The experimental data of the GTAW Ti40A are shown in Table 3. As the I increases from 70 A to 130 A, the HI is increased by 52.6 % (Fig. 8a). An increased I causes Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 263 Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality a higher instant current; hence, more heat input is produced. As V increases from 22 V to 26 V, the HI is increased by 39.2 % (Fig. 8a). An increased V causes a higher arc voltage, leading to higher heat input. The similar impacts of the I and V on the HI were explained in the works of [7], [15], and [16]. As F increases from 12 L/min to 20 L/min, the HI is decreased by 24.3 % (Fig. 8b). A higher F increases the amount of shielding gas, leading to a reduction in the instant current; hence, the HI reduces. As the G increases from 1.5 mm to 4.5 mm, the HI is decreased by 21.3 % (Fig. 8b). A higher G increases the electrode stick out, leading to a higher resistance. The instant current decreases, and the HI decreases. Similar influences can be found in the literature [19]. The TS improves by 18.6 % when the I rises from 70 A to 130 A (Fig. 9a). Due to the poor diffusion between the two plates caused by the low energy input supplied at a low I, a low TS was formed. A greater I results in a better fusion and an improvement in the TS by increasing the energy delivered to the base metal. As V increases from 22 V to 24 V, the TS is increased a) b) Fig. 5. Experimental outcomes at the No. 3; a) tensile strength, and b) micro-hardness a) b) Fig. 6. Experimental outcomes at the No. 31; a) tensile strength, and b) micro-hardness Fig. 7. Analysis of the micro-structure of different zones Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 264 Van, A.-L. – Nguyen, T.-C. – H.-T. – Dang, X.-B. – Nguyen, T.-T. by 9.4 %, while a further V decreases by around 5.3 % in the TS (Fig. 9a). A weak joint is created because a low V causes a low heat input and a poor specimen combination. The heat input rises with a V increase, strengthening the joint. A further V, however, transfers too much energy into the base material. The overheating temperature may lead to a higher grain size; hence, the TS decreases. Similar impacts of the I and V can be found in the works of [7], [13], and [16]. As F increases from 12 L/min to 16 L/min, the TS is increased by 7.5 %, while a further F decreases by around 11.1 % in the TS (Fig. 9b). A higher F may cause a proper welding condition, leading to a lower grain size; hence, the TS decreases. However, an excessive F decreases the heat input transferred into the base material, leading to an improper fusion; hence, a weak joint is produced. As the G increases from 1.5 mm to 4.5 mm, the TS is reduced by 10.1 % (Fig. 9b). Higher G results in an incorrect fusion due to lower input energy transferred to the base material; hence, the TS decreases. The welding gap length and the energy input have an inverse relationship. Consequently, at the lowest G, the TS can be maximized. Similar results were presented in the literature [20]. The MH decreases by 23.3 % when the I rises from 70 A to 130 A (Fig. 10a). A rise in the I causes the energy input at the welding zone to increase, which raises the grain size and causes the MH to drop. The MH decreases by 22.5 % when V rises from 22 V to 24 V (Fig. 10a). A rise in the V results in a higher energy input, which in turn leads the welding area to become coarser; as a result, the MH reduces. Similar results were presented in the literature [5] and [20]. The MH rises by 23.2 % when F increases from 12 L/min to 16 L/min (Fig. 10b). A greater F is linked to lower energy input, which causes solidification to happen more quickly and produces a higher MH. The MH increases by 18.5 % as G increases from 1.5 mm to 4.5 mm (Fig. 10b). Higher heat input at a low G could result in larger grain sizes and a low MH. Lower energy input results from a higher G. The result is a tiny grain size, which raises the MH. Similar outcomes can be found in the work of [20]. 