Structured design of parametric fault candidates

Detection and isolation of faults in technical systems is essential to ensure safe and reliable operation. Faults might lead to dangerous operating conditions or unplanned breakdowns of processes. The overall goal of this work is to devise a set-based approach for the challenging task of fault diagnosis in nonlinear dynamical systems with particular consideration of the structured design of fault candidates, robust diagnosability and applicability in practice. To this end a set-based framework for fault detection and isolation is employed. The design procedure focuses on the parameter space and uses sets to model uncertain and qualitative knowledge. It is shown how uncertainty can be incorporated in the design of fault candidates such that parametric faults based on qualitative and quantitative knowledge can be described. Furthermore, it is elaborated how diagnosability can be achieved by suitable design of fault candidates. To provide a proof-of-concept for a practical application, the approaches are validated with hydraulic axial-piston pumps as a non-trivial industry-scale example. The theoretical contributions are adopted using real-world data from test bench experiments to design fault candidates for a set of qualitatively derived fault cases and to perform fault detection and isolation by employing guaranteed model invalidation. By doing so, the applicability of set-based approaches in demanding practical applications is demonstrated for the design of fault candidates as well as for fault detection and isolation.