基于非张量积小波网络的模拟电路故障诊断

Analog circuit fault diagnosis based on non-tensor product wavelet network

为了克服张量积小波神经网络由于张量积小波缺乏自由度和具有很强的方向性等缺陷所引起的不足,提出了用非张量积小波取代张量积小波作为小波神经网络的激励函数构造非张量积小波神经网络的方法。先构造一个非张量积尺度函数,再根据多分辨分析理论得出该尺度函数的非张量积小波函数,把所构造的非张量积尺度函数和小波函数共同作为小波神经网络的激励函数。把该网络应用于模拟电路故障诊断,仿真结果表明,非张量积小波神经网络的效果比相应的张量积小波神经网络要好得多。

In order to solve the defects： that tensor product wavelet lacks degree-of-freedom and strong directionality in tensor product wavelet network, a non-tensor product wavelet network method is proposed, which is constructed using non-tensor product wavelet to replace tensor product wavelet. Firstly, a non-tensor product scaling function is constructed ; then a non-tensor product wavelet function of the scaling function is obtained according to the multireso- lution analysis theory in wavelet space; the non-tensor product scaling function and wavelet function are adopted as the activation functions for the wavelet network. Simulation results indicate that the non-tensor product wavelet network can diagnose the fault in analog circuit better than tensor product Wavelet network.