A novel, on-line, three-dimensional image analysis control approach for non-destructive testing of spot welds and a discussion of a current off-line testing technique

At present, the quality of spot welds on a production line is monitored by the use of an off-line ultrasonic non-destructive testing technique. This work describes the use of a novel feed back on-line non-destructive control technique that is based on the examination and analysis of the three dimentional shape, features and area of the spot weld. This novel technique also requires the use of a neural network modelling and an intelligent system's analysis. Experimental data, from a design of experiments on the performance of the spot welds and their shapes, have been selected and analysed using the neural network model. The predictions of the neural network model using the test data compare favourably with the experimental data. The correlation coefficient was found to be better than 0.9 with a root mean square (r.m.s.) error of less than 0.065. With most spot-welds, the predictions of the ultrasonic compare unfavourably with the experimental data and those of the neural network. However, the predictions of the neural network give a more reliable diagnosis of the spot weld's mechanical strength. Hence, the on-line method of analysing the shape of a spot weld seems to offer maximum scope for predicting and controlling the quality of the spot weld.