Residual Adaptive Algorithm Applied in Intelligent Real-time Calculation of Current RMS Value During Resistance Spot Welding

摘要

To solve the large residual problems, which may occur during feed-forward neural network weight training, a comprehensive residual adaptive algorithm is proposed to give a better stability compared to standard Levenberg-Marquardt (L-M) algorithm and has less computational complexity than classical Newton method. The comparison with standard L-M algorithm checks the better performance of this algorithm. Then the well-trained neural network is embedded into a DSP controller to perform real-time calculation of current RMS value during resistance spot welding. Experimental result shows the validity of the residual adaptive algorithm and the feasibility of an intelligent current measuring method.