Performance Evaluation System for Industrial Plants using the ART2 Network That Considers Cluster Size

Abstract

We have developed a performance evaluation system for industrial plants that uses data clustering technology. The proposed system, which can be used to analyze plant operational data by using a key performance indicator (KPI), consists of a data-clustering function for operational data with the KPI and a visualization function of clustered results. The data-clustering function is based on the ART2 network, one of the adaptive resonance theory (ART) networks. It can correlate plant operational data to the KPI by classifying the operational data. By using an improved data-clustering algorithm in which different vigilance parameters can be set for each category, the operational data can be classified into fewer categories, thus improving the clustering performance. The visualization function of clustered results illustrates the relation between operational data and the KPI by using a 3D graph with a z-axis indicating the KPI and an x-y plane showing where the gravity centers of the categories are mapped by multidimensional scaling (MDS). By implementing these functions, the system models and visualizes the relation between categories and KPI on a 3D graph.