PURPOSE: To provide an abnormality content deciding method for the fault predicting device which can accurately decide the contents of abnormality prediction when abnormality is predicted.

CONSTITUTION: When a fault is predicted, current decision data are stored (S12) and existent relative data are searches for similar data (S13). When the decision data are similar to the existent data, the abnormality contents are displayed (S14) and it is decided whether or not they match the actual abnormality contents (S16). When there are not the similar data, 'abnormality content is not fixed' is outputted (S15). If the actual abnormality contents are not matched at this time, abnormality contents are inputted (S17) and the abnormality contents and relative data at the point of time of the prediction are related and stored (S18). Those flows are repeated and while storing and learning various abnormality contents and their data during the operation, prediction enabling the specification of abnormality contents is increased.