The Canadian Journal of Chemical Engineering · 2020 · 13 citations · 26 references
EngineeringComputer ScienceIndustrial EngineeringProcess ControlProcess DynamicsProcess Systems EngineeringCondition MonitoringDiagnosisFault DetectionFault DiagnosisPrincipal Component AnalysisProcess AutomationData-driven Fault DetectionGenetic Algorithm OptimizationGenetic Algorithm Variable SelectionIndustrial Process MonitoringProcess Anomaly DetectionProcess Fault DetectionProcess Fault DiagnosisSupport Vector Data Description
Abstract In this study, a the two‐step support vector data description (TS‐SVDD) method is proposed to handle the problem of fault detection for dynamic, non‐linear, and non‐Gaussian processes. First, the dynamic structure of the data is identified and the data is divided into two components: innovation component and dynamic component. Then, the innovation component is used to make the SVDD model for fault detection. Moreover, in order to overcome the issue with two‐step principal component analysis (TS‐PCA) that the choice of parameters q and D affects the fault detection effect of the methods, a genetic algorithm (GA) is used to optimize the parameters. The proposed method combines the advantages of TS‐PCA in processing dynamic process data and SVDD in dealing with non‐linear and non‐Gaussian process data. In order to evaluate the effectiveness and superiority of the proposed method, TS‐SVDD is applied to the Tennessee Eastman (TE) process and the intelligent industrial processes control test facility (I 2 PC‐TF), and the fault detection performance is compared with TS‐PCA and SVDD in terms of dault detection rate (FDR) and false alarm rate (FAR). The results show that TS‐SVDD has a better monitoring performance.
26