Fault Diagnosis Of Tennessee Chemical Process Based On Multivariate Statistical Method

Nowadays, the function of industrial process control system tends to be large-scale and complicated constantly. Consequently, once the control system breaks down, it would lead to a huge economic losses even injury. In order to maintain the system working efficiently and reliably, it is important to develop a fault diagnosis system, which can monitor the process, detectect the fault and judge the source and type of fault in real-time. Multivariate statistical method is an important research branch of fault diagnosis area. It has the characteristic that it does not rely on mathematical models but the process data.In this paper, some fault diagnosis methods using multivariate statistical theory are studied systematically, which based on Tennessee-Eastman (TE) process data.1) Fault diagnosis based on principal component analysis (PC A) and kenel PC A (KPCA) are introduced.2) Fault diagnosis based on independent component analysis (ICA) is introduced. It is integrated with PCA (PCA-ICA) for fault detecting. What’s more, we improve a new fault diagnosis method-continous string matching (CSM) and integrate PCA-ICA and CSM for fault diagnosing.3) The theory of canonical variate analysis (CVA) is introduced, and then, the integration method canonical variate-independent component analysis (CV-ICA) is proposed. TE process is considered as a dynamical process here. Therefore, we apply CVA to solution the canonical variate space and do ICA on the space.The simulation shows that KPCA has a good performance than PCA on fault detective radio and fault reconition, but KPCA is not always a valid method on detecting partial special faults. What’s more, the method that integrates PCA-ICA and CSM has a better performance on fault detecting and reconition efficiency. However, the proposed method CV-ICA has a high fault detective radio on detecting some faults which can’t be detected by some classical methods.