Unified model-based fault diagnosis for three industrial application studies

This paper proposes a unified scheme for fault detection and isolation (FDI) that integrates model-based and multivariate statistical methods. For creating suitable models, subspace model identification is utilized together with state-observers to track the measured process operation. To describe and analyze the impact of fault conditions, the scheme utilizes input reconstruction and unknown input estimation to generate multivariate residual-based statistics. In contrast to existing work, the paper presents three industrial application studies involving sensor faults, as well as process and actuator faults which result from measured and unmeasured disturbances.

Research highlights
► A unified scheme for Fault Detection and Isolation is proposed.
► Unified scheme relies on integration of model-based and multivariate statistics based techniques.
► Integration of methods alleviates limitations of individual approaches.
► The scheme is designed to detect sensor, actuator and complex process faults.
► Three industrial cases are presented to demonstrate the range of potential FDI applications.