Robust Model-Based Fault Detection Using Monte Carlo Methods and Highest Density Regions
European Conference of the Prognostics and Health Management Society (PHM 2021)
Contribution to Conference
One of the major problems of model-based fault detection is to account for model and measurement uncertainties in order to robustly detect occurring faults. This paper presents a method which utilizes Monte Carlo simulations to solve this problem for hybrid nonlinear models. By sampling the a-priori and statistically identified uncertainty distributions, corresponding residual values are obtained. The distributions of these residuals are analysed using highest density regions to obtain information about the probability of receiving the observed measurements given a fault-free model. In addition to the basic method, an extended method utilizing explicit fault models is presented. Both methods are implemented in form of an algorithm and, in order to provide a proof of concept, applied to the model of a cooling system for an unmanned aerial vehicle.