Research On Fault Diagnosis Method Of Dynamic Industrial Process

The common process data are dynamic in nature, which can be caused by the complicated process mechanism, random errors and process disturbances, the existence of inertial link and energy storage link, the extensive adoption of feed-back control and the requirement of real-time sampling, er al. However, it is difficult to build a process model for disturbance detection. So, the studies on dynamic fault diagnosis methods in this thesis is multivariate statistical process control (MSPC) field mainly based on data driven method. MSPC method is a kind of fault diagnosis method including off-line modeling analysis of historical data and the online monitoring of real-time data, and can effective evaluate the running state of industrial process and fault detection. In this thesis, the main contents are as follows:(1) Summarize the causes and effects of Industrial process data dynamic in detail; simple introduce the process of fault diagnosis and classification of disturbance diagnosis methods; describe the study status of the fault diagnosis methods on dynamic data and summarize the several aspects of dynamic data fault diagnosis methods improvement.(2) Mainly introduce dynamic principal component analysis (DPCA) and dynamic independent component analysis method (DICA), the un-quality process monitoring methods which are widely used. Firstly, introduce the disturbance detection method based on DPCA; secondly, because the DPCA can not deal the data with non-gaussian distribution, introduce the DICA fault detection method in detail; at last, state that DICA has the advantage in detecting small fault which is hard to detect for DPCA through the TE simulation.(3) Proposes a dynamic fault diagnosis method, the combination of subspace identification and dynamic independent component analysis (SI-DICA). Firstly, prove the correlation of latent variables of the DICA in theory and the simulation proving that the artificial dynamic characteristics will cause the phenomenon of continuous false alarm; secondly, introduce the subspace identification methods, canonical variable analysis (CVA); at last, proposes the SI-DICA dynamic fault diagnosis method, the combination of subspace identification and dynamic independent component analysis, and through simulation research, the effectiveness of the method, avoiding the phenomenon of false alarm continuously, reducing the false alarm rate, further improving the fault detection rate and reduce the detection delay at the same time is proved.(4) Make an imagination on the improvement in fault diagnosis method proposed in this thesis, and summarize the possible future development trend on the fault diagnosis method in the future.