Application of experimental data-driven neural network models for assessing the weldabilities of CFRP/DP780 with REW

Abstract

In the present work, the welded joints examined the effects of welding parameters on tensile-shear load and fracture characteristics and also identified the welding phenomena through the dynamic resistance of the welds. The optimal condition achieved the maximum tensile-shear load of 4.6 kN. Under higher current conditions, such as 100 ms and 5.0 kA, the dynamic resistance quickly decreased following peak resistance due to expulsion. Deep learning techniques were used to develop weld quality prediction models based on experimental data. Compared to the shallow neural network and deep neural network models, the convolutional neural network model was found to be the most appropriate for predicting tensile-shear load and failure mode due to its highest coefficient of determination (0.99) and accuracy (99%).