Prediction and fitting of weld morphology of Al alloy-CFRP welding-rivet hybrid bonding joint based on GA-BP neural network

摘要

In the riveting-welding hybrid bonding of aluminum alloy and the Carbon Fiber Reinforced Polymer (CFRP), the welding joint morphology was one of the most important factors for the welding property. In the hybrid bonding process, laser-induced tungsten inert gas hybrid welding technology was used as the welding source, which had a plurality of parameters, such as laser power, arc current, defocused distance and welding speed and etc. The variation of each parameter could directly influence the welding properties. The prediction model of the laser power, welding current, defocused distance and welding speed on the profile of the welding joint morphology was established by using the BP neural network optimized through the genetic algorithm (GA-BP). The contour data of the welding joint was acquired by establishing the polar coordinate system. The geometric morphology of the welding joint was ideally fitted by using the cubic interpolation fitting in MATLAB. The results showed that the fitting line was close to the actual profile of the welding joint, and the prediction accuracy of the GA-BP was favorable. The mean absolute percentage error (MAPE) of each group of data did not exceed 3% while the standard deviation (STD) of that was less than 0.1. The study provided a new method to grope for energy transfer and real-time monitoring of laser induced TIG hybrid welding quality.