摘要
Nonlinear methods are used to analyze current signal of spot welding and the minimum embedding dimension, correlation dimension, the optimal time delay and the largest Lyapunav exponent of current signal time series are calculated in this paper. The chaotic character of current signal time series is discovered. Then a chaotic neural network is built and used to predict the future current signal. Means of residual error out of the network are used as eigenvalue of current signal during spot welding. It is shown that spatter can greatly affect the means of residual error of spot welding after analysis, the mean values of output errors of signal contaminated by spatter noise are more than 0. 08, but the mean values of output errors of the signal with no spatter noise are less than 0. 04, so mean of residual errors can be employed as the character of spatter.
Nonlinear methods are used to analyze current signal of spot welding and the minimum embedding dimension, correlation dimension, the optimal time delay and the largest Lyapunav exponent of current signal time series are calculated in this paper. The chaotic character of current signal time series is discovered. Then a chaotic neural network is built and used to predict the future current signal. Means of residual error out of the network are used as eigenvalue of current signal during spot welding. It is shown that spatter can greatly affect the means of residual error of spot welding after analysis, the mean values of output errors of signal contaminated by spatter noise are more than 0. 08, but the mean values of output errors of the signal with no spatter noise are less than 0. 04, so mean of residual errors can be employed as the character of spatter.
基金
National Natural Science Foundation of China(No50575159)
project of Chinese Ministry of Education(No106049,20060056058)
Natural Science Foundation of Tianjin (06YFJMJC03400)
