Monitoring-performance-indicator-related industrial process monitoring with a monitoring index identification model

•A process monitoring framework MII is proposed.•MII is an extensible framework and can increase ADR and decrease FAR simultaneously.•MII method has no limitations on the specific data distribution.•Monitoring results show MII can enhance the performance of MPI. The interaction and balance between anomaly detection rate and false alarming rate in industrial process monitoring cannot be ignored. In contrast to traditional methods, in this study, the final monitoring index in the monitoring task is integrated into the objective function for the first time. A method called monitoring index identification (MII) is proposed to improve the edge discrimination between normal and abnormal monitoring indices. The contributions are as follows: First, we maximize the compression of the range of normal monitoring indices and maximize the separation of the normal and abnormal monitoring indices, and then propose a general framework MII that can accommodate different indices in the monitoring task. MII significantly reduces the overlap region between normal and abnormal monitoring indices while increasing the anomaly detection rate and reducing the false alarming rate. Second, MII is extended based on the conventional linear/nonlinear process monitoring model and can ignore the model limitation caused by the data distribution assumption. Third, the objective function is constructed from the perspective of monitoring indices, which is helpful in capturing the low-dimensional projection direction of the monitoring indices. Furthermore, statistics-based MII/KMII process monitoring methods are developed. The experimental results illustrate that the proposed method performs well in balancing the anomaly detection rate and false alarming rate to improve the monitoring performance indicator.