1995

Kurki, Matti

This dissertation presents an approach to model based fault diagnosis for mechatronic systems. This approach is strongly based on the utilization of currently existing models, such as control system and topological models. The need to develop additional complicated models (qualitative, causal, etc.) for fault diagnosis ruled out the possibility of applying shallow approaches and also many promising model based approaches. Extensive utilization of the control system information and tight integration of fault diagnosis with control are the innovative ideas of the thesis. Descriptions of the control software include essential knowledge from the viewpoint of fault diagnosis, such as dynamic control state information, operating context, variables, and parameter values. Additionally integration with the executable control system offers the following advantages for fault diagnostics: facilitates integration and embedding, ensures quick and flexible development and opens up opportunities for "few sensor" applications. This research has proven the usefulness of the ideas of model based diagnosis performed by generic tasks. Experience accumulated from different domains indicates that diagnostic tasks developed in this work are indeed generic. The diagnostic tasks offer a solution to the whole process which includes preprocessing, fault detection, localization and recovery. The fault detection approach is based on modeling of the normal operation and behavior of the diagnosed system where failures are detected as a complement of the normal operation and behavior. Sensitivity to the operating context derived from behavioral models and a learning mechanism are primary matters in fault detection. Fuzzy logic, the connection of rules to models, a case based approach, and assistant information are used as technologies for fault localization and recovery. The model based fault diagnosis approach has been applied to a number of real life mechatronic systems. The approach can be easily adapted to different applications and environments. The approach has noticeably shortened the development time of fault diagnosis systems and resulted in improved fault diagnosis features.