Intelligent Monitoring of Laser Welding by Neural Network Model and Multivariate Analysis

An intelligent monitoring device for laser welding in small-size pipes is described. It monitors the reflected laser light and the welding speed, etc, learns the normal criteria and the anomalous criteria of welding, and discriminates between normal and anomalous welding. Using the learned criteria, it categorizes the anomaly as one of several types. As a result of evaluation tests, the degree of correspondence between this device and nn expert is about 90%. This paper describes the new methods, the multivariate analysis for discriminating between normal and anomalous welding, and a neural network model for distinguishing the types of anomalies.