Fault Detection For Multivariable Systems Based On PCA

Fault detection based on principal component analysis (PCA) has received muchattention in the field of multivariate statistical monitoring. In order to achieve the goalof fault detection with complex systems, the multivariate statistical analysis is used toconduct the large amounts of real time data. It is one of the most popular methodsbased on data driven approach, which meets the requirements of the modern complexsystem because it does not rely on mathematical and precise models. Therefore, themethod develops very fast. Now, it is the key technology of ensuring the safety andreliability of the modern industrial system. It has important theoretical and practicalvalue. However, the traditional principal component analysis is only applicable tolinear system. There are many problems remain to be research in nonlinear system. Inthe thesis, The fault detection technology and research progress are given and faultdetection of nonlinear system based on data driven approach is studied. The mainwork and research results of the thesis are as follows:First, the basic principles of PCAand fault design of PCAwith linear system areintroduced, and then the fault detection algorithm based on Gaussian kernel functionwith nonlinear system is put forward. It is an improved method based on traditionalPCA. The proposed approach implements nonlinear transformation by Gaussiankernel function to map the nonlinear input space into linear characterization space.The simulation results show the effectiveness of the proposed method.Second, the fault detection performance of nonlinear system is studied. The faultdetection algorithm based on improved wavelet kernel function is presented. Inaccordance with the Mercer’s theorem, a new kernel function is constructed andproved, namely, the wavelet kernel function. Firstly, the wavelet threshold denoisingis applied to denose sampling data, which improves the accuracy of the model. Then,the preprocessed data is classified by wavelet kernel function to detect the faults. Tocertify the characteristic of the approach, the proposed method is applied to anonlinear simulation system. The simulation results show that the proposed methodimproved the accuracy of fault detection.