Implementation of Machine Learning Algorithms for Weld Quality Prediction and Optimization in Resistance Spot Welding

The manufacturing industry constantly aims to improve product quality while improving production speed and lowering production costs. Resistance spot welding (RSW) is widely used in the automotive industry to join thin sheets of coated and uncoated materials. Manufacturers measure weld quality by performing destructive tests like peel and with the help of metallographic examination, which is time-consuming. Further, critical welding parameters need to be optimized to achieve consistent and predictable weld quality. This work addresses the effects of the three critical welding parameters: welding current, welding time, and electrode force on RSW of 1.40-mm-thick DP780 steel sheets. The weld quality indicators studied are nugget diameter (from the peel test), peel strength, tensile shear strength, and the mean dynamic contact resistance. Artificial neural network and adaptive neuro-fuzzy inference system models were used to predict the weld quality indexes, and the prediction accuracy was found to be 99.36 and 99.98%, respectively. A mathematical model was developed using regression analysis to correlate the welding parameters and weld quality indicators. The multi-objective optimization of the welding parameters was done using the genetic algorithm, and its results were validated experimentally. It was found that the welding current had the most significant impact on the weld quality, followed by the electrode force and the welding time.