Fault Diagnosis Method Based On Statistical Features Extracted And Applied Research

Due to many reasons of system operation condition, large scale system is always bothered by different kind of faults. It is highly required to improve the performance of abnormal monitoring, which can maintain the safe and economy operation of automatic system. Since large scale systems are becoming more complicated and more automatic, it is difficult to establish an accurate physical model that is useful for abnormal monitoring. Using huge amount of data that can be obtained by DCS, this thesis focuses on establishing some new fault detection and diagnosis methods. Some theory research is carried out in the first to provide theory foundation for data-driven abnormal monitoring. The main innovations of this thesis are significant in both theory and application.Involving around multivariate statistical information extraction, some mathematical tools, such as wavelet analysis, space projection theory, spectral decomposition of matrix, correlation analysis etc., are used to develop the research on 3 aspects: multi-scale abnormal detection, knowledge guided data driven fault diagnosis and fault propagation. The main contribution of this paper follows as:1. Using spectral decomposition of a matrix and multi-scale representation of spectral as well as multi-scale transform of a signal, a quasi multi-scale PCA (MSPCA) method is proposed to analyze the reason why multi-scale detection method does well than single scale method. Combing with relative PCA, a quasi MSRPCA algorithm is also developed for abnormal detection.2. Designated component analysis (DCA) is introduced to solve the pattern compounding problem of PCA. A DCA projection frame, which is the theory foundation for DCA based multiple fault diagnosis, is established in the first. For the case when designated pattern are not orthogonal to each other, a progressive DCA analysis method is developed for the generally application of DCA based multiple faults diagnosis.3. A multi-level small faults diagnosis method under the DCA projection frame is established for small fault diagnosis. In addition, an extended DCA algorithm is developed to avoid the shortage that DCA is only validated for faults defined in advance.4. A DCA based fault propagation analysis method is proposed. It is proved in the thesis that correlation between Input/Output designated component(DC) can be used to guide the determination of fault propagation matrix. In addition, a DC regress model is established to predict the imperil level of the root fault.5. Some of the above mentioned methods are used for fault detection and diagnosis of an ocean ship's main diesel engine.