Abstract: A new method is developed to monitoring joint quality based on the information collecting and processing in the spot welding. First, 12 parameters relating to weld quality are mined from electrode displacement signal on the basis of different phase of nugget forming marked by simultaneous dynamic resistance signal. Secondly, through the correlation analysis of the parameters and tensile-shear strength of spot-welded joint taken as evaluating target, different characteristic parameters are reasonably selected. At the same time, linear regression, nonlinear regression and RBF (radial basis function) neural network models are set up to estimating weld quality between the selected parameters and tensile-shear strength. At last, the validity of the proposed models is citified. The results show all of the models can be used to monitoring the joint quality. For RBF neural network model, which is more effective to monitoring weld quality than the others, the average error validated is 2.28% and the maximal error validated is under 10%.

Key words: Correlation analysis, Quality monitoring, Radial basis function (RBF) neural network, Regression analysis, Resistance spot welding