This project addresses the topic of fault diagnosis of mechanical systems. Unexpected machinery breakdown can lead to lost production, expensive repair work, and safety problems. The two traditional methods of maintainance involve either letting machines run to breakdown or scheduling repairs at predetermined intervals. An alternative is to monitor machine condition in process, and implement repairs only when necessary. The research focuses on this fault diagnosis approach and attempts to cope with fault signature variability, the main source of difficulty for existing diagnostic systems. In the proposed method, multi valued influence matrix (MVIM), diagnostic signals are monitored on line, and flagged upon the detection of an abnormality. The method offers a methodical approach to diagnostic reasoning which is suited to on line application and also establishes a framework to select and optimize diagnostic sensors. Two diagnostic problems, tool breakage in machining and fault diagnosis of helicopter power trains will be used to test the MVIM method.