3.2 ANOVA Analysis for Welding Responses ANOVA results for the HI model are shown in Table 4. The I, V, F, G, IF, VG, FG, I 2 , V 2 , F 2 , and G 2 are significant terms for the HI model. The contributions of the I, V, F, and G are 22.32 %, 15.38 %, 13.31 %, and 11.12 %, respectively. The contributions of the IF, UG, and FG are 3.43 %, 14.44 %, and 3.83 %, Table 3. Experimental results of the GTAW operation No. I [A] V [V] F [L/min] G [mm] HI [J/mm] TS [MPa] MH [HV] Experimental data for developing the models 1 70 22 12 1.5 762.19 408.68 254.2 2 90 23.5 12 2.5 780.52 427.87 266.5 3 110 25 12 3.5 857.51 431.53 255.6 4 130 26 12 4.5 980.14 432.28 232.7 5 70 23.5 12 1.5 842.05 436.40 251.1 6 90 22 12 2.5 698.62 402.24 273.6 7 110 26 12 3.5 888.90 416.15 238.2 8 130 25 12 4.5 947.38 449.05 252.9 9 90 25 15 1.5 797.86 468.61 259.7 10 70 26 15 2.5 745.81 424.55 253.9 11 130 22 15 3.5 765.12 457.12 273.3 12 110 23.5 15 4.5 680.93 448.88 301.4 13 90 26 15 1.5 826.23 455.23 245.3 14 70 25 15 2.5 716.91 437.75 267.9 15 130 23.5 15 3.5 848.64 479.46 258.2 16 110 22 15 4.5 596.62 426.35 315.9 17 130 22 17 1.5 693.12 479.76 256.8 18 110 23.5 17 2.5 718.72 468.79 279.5 19 90 25 17 3.5 656.18 438.71 288.9 20 70 26 17 4.5 493.63 400.91 294.9 21 130 23.5 17 1.5 772.97 503.87 244.1 22 110 22 17 2.5 637.94 444.34 291.5 23 90 26 17 3.5 685.98 424.17 271.5 24 70 25 17 4.5 463.31 415.23 311.9 25 110 25 20 1.5 729.35 469.54 274.1 26 130 26 20 2.5 870.87 458.71 219.3 27 70 22 20 3.5 388.56 388.73 354.4 28 90 23.5 20 4.5 493.01 422.25 349.7 29 110 26 20 1.5 756.91 455.38 256.4 30 130 25 20 2.5 841.92 474.27 239.8 31 70 23.5 20 3.5 468.43 414.76 344.1 32 90 22 20 4.5 411.09 398.31 364.2 Experimental data for testing models 33 90 22 20 4.5 411.09 398.31 364.2 34 80 24 13 3 711.62 424.53 272.4 35 100 23 14 2 739.42 455.12 271.8 36 120 24.5 16 4 804.95 467.22 268.6 37 75 24 18 2 649.06 452.62 302.3 38 85 22.5 19 3 524.92 421.51 329.2 39 95 25.5 13 4 741.25 414.83 266.1 40 105 24.5 16 2 769.22 474.26 263.7 41 115 25 18 3.5 779.18 459.98 267.5 42 95 22 19 4.5 450.74 408.80 351.7 43 105 23 20 3 629.07 440.35 316.3 44 95 25.5 15 2 803.78 456.12 252.2 44 120 23.5 18 3.5 748.98 467.86 283.3 45 130 24 16 4.5 851.04 479.81 270.4 46 105 26 17 3 778.37 442.05 251.8 47 90 24 15 3 688.59 444.37 283.4 Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 265 Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality a) b) Fig. 8. The impacts of process parameters on the HI; a) HI versus I and V, and b) HI versus F and G a) b) Fig. 9. The impacts of process parameters on the TS; a) TS versus I and V, and b) TS versus F and G a) b) Fig. 10. The impacts of process parameters on the MH; MH versus I and V, and b) MH versus F and G respectively. The contributions of the I 2 , V 2 , F 2 , and G 2 are 4.51 %, 3.18 %, 3.68 %, and 4.12 %, respectively. As a result, the I is the most effective factor, followed by the V, F, and G, respectively. ANOVA results for the TS model are shown in Table 5. The I, V, F, G, IV, IF, IG, VG, FG, I 2 , V 2 , F 2 , and G 2 are significant terms for the HI model. The contributions of the I, V, F, and G are 22.18 %, 3.79 %, 5.15 %, and 15.16 %, respectively. The contributions of the IV , IF , IG, VG, and FG are 2.01 %, 1.31 %, 3.93 %, 1.45 %, and 2.32 %, respectively. The contributions of the I 2 , V 2 , F 2 , and G 2 are 4.43 %, Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 266 Van, A.-L. – Nguyen, T.-C. – H.-T. – Dang, X.-B. – Nguyen, T.