Research On Fault Diagnosis Of Dynamic Principle Component Analysis

In order to ensure the safety of the production and the people, improve the economic benefit, online monitoring has become an important research direction in the field of process control. A large number of process variables in the industrial field can be measured, handled and monitored because of the rapid development of modern science and technology, we can acquire a lot of implicit information by analyzing the data of these variables. Fault diagnosis method based on data is one method that monitors process condition through analyzing data. And it has been widely studied and applied because it does not rely on mathematical models.In this paper, improved algorithms are set up based on traditional principal component analysis method.The main studying work is shown as follows:Firstly, introduce the fault diagnosis methods that used commonly, and analyzes their application field, advantages and disadvantages.The basic principle and algorithm of Dynamic Principle Component Analysis (DPCA) are introduced in detail.Taking Tennessee Eastman chemical industrial process (TE) as the background, the simulation study is made on typical TE faults by the method of PCA and DPCA.Secondly, combine kernel function and dynamic principle component analysis theory, put forward a fault diagnosis method based on dynamic kernel principle component analysis (DKPCA). The method is fully taken into account the dynamic non-linear characteristics of the actual system, using the ability of kernel function in dealing with non-linear data to construct a dynamic nonlinear fault diagnosis method. In addition, make an improvement to DKPCA by using the feature vector extraction which reduces the heavy computation burden to short the diagnostic time. The simulation results in TE process show that the method has obvious effect in reducing the false positive rate and low false negative rate. Thirdly, a new fault diagnosis system is designed by combining dynamic kernel principal component analysis and Support Vector Machine (SVM). The system makes full use of the advantage of dynamic kernel principal component analysis in fault detection and support vector machine in classification. The simulation shows a good performance of the system in the aspect of fault diagnosis.