Systematic evaluation of pulsed laser parameters effect on temperature distribution in dissimilar laser welding: A numerical simulation and artificial neural network
Chuan Sun,Mohammad Hossein Razavi Dehkordi,Mohammad Javad Kholoud,Hamidreza Azimy,Z. Li
DOI: https://doi.org/10.1016/j.optlastec.2023.109407
2023-03-30
Abstract:The heat transfer mechanism and temperature distribution in laser welding applications have a great impact on the quality of the weld bead geometry, mechanical properties and the resultant microstructure characterizations of the welding process. In this study, the effects of pulsed laser welding parameters including the frequency and pulse width on the melt velocity field and temperature distribution in dissimilar laser welding of stainless steel 420 (S.S 420) and stainless steel 304 (S.S 304) was investigated. A comprehensive comparison was conducted through the numerical simulation and artificial neural network (ANN). The results of numerical simulation indicated that buoyancy force and Marangoni stress are the most important factors in the formation of the flow of liquid metal. Also, increasing the pulse width from 8 to 12 ms due to increasing the pulse energy, the temperature in the center of the melt pool increased about 250 °C. This leads to increasing the convective heat transfer in the molten pool and heat affected zone (HAZ). The temperature difference at a distance of 1 mm from the beam center at both metals at a frequency of 15 and 20 Hz is bout 58 and 75 °C, respectively. Furthermore, reducing the frequency to 5 Hz, due to diminishment of thermal energy absorption time, has clearly decreased the weld penetration depth in the workpiece. According to the ANN results, increasing both pulse duration and frequency has the significant effect on increasing melting ratio from 0.4 to 0.8 compared to the other input parameters. The ANN results confirmed that under the same input conditions, because of the differences in thermal conductivity coefficient, absorption coefficient and melting point of the two pieces, S.S 304 has experienced higher temperatures about 10% more than S.S 420. Also, among the 13 back propagation learning algorithms, the Bayesian regularization algorithm had the best performance. Among the number of different neurons in the hidden layer, comparison was performed to prevent network overfitting. The maximum relative error of network output data and target data for S.S 304 and S.S 420 temperatures and melting ratio were 7.297, 10.16 and 11.33%, respectively.
optics,physics, applied