Detection Of Defects Based On The Friction Welding Process Parameters For Neural Network Prediction And Joint Performance

The friction welding-a modern connecting method is used in the various fields more and more widely and so the study of this field is now becoming more and more imperative.In this paper, on the base of analyzing and experimental studying on the process in the forming process of the friction welded joints, BP network is used to construct the neural network prediction system of the major technical parameters (friction-time, friction force and so on) and the capability of the friction welding joints. During the training of the neural network, the influence, which is caused by the various network parameters, on the error function is discussed emphatically and the traditional BP algorithm is improved by importing the re-adjusting coefficient fiand error function coefficient a. The forecast value can meet the actual value well.In addition, using ultrasonic testing machine test the friction welding joints and the modern analyzing method -wavelet packet analysis is used to class the reflected echo scan signal into three sorts: good welding, un-welding and weak defect. After classing, the selected characters of signal by wavelet packet analysis and amplitude-frequency analysis are used to construct the classing-prediction neural network.At last, in order to convenient for practice application, Visual C++ is applied to integrate these application programs into an application-software.