基于小波分析与概率神经网络的化工过程故障诊断

A chemical fault diagnosis performed in the light of the probabilistic neural network and wavelets analysis

针对复杂化工过程,提出了一种结合小波分析与概率神经网络(PNN)的故障诊断方法(HWPNN方法),即利用Haar小波分析对过程原始数据进行消噪处理,然后将重构的逼近系数作为输入样本送入概率神经网络完成故障诊断。将HWPNN方法应用于TE过程(一个化工生产过程,由Tennessee Eastman公司控制小组提出)的15种故障进行实验,并与将原始数据直接送入概率神经网络作故障诊断的PNN方法进行了比较,实验结果表明HWPNN方法的故障诊断的准确率明显高于PNN方法。HWPNN方法的诊断准确率达到了100%,是一种可行而有效的化工过程的故障诊断方法。

This paper presents a method to integrate the wavelets analysis and the probabilistic neural network for a fault diagnosis（HWPNN） in the complicated chemical production. The Haar wavelets analysis is used to filter out the noise of the raw data and the low frequency coefficients are taken as the input samples for the probabilistic neural network, which is used to judge the fault type of an input vector. The HWPNN method is applied to the fault diagnosis of the Tennessee Eastman （TE） model . Compared with the PNN method that the input samples are the raw data, the HWPNN method has advantages of fault identifiability and diagnosis accuracy . In the simulation process , fifteen faults of TE model are tested. The diagnosis is up to the accuracy of 100%. The HWPNN method is practical and effective in the fault detection of the chemical processing.