论文部分内容阅读
针对自动机振动信号短时、非平稳、高冲击的特性,本文提出一种运用固有时间尺度分解(ITD)样本熵和概率神经网络(PNN)进行故障诊断的方法。首先将ITD引入自动机的故障诊断中,通过对ITD分解得到前五层重构信号提取的时频特征来验证ITD方法的有效性,并对信号进行样本熵提取,把其作为特征向量分别用概率神经网络和BP神经网络对自动机进行故障模式识别。实验结果表明:概率神经网络相对于BP神经网络可以提高故障分类的正确率,从而验证了ITD样本熵与PNN的自动机故障诊断方法的优越性。
Aiming at short-time, non-stationary and high-impact vibration signals of automata, this paper presents a method of fault diagnosis using inherent time scale decomposition (ITD) sample entropy and probabilistic neural network (PNN). Firstly, ITD is introduced into the fault diagnosis of automaton. The ITD decomposition is used to obtain the time-frequency features extracted from the first five layers of reconstructed signals to verify the validity of the ITD method. The signal entropy is extracted and used as the eigenvector Probabilistic Neural Network and BP Neural Network for Fault Pattern Recognition of Automata. Experimental results show that probabilistic neural network can improve the accuracy of fault classification compared with BP neural network, and verify the superiority of ITD sample entropy and PNN automaton fault diagnosis method.