A Kohonen SOM based, machine health monitoring system which enables diagnosis of faults not seen in the training set

A unique system is described which uses Kohonen self organising networks and an expert system to enable predictive maintenance of rotational machinery. The system can be trained on vibration data recorded from a machine operating at full health throughout its normal operating envelope. Once trained, the system can monitor the machines vibrations detecting and diagnosing fault conditions. In this way, faults are diagnosed which were not included in the training set, thus overcoming the problems involved with collecting fault data.