Fault Detection And Diagnosis In Industrial Process Based On Improved PCA

As the rapid development of technology, the automation level of modern factories is becoming higher and higher, meanwhile the nonlinearity, time varying and uncertainty of the variables in the system are becoming more prominent, which is a great challenge for fault diagnosis technology. Fault diagnosis based on PCA is widely adopted in modern industry, but the defects of PCA theory itself lead to high missing report and false alarm rate.This paper takes PCA as main line, and conducts thorough and deep research on PCA based fault diagnosis, and completes the following job.1. The purpose and importance of fault diagnosis are summarized, and the classification of fault diagnosis are also illustrated in detail.2. PCA theory which is widely adopted in fault diagnosis is introduced in detail The defects of PCA are also stated, which provide direction for further improvement.3. For the assumptions made in PCA deduce conflict with the real data from industrial field, kernel PCA based fault diagnosis is proposed. The application of it in TE process verify its efficiency and superiority.4. For the complexity and time consuming defects existing in kernel PCA based fault diagnosis, Feature Vector Selection(FVS) is combined with kernel PCA to reduce the operation amount and save operation time. The application of the combined FVS-KPCA in TE process clearly indicate its superiority. For the inadequate performance of kernel PCA in fault identification, this paper presents a novel method based on caculating different varables'contribution to SPE and T 2. This method is also verified in TE process.