Real-time Monitor System Based On Data Fusion Technology For Laser Welding

The technology of real-time monitor of laser welding process was extremely necessary for the guaranteeing of welding quality. Introducing the multi-sensor system can solve the problems, such as many targets needed to be detected and much information needed to be got, which were brought by the complexity of laser welding process. In this research, multi-sensor data fusion technology was applied in real-time monitoring system for laser welding process due to the characteristics of laser welding process. The functions of feature extract and data fusion of monitoring system were implemented by virtual instrument technology.At first, the characteristics of detecting targets in laser welding were presented. The ultraviolet, infrared and acoustic emission sensors were used for acquisition of the feature signals in the process of laser welding. Through the optimization of signal processing circuit of multi-sensor system, an extensible, flexible and all-purpose multi-sensor hardware platform was developed. On the basis, a framework of multi-sensor data fusion system with parameters feedback was proposed. These raw signals being acquired by multi-sensor system were sent to computer, in which the features of signals were extracted by algorithm of discrete filter banks, that the objects of studying were transformed to feature space from data space. The problem of data fusion was converted to the problem of pattern recognition. In this research, neural network algorithm was investigated to solve data fusion problem, which well fulfilled the task of system planning. In order to solve the problem that the samples in training data set was lacked in practice application, a novel learning method based on Statistical Learning Theory called Support Vector Machine or SVM was proposed. At last, the different performance between neural network and SVM was discussed by analyzing testing data.The real-time monitoring system for laser welding was installed in one key lab and was running well. The testing results proved that this system was hard to detect the defects which were not full penetration during welding if only one sensor was used in measurement apparatus. The monitor system had been improved in judging the quality of welding after adopting the technique of multi-sensor data fusion. The rate of data classification would improve if the dimensions of feature space increased.This research indicated that the multi-senor data fusion technique can improve stability and validity. The data fusion was a useful method in real-time monitoring for guaranteeing the welding quality.