A new criterion for quality monitoring of pulsed laser spot welding using an infrared sensor - Part 2: quality estimation using an artificial neural network

This paper suggests a method for estimating weld quality using a radiation feature. In one experiment the dimensions of the weld joint were examined u sing the radiation feature. Results show that the feature can be used to es timate weld dimensions. In another experiment, weld strength was estimated using the feature. Since it would be laborious to examine a large number of radiation features and find the explicit relationship, an artificial neura l network (ANN) was employed. In experiments, the significant welding param eters were varied within a controllable range and 640 laser spot welds were used for ANN learning. The correlation coefficient between the estimated a nd the measured strength was as high as 0.98 for learned parts. The other 1 80 welds were used to appraise the learned ANN. The correlation coefficient between the estimated and the measured strength was as high as 0.95 for th e unstudied parts and the mean square error of estimation was as low as 0.7 8 kgf.