Diagnóstico de falhas em processos industriais via causalidade de Granger

Abstract

Faults that propagate in industrial plants can cause many control loops to operate outside of regions of security and greater economic profits. In this context, fault diagnosis in industrial plants is an essential feature for the company’s competitiveness. In this paper, a study in the area of fault detection, identification and diagnosis is presented. A bibliographic revision is performed about the existing techniques and a method to indicate the source of faults directly from operating data is presented. The control loops affected by the disturbance are identified based on control charts, principal component analysis and Hotelling’s T2 statistic and the causal relationships between them are detected via Granger causality. The methodology was applied to two case studies: a thermoelectric power plant subject to disturbances due to switching fuels used to generate energy; and a pelletizing furnace, that has a many control loops interactions caused by energy reuse. The proposed algorithm correctly indicated the sources of faults that propagated through the control loops.