Knowledge based monitoring and diagnosis in machine maintenance

Maximum availability of cost intensive machinery is highly dependant of minimizing downtime caused by maintenance. Software systems for monitoring and diagnosis can help to reduce downtime. Permanent monitoring of a machine guarantees a reliable detection of most faults. An interactive diagnosis system can guide the maintenance staff to accomplish their tasks. A system architecture which supports permanent monitoring, interactive diagnosis as well as repair at site needs knowledge representation techniques which are specifically designed for these tasks. This paper presents a suitable knowledge representation and its inference techniques. The design of knowledge representation terms has been guided by industrial standards of fault tree analysis in order to ease direct model acquisition by the domain expert. The term `Fault´ is the main concept in the model acquired. Specific extensions define objects that describe the interface to sensors of the machine and to the end user, the maintenance engineer. A special calculus is the foundation of an inference engine that can be used in monitoring and interactive diagnosis as well