Comparative Study Of Fault Diagnosis Methods For TE Process

Chemical processes is very important for industrial production. In recent years,with the rapid development of technology, the chemical processes have more advancedequipment, more complex production processes, more amounts of investment, and moreadvanced level of intelligence, so a particularly important question is how to achieve itssecurity and reliability. And one of the most important method is to use fault diagnosistechnology to improve the safety and reliability of chemical process. So for chemicalprocesses fault diagnosis technology is very significance. In this thesis, regarding atypical chemical production process TE (Tennessee Eastman) process as the researchobject, providing two modified kernel clustering algorithm, improving the accuracy andrapidity of fault diagnosis system.In this thesis, firstly summarizes the significance and development of faultdiagnosis, and summarizes the existing TE process fault diagnosis methods, then has adetailed exposition of TE process.Then, for the accuracy of TE process fault diagnosis system, has a comparativestudy of existing TE process fault diagnosis methods, proposed an modified kernelclustering algorithm, namely kernel principal component analysis (KPCA) and fuzzykernel C-means clustering (KFCM) algorithm, and simulations are applied to TEchemical process The comparison shows that kernel principal component analysis(KPCA) and fuzzy kernel C-means clustering (KFCM) algorithm is better than fuzzykernel C-means clustering method.Again, for the rapidity of TE process fault diagnosis system, has a comparativestudy of existing TE process fault diagnosis methods, proposed an modified incrementalkernel clustering algorithm, namely kernel principal component analysis (KPCA) andincremental fuzzy kernel C-means clustering (IKFCM) algorithm, and simulations areapplied to TE chemical process. The comparison shows that kernel principal componentanalysis (KPCA) and incremental fuzzy kernel C-means clustering (IKFCM) algorithmis better than incremental fuzzy kernel C-means clustering method. Finally, build the process monitoring system application platform at LabVIEWenvironment for comparative study of the two modified methods.