-T. I 2 , V 2 , F 2 , and G 2 are 9.46 %, 4.91 %, 3.82 %, and 4.27 %, respectively. Table 6. ANOVA results for the MH model So SS MS F Value p-value Model 25581.89 1827.28 38.68 < 0.0001 I 32604.92 32604.92 690.20 < 0.0001 V 33895.79 33895.79 717.52 < 0.0001 F 31334.87 31334.87 663.31 < 0.0001 G 25588.41 25588.41 541.67 < 0.0001 IV 6933.23 6933.23 146.77 0.0012 IF 18426.15 18426.15 390.05 < 0.0001 VF 3726.87 3726.87 78.89 0.0024 VG 4955.28 4955.28 104.90 0.0019 FG 3622.77 3622.77 76.69 0.0026 I 2 19696.20 19696.20 416.94 < 0.0001 V 2 10222.87 10222.87 216.40 < 0.0001 F 2 7953.43 7953.43 168.36 < 0.0001 G 2 8890.36 8890.36 188.20 < 0.0001 Res. 661.33 47.24 4407.39 Cor. 26243.22 R 2 0.9748 R 2 = 0.9748; Adj. R 2 = 0.9742; Pre. R 2 = 0.9586 Table 7 presents confirmations of the precision of the HI, TS, and MH models. The small deviations (less than 5 %) indicate the allowable validity of the RBFN correlations. 3.3 Optimal Outcomes Produced by the ANSGA-II The weights of the HI, TS, and MH are 0.36, 0.33, and 0.31, respectively. Fig. 11 shows the Pareto graphs produced by ANSGA-II. Consequently, a low HI causes a reduction in the TS (Fig. 11a), while a larger HI results in an enhanced MH (Fig. 11b). Accordingly, the optimal point with the highest PI i is chosen as the best one (Table 8). The best results produced by the I, V, F, and G are 89 A, 23 V, 20 L/min, and 1.5 mm, respectively (Table 10). Whereas the TS and MH are improved by 1.2 % and 19.8 %, respectively, the HI is down 18.4 %. 3.4 Optimal Outcomes Produced by the NSGA-II To prove the strength of the proposed approach, the conventional NSGA-II is applied to find optimal data. The optimal values of the I, V, F, and G are 81 A, 22 V, 19 L/min, and 1.4 mm, respectively (Table 8). The corresponding values of the HI, TS, and MH are 18.17 %, 14.82 %, and 4.74 %, respectively. As a result, the I is the most effective factor, followed by the G, F, and V, respectively. Table 4. ANOVA results for the HI model So SS MS F Value p-value Model 424549.38 30324.96 41.02 < 0.0001 I 380583.72 380583.72 514.77 < 0.0001 V 262248.10 262248.10 354.71 < 0.0001 F 226952.03 226952.03 306.97 < 0.0001 G 189609.81 189609.81 256.46 < 0.0001 IF 58485.76 58485.76 79.11 0.0072 IG 246219.93 246219.93 333.03 < 0.0001 FG 65306.26 65306.26 88.33 0.007 I 2 76901.10 76901.10 104.01 0.0062 V 2 54222.95 54222.95 73.34 0.0071 F 2 62748.57 62748.57 84.87 0.0071 G 2 70251.12 70251.12 95.02 0.0064 Res. 10350.62 739.33 Cor. 434900.00 R 2 = 0.9762; Adj. R 2 = 0.9612; Pre. R 2 = 0.9536 Table 5. ANOVA results for the TS model So SS MS F Value p-value Model 19509.10 1393.51 49.50505051 < 0.0001 I 19830.89 19830.89 704.47 < 0.0001 V 3388.60 3388.60 120.38 < 0.0001 F 4577.74 4577.74 162.62 < 0.0001 G 13554.39 13554.39 481.51 < 0.0001 IV 1797.12 1797.12 63.84 0.0062 IF 1171.26 1171.26 41.61 0.0094 IG 3513.77 3513.77 124.82 0.0034 VG 1296.43 1296.43 46.05 0.0096 FG 2074.29 2074.29 73.69 0.0058 I 2 3960.81 3960.81 140.70 < 0.0001 V 2 16728.41 16728.41 594.26 < 0.0001 F 2 13250.40 13250.40 470.71 < 0.0001 G 2 4237.98 4237.98 150.55 < 0.0001 Res. 394.08 28.15 3176.16 Cor. 19903.18 R 2 = 0.9802; Adj. R 2 = 0.9724; Pre. R 2 = 0.9602 ANOVA results for the MH model are shown in Table 6. The I, V, F, G, IV, IF, VF, VG, FG, I 2 , V 2 , F 2 , and G 2 are significant terms. The contributions of the I, V, F, and G are 15.66 %, 16.28 %, 15.05 %, and 12.29 %, respectively. The contributions of the IV, IF, VF, VG, and FG are 3.33 %, 8.85 %, 1.79 %, 2.38 %, and 1.74 %, respectively. The contributions of the Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 267 Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality 638.31 J/mm, 450.22 MPa, and 316.8 HV. However, the conventional NSGA-II provides a higher HI and lower TS, as well as MH. In comparison to the traditional NSGA-II, it can be stated that the ANSGA- II produces superior optimal results. 