Structural analysis of models for fault diagnosis
Prof. Erik Frisk (Linköping University; Sweden)

Model based fault diagnosis concerns detecting and locating faults by comparing expected and observed process behavior. Models of industrial applications are often complex and model-based techniques therefore often faces a general, large-scale, and non-linear differential-algebraic model consisting of hundreds or thousands of equations, possibly written in a language like Modelica.

Methodological treatment of such complex models often requires specialized techniques for specific classes of systems. One successful way to manage the complexity is to utilize the model structure using graph-based algorithms. This talk will show how structural analysis can be a powerful tool for computer-aided analysis and synthesis of diagnosis algorithms. Making it possible to find answers to fundamental questions, e.g., which faults can be detected or isolated for a given model or which sensors should be added to achieve a desired fault isolability performance specification. After analysis, structural analysis can also guide design of fault diagnostic observers and fault detection filters. This talk will present methodology and computer tools that scales with model size and are applicable to a wide set of diagnosis problems and results from experiments on an automotive engine will be demonstrated.