Study On Quality Control In Small Scale Resistance Spot Welding

Small scale resistance spot welding is a joining technique developed in recent ten years. The worksheets to be welded are much thinner, welding is conducted in a more diversified way, no cooling water is provided during the welding process, and a higher precision of welding parameters is demanded. Simply "scaling down" welding parameters from conventional resistance spot welding to small scale is not feasible. Considering the increasing demand for microjoining, study on quality control is of great importance in designing reasonable welding lobe and realizing the automatic and intelligent welding. Signals of dynamic resistance, electrode voltage and welding parameters were used in this study. The experimental design, regression analysis and neural network techniques were applied for weld quality prediction, classification and parameter optimization. The main work and conclusions are as follows:(1) The dynamic resistance curve measured in small scale resistance spot welding was considered first. The weld quality prediction model was put forward based on neural network and features extracted from dynamic resistance curve. Dynamic resistance measurement was proved very easy here. Features were extracted after analyzing the relationship between dynamic resistance and nugget growth. The features having high correlation with weld quality were designed as independent variables in multiple linear regression and neural network analysis. Weld quality was estimated accordingly, and a high prediction accuracy could be realized.(2) In order to realize the real-time output of weld quality, an attempt was made using dynamic resistance and principal components. Principal component analysis was carried out on discrete dynamic resistance directly. Principal components with higher variance contribution were selected as inputs in regression analysis and neural network model. The time needed for data processing was greatly reduced. Results of weld quality prediction and classification showed that the on-line and real-time monitoring could be achieved more easily.(3) Characteristics of electrode voltage variation in small scale resistance spot welding was then considered. The quality monitoring model was proposed utilizing electrode voltage signal and neural network. Features extracted from the voltage curve were used as inputs in the generalized regression neural network model. The on-line and real-time estimation of failure load could be achieved. A discrete Hopfield neural network was also developed for weld quality classification through voltage curve pattern recognition. Weld quality level was accurately classified by both methods. New research way for weld quality control could be recommended by using the electrode voltage curve.(4) Considering a limited variation of quality indicators in small scale resistance spot welding, the parameter optimization approach on multiple quality indicators was conducted. Principal component analysis was first made on quality indicators. Weighted principal components were adopted as the composite quality index. Regression model was established by using welding parameters and weighted principal components as independent and dependent variables respectively. The optimal welding parameters combination was validated through experiments. Weld quality was found significantly improved.