Principal Component Analysis In The Application Of The Power Plant Fault Diagnosis

As the increasing demand of the stability and safety of a system, fault detection and diagnosis is becoming more important in state monitoring. In industrial process, it is beneficial to the industrial process’s economics and safety that detecting faults timely, efficiently and precisely. Detecting faults in time can lower maintenance costs and avoiding the broke caused by long term faults. If not detecting the faults timely, it will not just cause economic losses, but even threaten the life safety of the site staff.Principal component analysis (PCA), as a diagnosis method driven by data, needs the data of stable condition to build a PCA model. The limits of statistics and statistics is fixed in traditional PCA. However, it will cause a lot of fault alarms in transient condition using fixed control limits and it will affect the performance of PCA. Meantime, the data acquired in the process contain a lot of noises, it will lower the performance of PCA, too.This essay introduced the basic theory information of fault diagnosis, such as faults classification, fault diagnosis methods and the classification of methods. Also I research the mechanism and method of fault diagnosis by PCA deeply. In view of high rate of false alarms of the principal component analysis in the transient process, the variance adaptive control limit of statistics is studied. However, using the variance adaptive control limit increases the miss alarms of faults, so the studying of the EWMA filter is developed. The variance adaptive control limit of principal component analysis method is used in power plant fault diagnosis first time and reduces the fault alarms of the statistics in principal component analysis in transient process effectively. And the combination of the variance adaptive control limit of principal component analysis method and EWMA filter improves the performance of small fault detection capacity of principal component analysis effectively under the condition of without increasing the false alarm.The main research work is as follows:1Investigating the basis knowledge of fault diagnosis and studying the classification of faults and diagnosis methods. For different kind of fault diagnosis methods, there are two kinds of fault diagnosis methods are introduced in this part:1) fault diagnosis based on model,2) fault diagnosis based on data. And I compare and analyse the advantage and disadvantage of the methods.2The classical principal component analysis is deeply studied in this thesis. To solve the problem that classical PCA will make fault alarms in detecting process, the variance adaptive control limit of statistics and the Exponentially Weighted Moving Average (EWMA) are studied. Also the thesis proposed a fault diagnosis method combining the variance adaptive control limit PCA and EWMA, not only decrease the fault alarm rate, but also improve the performance of detecting the small faults of PCA.3Numerical simulation experiments is taken to verify the detecting performance of combination of variance adaptive control limit PCA and EWMA. And in the end, the proposed method is verified through the power plant flame data.