Fault Diagnosis Using Novel Class-Specific Distributed Monitoring Weighted Nal¨ve Bayes: Applications to Process Industry

Safety management of the process industry plays a significant role in protecting life and property. Fault diagnosis techniques have been widely utilized for safety management of the process industry. However, an acceptable fault diagnosis accuracy is difficult to achieve due to the large scale and the high integration of modern industrial processes. To deal with this issue, in this paper a novel class-specific distributed monitoring weighted nal¨ve Bayes (CDMWNB) method is proposed to improve the fault diagnosis performance of complex processes. In the proposed CDMWNB method, first, the whole process should be divided into subblocks by decomposition; second, dynamic independent component analysis (DICA) is used to obtain the I² statistic and the control limits (CLs) in each subblock; and finally, the proposed CDMWNB method can be developed for fault diagnosis. To prove the effectiveness of the proposed CDMWNB method, case studies of fault diagnosis using the Tennessee Eastman (TE) benchmark process are carried out. The effectiveness and feasibility of the proposed CDMWNB method are proven by simulation results.