Benchmarking on Approaches to Interval Observation Applied to Robust Fault Detection

摘要

Model-based fault detection is based on generating a difference, known as a residual, between the predicted output value from the system model and the real output value measured by the sensors. If this residual is bigger than a threshold, then it is determined that there is a fault in the system. Otherwise, it is considered that the system is working properly. However, it is very important to analyse how the effect of model uncertainty is taken into account when determining the optimal threshold to be used in residual evaluation. In case that uncertainty is located in parameters (interval model), an interval observer has been shown to be a suitable strategy to generate such threshold. However, interval observers can present several problems that in order to be solved, existing approaches require computational demanding algorithms. The aim of this paper is to study the viability of using region based approaches coming from the interval analysis community to solve the interval observation problem. Region based approaches are appealing because of its low computational complexity but they suffer from the wrapping effect. On the other hand, trajectory based approaches are immune to this problem but their computational complexity is higher. In this paper, these two interval observation philosophies will be presented, analysed and compared using in two examples.