Fault Detection and Localization Using Modelica and Abductive Reasoning

Modelica is an object-oriented and domain-independent programming language that is excellently suited for modeling and simulating a wide range of systems. In this chapter, after briefly discussing the use of Modelica for representing hybrid systems, we show how to use corresponding simulation results for detecting and isolating faults. To this end, we present three approaches to comparing simulated signals with actually observed behavior. This includes the use of average values and pre-defined tolerances, temporal band sequences, and the Pearson correlation coefficient. Once we identify significant deviations from expected behavior, we are, of course, interested in identifying their cause. For this task, we show how to add fault models to the Modelica system model, so that we can simulate the corresponding faulty behavior. From the described faults and their simulations, we then derive an intuitive cause-and-effect model that we can use for fault localization using abductive diagnosis. Aside describing the foundations, we illustrate the concept with examples and outline also limitations and applicability in practice.