Research Of Small Fault Detection Algorithm Based On Improved Principal Component Analysis

With the development and innovation of science and technology, mankind has experienced the age of handicraft and large industrial machines, the modern industry and gradually moves to the direction of large-scale, integrated and automated. Industrial production processes are often changeable conditions, complex structure and difficult to maintain, these adverse factors to the security and reliability of the system is a big challenge. In order to ensure the continuous and stable operation of industrial system, it is needed to establish an effective mechanism to real-time monitoring the running status of the system.Multivariate statistical process control provides an effective means for fault detection, principal component analysis(PCA) is one of the most common methods used. Accidents are often started from the small fault, if can timely detect these fault occur, providing early warning, it is possible to avoid a series of unnecessary losses. In this paper, the characteristics of small fault, based on the traditional PCA process monitoring algorithms make further improvements, combined with the simulation platform numerical simulation and actual industrial process research to verify the improved algorithm than the traditional PCA algorithm is superior.The main research contents of this paper are as follows:(1) Introduces the basic principle of the traditional PCA algorithm and monitoring method. Linear PCA is extended to nonlinear system, the kernel principal component analysis(KPCA) detailed description of the derivation process method is carried out, and compares the characteristics of the two methods.(2) For industrial process in the common two kinds of small faults, using the moving window strategy, window sampling data combined to cumulative error, makes the fault sample data and the non-fault sample data differentiation more obvious, to achieve amplification of small faults. The method can more timely detect the occurrence of small defects. Finally, it is applied to the numerical simulation and penicillin fermentation process simulation test. The results showed that the amplification of small faults, improved algorithm reduces the false negative rate and improve the detection accuracy.(3) In nonlinear systems, in the calculation of T2 statistics using KPCA method, there are differences of each principal component of the fault sensitivity issue, proposes a KPCA fault detection method based on correlation analysis. This method considers the correlation between measured variables and T2 statistics, the extraction of fault information is more sensitive to the principal component, then establishment of new T2 statistics and control limits. Through the simulation research on the numerical simulation and TE process, shows that the monitoring effect is better than the traditional KPCA method. This algorithm improves the monitoring performance.