Study Of Fault Diagnosis And Detection In TE Process Based On Artifical Neural Network

As the rapid development of modern industrial technology, the productive process become continuous, multitudinous and nonlinear with strong coupling, time variable and parameter uncertainty, therefore, the fault occur probability increases, which makes some of the currently fault diagnosis methods can not meet the requirements of process monitoring. It is urgent to develop and improve the current fault diagnosis method and coping strategy to guarantee the stable operation of the productive process.This paper mainly study on complicated industrial fault diagnosis based on artificial neural network, and has completed the following work:1. The current research situation and development trend of all typical fault diagnosis method are summarized. Advantages, disadvantages and applied objects of each method are also analyzed.2. Fault diagnosis method based on BP neural network and probability neural network are studied. The traditional BP algorithm has a low convergence rate, and the target function is easily immerging in partial minimum. This paper introduces momentum factor and variable learning rate to improve the performance of BP algorithm; in probability neural network, the performance of the network is determined by the distribution density SPREAD of the radial basis function. The proper SPREAD value is determined through comparative tests, and the fault diagnosis results is better.3. An improved fault diagnosis method based on PCA and neural network is studied. For the defect that too many input sample will seriously effect diagnosis results, this paper adopts PCA to process the data first to realize dimensionality reduction. Then, the treated data is taken as the input of the network. This method reduces the network model scale and accelerate operation speed. It also reduces the false alarm rate and missing alarm rate, therefore improves the fault diagnosis accuracy of neural network.The effectiveness validation of all algorithms in this paper is conducted in TE process.