3.5 Novelty and Applications of the Findings The novelty of this work can be expressed as follows. This work proposed an efficient optimizing algorithm entitled ANSGA-II, which could be effectively applied to solve complicated issues and find global results instead of traditional algorithms. The trade-off analysis between the HI, TS, and MH was successfully solved using optimal parameters. Table 7. Comparative data for RBFN models No. HI [J/mm] TS [MPa] MH [HV] Actual Pre. Error Actual Pre. Error Actual Pre. Error 33 411.09 409.56 0.37 398.31 397.36 0.24 364.2 366.7 -0.69 34 711.62 709.62 0.28 424.53 426.56 -0.48 272.4 274.4 -0.73 35 739.42 741.86 -0.33 455.12 453.21 0.42 271.8 269.3 0.92 36 804.95 806.32 -0.17 467.22 465.98 0.27 268.6 266.7 0.71 37 649.06 650.38 -0.20 452.62 454.82 -0.49 302.3 304.6 -0.76 38 524.92 522.94 0.38 421.51 419.36 0.51 329.2 331.2 -0.61 39 741.25 739.62 0.22 414.83 416.85 -0.49 266.1 268.5 -0.90 40 769.22 771.25 -0.26 474.26 476.29 -0.43 263.7 265.4 -0.64 41 779.18 777.36 0.23 459.98 461.32 -0.29 267.5 265.4 0.79 42 450.74 452.23 -0.33 408.81 410.39 -0.39 351.7 349.6 0.60 43 629.07 628.36 0.11 440.35 438.36 0.45 316.3 318.4 -0.66 44 803.78 805.24 -0.18 456.12 458.81 -0.59 252.2 254.3 -0.83 44 748.98 750.66 -0.22 467.86 465.39 0.53 283.3 281.2 0.74 45 851.04 850.36 0.08 479.81 481.25 -0.30 270.4 272.6 -0.81 46 778.37 779.86 -0.19 442.05 446.08 -0.91 251.8 249.6 0.87 47 688.59 690.46 -0.27 444.37 446.52 -0.48 283.4 285.6 -0.78 Table 8. Optimization results produced by the ANSGA-II and NSGA-II Method Optimization parameters Responses I i I [A] V [V] F [L/min] G [mm] HI [J/mm] TS [MPa] MH [HV] Initial values 100 25 15 3.0 764.35 449.47 264.6 ANSGA-II 89 23 20 1.5 623.46 454.77 317.1 0.936 NSGA-II 81 22 19 1.4 638.31 450.22 316.8 0.869 Improvement by ANSGA-II [%] -18.4 +1.2 +19.8 a) b) Fig. 11. Pareto fronts generated by ANSGA- II; a) HI versus TS, and b) HI versus MH Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 268 Van, A.-L. – Nguyen, T.-C. – H.-T. – Dang, X.-B. – Nguyen, T.-T. The optimality can be used to obtain a sustainable GTAW process. The highly accurate models of the HI, TS, and MH were developed using the ANN approach, as compared to the conventional RSM ones. The proposed optimization technique comprising RBFN-ANSGA-II-EAMR can be used to address optimization problems related to various GTAW operations and machining processes. The applications of the findings can be expressed as follows. The findings can be utilized to develop an expert system that will allow the GTAW to operate in many industries. The practical HI, TS, and MH values of the GTAW Ti40A can be predicted using the RBFN models. The optimal data can be utilized to improve quality indicators and energy efficiency of the practical GTAW Ti40A. By leveraging the effects of GTAW inputs on the output, the technological knowledge of the GTAW process can be improved significantly. The range of output objectives may be considered crucial technical advice for welding researchers. 4 CONCLUSIONS The objective of the current study was to select the optimal GTAW inputs (I, V, F, and G) in order to decrease heat input (HI) and increase welding quality (TS and MH). The ANSGA-II was utilized to produce feasible solutions, and the RBFN method was applied to recommend GTAW solutions. The WPCA and EAMR were applied to calculate the weights and select the best optimal results. The conclusions are presented as follows: 1. The highest F and G values were recommended, but to reduce the HI, the low values of I and V were used. The medium values of V and F were addressed, and the highest I and lowest G were utilized to improve the TS. The lowest I and V were utilized, while the highest F and G were used to optimize the MH. 2. The HI, TS, and MH increase from 388.56 J/mm to 980.14 J/mm, 388.73 MPa to 530.87 MPa, and 319.3 HV to 364.2 HV, respectively, for the GTAW parameters considered. 