Genetic Algorithm Based Information Retrieval in Fault Diagnosis Systems

Abstract

Plant superior management systems need a reliable management of multiple freelance faults, which is crucial for supporting plant operators’ decision-making. In this regard, the MultiLabel approach with Support Vector Machines (ML&SVM) used as the base learning formula was recently proposed for addressing the problem of coinciding fault identification in the chemical process mistreatment coaching sets comprised solely of single fault data. A new method for improving fault diagnosis performance has been proposed, which includes feature extension using the variance and linear trend of the data sets, feature reduction using Genetic Algorithms (GA), and then applying the selected options to the new training set. The identification performance was tested on the Tennessee discoverer benchmark and it absolutely was measured and compared mistreatment the F1index. Excellent results were obtained for seventeen of total of the twenty faults in the case study, while three faults (3, 9, and fifteen) could not be classified properly. As a result, the approach's capability and limitations are finally discussed.