IJCAS

Fault Detection and Isolation (FDI) is a crucial and challenging problem in many industrial applications and continues to be an on-going research issue in the control community. In the literature, model-based techniques are mostly employed to generate residuals for diagnosis and decision-making. In this paper, we focus on the FDI problem using a novel based-clustering approach. The key idea is to restrict each fault to be a data cluster with high-density gathering the most similar objects. In this way, the algorithm does not require prior analytic models to start. It uses rather a density measurement to detect and isolate cluster’s regions. The overall algorithm is expanded around two fundamental steps: cluster domain description and density-based clustering. To address properly the requirements of system control and monitoring, the algorithm is designed to work in real-time as observations are acquired and it is endowed with specific tools for data mining and feature extraction. A study case is proposed consisting of a plastic Injection Molding Machine (IMM) to prove the effectiveness of the method.

Keywords: Clustering, decision-making, density, diagnosis, FDI, IMM.