3. The I and V contributed the most to the HI and MH models. The I and G were named as the most effective parameter in the TS model. 4. The I, V , F, and G have optimal data of 89 A, 23 V , 20 L/min, and 1.5 mm, in that order. While the HI was saved by 18.4 %, the TS and MH improved by 1.2 % and 19.8 %, respectively. 5. The Pareto graphs produced by the ANGA-II could be used to select optimal parameters and responses for different GTAW purposes. 6. Compared to the conventional NSGA-II, the developed ANSGA-II might be used to tackle complex problems and produce better results. 7. The ANSGA-II could be utilized to obtain global data instead of conventional algorithms. 8. The designed and fabricated fixture can be utilized in other GTAW operations. 9. Improving HI, TS, and MH are practical benefits to the GTAW Ti40A operation. 10. The impacts of the GTAW factors on air pollution and elongation will be explored in future work. 5 REFERENCES [1] Sonar, T., Balasubramanian, V., Malarvizhi, S., Venkateswaran, T., Sivakumar, D. (2020). Multi-response mathematical modelling, optimization and prediction of weld bead geometry in gas tungsten constricted arc welding (GTCAW) of Inconel 718 alloy sheets for aero-engine components. Multiscale and Multidisciplinary Modeling, Experiments and Design, vol. 3, p. 201-226, DOI:10.1007/s41939-020-00073-3. [2] Sivakumar, J., Vasudevan, M., Korra, N.N. (2020). Systematic welding process parameter optimization in activated tungsten inert gas (A-TIG) welding of Inconel 625. Transactions of the Indian Institute of Metals, vol. 73, p. 555-569, DOI:10.1007/ s12666-020-01876-1. [3] Karpagaraj, A., Rajesh Kumar, N., Thiyaneshwaran, N., Siva Shanmugam, N., Cheepu, M., Sarala, R. (2020). Experimental and numerical studies on gas tungsten arc welding of Ti- 6Al-4V tailor-welded blank. Journal of the Brazilian Society of Mechanical Sciences and Engineering, vol. 42, p. 532, DOI:10.1007/s40430-020-02629-3. [4] Sivakumar, J., Korra, N.N. (2021). Optimization of welding process parameters for activated tungsten inert welding of Inconel 625 Using the technique for order preference by similarity to ideal solution methodology. Arabian Journal for Science and Engineering, vol. 46, p. 7399-7409, DOI:10.1007/s13369-021-05409-w. [5] Omprakasam, S., Marimuthu, K., Raghu, R., Velmurugan, T. (2022). Statistical modelling and optimization of TIG welding process parameters using Taguchi’s method. Strojniški vestnik - Journal of Mechanical Engineering, vol. 68, no. 3, p. 200- 209, DOI:10.5545/sv-jme.2021.7414. [6] Moghaddam, M.A., Kolahan, F. (2022). Modeling and optimization of A-GTAW process using Box-Behnken design and hybrid BPNN-PSO approach. Proceedings of the Institution of Mechanical Engineers, Part E: Journal of Process Mechanical Engineering, vol. 236, no. 3, p. 859-869, DOI:10.1177/09544089211050457. [7] Wan, Z., Meng, D., Zhao, Y., Zhang, D., Wang, Q., Shan, J., Song, J., Wang, G., Wu, A. (2021). Improvement on Strojniški vestnik - Journal of Mechanical Engineering 70(2024)5-6, 259-269 269 Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality the tensile properties of 2219-T8 aluminum alloy TIG welding joint with weld geometry optimization. Journal of Manufacturing Processes, vol. 67, p. 275-285, DOI:10.1016/j. jmapro.2021.04.062. [8] Vijayakumar, V., Sonar, T., Venkatesan, S., Negemiya, A., Ivanov, M. (2024). Influence of IP-TIG welding parameters on weld bead geometry, tensile properties, and microstructure of Ti6Al4V alloy joints. Materials Testing, DOI:10.1515/mt-2023- 