Principal Component Analysis Based Fault Detection Of Dynamic System

With the ever increasing of scale and complexity in the dynamic system, especially, which is responsible for productive task, it is pressing to produce higher quality products, to reduce product rejection rates, and to satisfy increasingly stringent safety and environmental regulations. Therefore, effective FDD (Fault detection and diagnosis) is of vital importance to the safe operation of dynamic systems. And it will develop new method and research topic for environment friendly project and sustainable development.A fault detection model based on Principal Component Analysis (PCA) was introduced, but the tranditional PCA-based fault detection algorithm did not work well in operating conditions change very frequently system, Its operating conditions change very frequently due to the changes of work points, which always lead to false alarms. We focus on this issue and present a recursive Dynamic PCA (RDPCA) based monitoring scheme for Imperial Smelting Process (ISP) to adapt process changes. A simplified RDPCA algorithm based on first-order perturbation analysis (FOP) was proposed, which is a rank-one update of eigenvalues and their corresponding eigenvectors of an observation covariance matrix. The computation cost is greatly decreased. We also present two new statistics for process monitoring in ISP to avoid numerical computation difficulty induced by the traditional statistics. Finally, we apply the proposed method to real data from ISP. The results show that the proposed scheme can be able to eliminate false alarms and detect faults efficiently.In this work, a fault detection algorithm integrated with PCA and subspace identification method (SIMPCA) is introduced. Unlike SIM based fault detection method, the presented algorithm without explicitly identifying system model and matrices. First, the appropriate data matrix is composed with input and output measurements. PCA is applied to extract the parity subspace, and then two statistics are established similarly with SPE and T2statistics. Through monitoring a real process benchmark model-CSTH, the SIMPCA based fault detection algorithm shows significantly performance compared fault-free condition and fault condition.