INTELLIGENT SPOT WELDING QUALITY MONITORING USING ADVANCED SIGNAL PROCESSING TECHNIQUES

Abstract

Resistance Spot welding is one of the most important metal joining techniques nowadays. The research on quality evaluation is one of the most important aspects. Traditional quality testing method is destructive, time-consuming and expensive. Many online non-destructive test schemes have been proposed using neural networks, finite element modeling, etc. However, these online monitoring methods require measurement of many physical parameters and are vulnerable to experiment condition change. In this thesis, only the easily-obtainable electrical signals were used and two schemes were proposed. The first scheme applied windowed feature extraction and SOM network to realize fast and accurate quality classification. The second scheme used a recurrent neural network for feature extraction, a sliding window recurrent neural network for HAZ size estimation and a SOM-type classifier for expulsion checking. The scheme considered more aspects of quality and was able to deal with weld of varying time with satisfying accuracy.