Multiple-Model Based Fault-Diagnosis: An Approach to Heterogeneous State Spaces

摘要

In the context of diagnosing technical systems, in addition to pure fault detection, it is also important to determine the fault location and the fault size (also known as fault identification). In many cases, model-based diagnostic methods are used for fault-localization and - identification, which are based on the simultaneous use of several mathematical models, so fault-localization and - identification can be performed in the sense of a multiple model estimation. Known approaches to multiple model estimation, such as generalized pseudo-Bayesian approaches or the Interacting Multiple Model approach, use a stochastic filter for each of the models, with the results of the individual stochastic filters being suitably aggregated. However, it has to be taken into account that the individual mathematical models have different dimensions and/or physically heterogeneous state spaces. Consequently, the aggregation, i.e. the weighted combination of the estimates can not be done without appropriate modification. Using a hydraulic cylinder, possible modification approaches for the interaction of filters applied to heterogeneous state spaces are explained.