0237. [9] Sonar, T., Balasubramanian, V., Malarvizhi, S., Venkateswaran, T., Sivakumar, D. (2021). Maximizing strength and corrosion resistance of InterPulsed TIG welded Superalloy 718 joints by RSM for aerospace applications. CIRP Journal of Manufacturing Science and Technology, vol. 35, p. 474-493, DOI:10.1016/j.cirpj.2021.07.013. [10] Sonar, T., Balasubramanian, V., Malarvizhi, S., Venkateswaran, T., Sivakumar, D. (2020). Development of 3-Dimensional (3D) response surfaces to maximize yield strength and elongation of InterPulsed TIG welded thin high temperature alloy sheets for jet engine applications. CIRP Journal of Manufacturing Science and Technology, vol. 31, p. 628-642, DOI:10.1016/j. cirpj.2020.09.003. [11] Singh, B.K., Chauhan, N., Mishra, A.K., Yadhuvanshi, A.A., Kumar, A., Ansu, A.K., Goyal, A. (2023). Experimental investigation of welding parameters to enhance the impact strength using gas tungsten arc welding. International Journal on Interactive Design and Manufacturing, DOI:10.1007/ s12008-023-01264-1. [12] Baek, D., Moon, H.S., Park, S.H. (2024). Optimization of weld penetration prediction based on weld pool image and deep learning approach in gas tungsten arc welding. International Journal of Advanced Manufacturing Technology, vol. 130, p. 2617-2633, DOI:10.1007/s00170-023-12855-3. [13] Baskoro, A.S., Widyianto, A., Prasetyo, E., Kiswanto, G. (2024). The Taguchi and response surface method for optimizing orbital pipe welding parameters in pulsed current gas tungsten arc welding (PC-GTAW) for SS316L. Transactions of the Indian Institute of Metals, DOI:10.1007/s12666-023-03254-z. [14] Baek, D., Moon, H.S., Park, S.H. (2024). In-process prediction of weld penetration depth using machine learning-based molten pool extraction technique in tungsten arc welding. Journal of Intelligent Manufacturing, vol. 35, p. 129-145, DOI:10.1007/s10845-022-02013-z. [15] Pandya, D., Badgujar, A., Ghetiya, N., Oza, A.D. (2022). Characterization and optimization of duplex stainless steel welded by activated tungsten inert gas welding process. International Journal on Interactive Design and Manufacturing, DOI:10.1007/s12008-022-00977-z. [16] Baskoro, A.S., Amat, M.A., Widyianto, A., Putra, A.D., Aryadhani, S.A. (2024). Investigation of weld geometry, mechanical properties, and metallurgical observations of activated flux tungsten inert gas (A-TIG) welding on 304 austenitic stainless steel. Transactions of the Indian Institute of Metals, vol. 77, p. 897-906, DOI:10.1007/s12666-023-03180-0. [17] Elangandhi, J., Periyagounder, S., Selavaraj, M., Saminatharaja, D. (2023). Mechanical and microstructural properties of B4C/W reinforced copper matrix composite using a friction stir-welding process. Strojniški vestnik - Journal of Mechanical Engineering, vol. 69, no. 9-10, p. 388-400, DOI:10.5545/sv-jme.2023.518. [18] Ibrahim, M.A., Çamur, H., Savaş, M., Sabo, A.K. (2022). Multi-response optimization of the tribological behaviour of PTFE-based composites via Taguchi grey relational analysis. Strojniški vestnik - Journal of Mechanical Engineering, vol. 68, no. 5, p. 359-367, DOI:10.5545/sv-jme.2021.7466. [19] Singh, V., Chandrasekaran, M., Samanta, S., Devarasiddappa, D., Arunachalam, R. (2021). Sustainability assessment of gas metal arc welding process of AISI 201LN using AHP-TLBO integrated optimization methodology. Journal of the Brazilian Society of Mechanical Sciences and Engineering, vol. 43, art. ID 68, DOI:10.1007/s40430-020-02786-5. [20] Bhattacharya, A., Singla, S. (2017). Dissimilar GTAW between AISI 304 and AISI 4340 steel: Multi-response optimization by analytic hierarchy process. Proceedings of the Institution of Mechanical Engineers, Part E: Journal of Process Mechanical Engineering, vol. 231, no. 4, p. 824-835, DOI:10.1177/0954408916641458.