ANN modeling and multiobjective genetic algorithm optimization of pulsed laser welding of Ti6Al4V alloy sheets with various thicknesses

According to the high cost and time-consuming nature of laser welding experiments, repetition of one experiment in a wide range of data is not feasible; so, achieving unexperimented data can be interesting. Hence, the high precision predictability of artificial neural networks (ANN) seems useful. ANN is an intelligent approach to solve different problems. In this study, the experimental data belonging to the pulsed laser welding of two Ti6Al4V sheets, one of them with 1 mm thickness and the other with 1, 1.5, and 3 mm thicknesses, were used to predict the dimensions of the heat-affected zone (HAZ) and the maximum temperature. Moreover, 12 learning methods of a backpropagation network was utilized to select the best one. The Levenberg–Marquardt method had the best performance by considering the mean square error. According to the ANN results, when the laser focus is at the vicinity of workpiece’s surface, the maximum temperature and HAZ width are achieved. It should be also mentioned that increasing thickness and welding speed results in decreasing width of HAZ. By comparing the ANN and experimental results, the maximum relative error for the temperature and HAZ width was obtained equal to 8.62% and 8.22%, respectively. Therefore, ANN can be employed as a tool to develop experimental results and predict indeterminate values in unexperimented ranges with very high precision. Furthermore, in order to optimize the parameters of laser welding, the multiobjective genetic algorithm was used to reduce the HAZ width. The genetic algorithm specified that the HAZ width can be reduced to 0.24 mm by increasing the velocity and thickness.