Smart Process Analytics for Process Monitoring

Process monitoring is critical to ensuring product quality and efficient, safe process operation. Data-driven modeling is used in the process industries to build fault detection systems. No single data-driven modeling method provides the best fault detection performance for all process systems, and the selection of the best data-driven modeling method for a specific process system requires substantial expertise. In this study, we propose Smart Process Analytics for Process Monitoring (SPAfPM), a systematic framework for automatic method selection and calibration of data-driven fault detection models. A set of candidate methods is pre-selected from a library on the basis of the characteristics of the data. A rigorous cross-validation procedure is then employed to compare the models obtained by these methods to select the best data-driven model for fault detection. The performance of SPAfPM is demonstrated in four case studies, including the Tennessee Eastman Process. •A systematic approach is proposed for automated selection of fault detection methods.•Candidate methods are preselected on the basis of characteristics of the data.•The preselection is informed by a combination of theory and Monte Carlo studies.•A rigorous cross-validation procedure selects the best model for fault detection.•The approach is demonstrated in four case studies.