Study On The Applications Of Fault Diagnosis Methods In Chemical Process

Chemical process monitoring and fault diagnosis, which is the intensive researchsubject in modern chemical process, assure the stability and security of this productionprocess. Overall characteristic of chemical reaction process were introduced in thepaper, importance and essentiality of the fault diagnosis in chemical process were alsoexplained. Furthermore, particular classifications and illustrations of fault diagnosismethods are given in this paper. Polyacrylonitrile productive process is used here as thespecific research object, according to the polymerization reaction characteristic and theprocess information the monitoring and fault diagnosis of this process are realized herein several diagnosis methods.A neural network is used here as a fault diagnosis model based on the relationshipbetween polymer quality and process variables of the polymerization reaction process.This method is proved available in fault diagnosis and the shortcomings in applicationsare indicated at the same time. Since the data of the process variables can be obtained,the principal component analysis (PCA) is employed to diagnose the fault. Thesimulation results show that this method is available. Unfortunately, PCA is based onthe hypothesis that all the process data follow a Gaussian distribution, which is difficultto be realized in chemical process. While the independent component analysis (ICA) isnot restricted to the above-mentioned condition. Allow for that, ICA method isintroduced to fault diagnosis in this system.In order to correctly estimate the fault detected data, Support Victor Machines isemployed here on the basis of ICA. In the application, reducing controlling limitsproperly can decrease the missing detected rate, and then the fault detected data can beestimated by SVM correctly. The simulation results show that the fault diagnosisapproach based on ICA and SVM improves the diagnosis capability and decreases theunnecessary loses caused by missing detected or fault detected in the factory.