Using Models for Dynamic System Diagnosis: A Case Study in Automotive Engineering.
Oliver Niggemann, Benno Stein, Thomas Spanuth, Heinrich Balzer · MBEES · 2009
Though various sophisticated concepts for the diagnosis of technical systems have been developed, diagnosis technology in practical applications often boils down to the use of simple heuristics, associative case memories, or manually designed decision trees. These approaches are robust, but restricted with respect to the complexity of the diagnosed system and the faults to be detected. An inviting direction, which is desired by practitioners and investigated by researchers, aims at the reuse of those models for diagnosis purposes, which were originally designed for system construction and simulation. A promising principle in this connection is model compilation: the model of the interesting system is simulated in various fault modes, and the resulting (huge) set of simulation data is analyzed with machine learning methods, yielding tailored diagnosis rules [Ste03]. To verify this approach, we present a case study from the field of automotive software development. The case study highlights strengths and weaknesses of model compilation. One key challenge is addressed in this paper: the identification of suited symptoms in the data.