PCA Based Hybrid Fault Diagnosis System Of Chemical Process

In the 21st century, the chemical industry had achieved explosive development with rapidly developing of the science and technology. So the chemical process becomes more and more large and complicated. As the chemical industry has the characteristic of high-risk, its accident would cause a severe disaster. How to ensure the security and stability of chemical process had become a hot-spot for the majority of chemical researchers. As a new technology, fault diagnosis technology is combined with the modern control theory, reliability theory, mathematical statistics, fuzzy theory, information processing, pattern recognition, artificial intelligence theory and so on. It is one of the important techniques to guarantee the safety of chemical process.This research was supported by the national 863 project (Qualitative and Quantitative Study of Complex Fault Diagnosis, NO. 2009AA04Z133). It focuses on the fault diagnosis of chemical process based on the principal component analysis and combined with other methods. Facing the fact of rich data but deficient information of chemical industry, firstly the PCA method was investigated and its advantages and disadvantages were studied by an example. Due to its low recognition ability, PCA was combined with Signed Directed Graph (SDG) and Expert System (ES). So PCA-SDG method and PCA-ES method were proposed to improve PCA method. The two methods were tested separately. Results show that the two methods give a better diagnosis performance. Based on these researches, the mixed fault diagnosis system of chemical process based on PCA was put forward. The system consists of real-time database, PCA-SDG modeling environment, PCA-ES expert knowledge management, mixed fault diagnosis module and so on. The local data collected by real-time database are used for fault detection. If a fault appeared, PCA-SDG and PCA-ES would identify it.This system was applied in a domestic PTA (terephthalic acid) device. The application result shows that it could detect and distinguish fault accurately. Once a fault appeared, the detailed fault reason would feedback to the user. Fault would be found and removed as early as possible. On the next step, this work will do further study on the fault diagnosis method, the fault prediction technology and their mutual supplement to guarantee the safety of chemical production process.