A Data-Driven Key Performance Indicator-Related Monitoring Scheme for Dynamic Nonlinear Systems

This brief addresses the data-driven fault monitoring issues for nonlinear systems with dynamic operations by taking into account the key performance indicators (KPIs). To reach this objective, the KPIs related to the other variables of the nonlinear systems are first established by means of the Takagi-Sugeno fuzzy technique. Then, the input/output data model of the KPIs is constructed in fuzzy form. By applying the subspace-aided method, a fuzzy-model-based KPI predictor is obtained. On this basis, a data-driven realization algorithm of the KPI prediction residual generator is proposed with available system measurements. Towards KPI-related fault monitoring purpose, the control limit is computed by utilizing the kernel density estimation method, with which an online KPI-related monitoring scheme for dynamic nonlinear systems is presented. Compared with the existing methods, the merits of this scheme lie in that it can handle the nonlinear and dynamic characteristics of data, so as to improve the detection performance of KPI-related faults. Meanwhile, the KPI prediction model can achieve interpretability by introducing the fuzzy modeling. An experimental study on the ship propulsion system is finally given to demonstrate the developed results.