| 摘要 |
PROBLEM TO BE SOLVED: To accurately deduce the internal state of a target to be inspected, without requiring much skill. SOLUTION: The inspection target X is struck by the hammer part 21 provided to vibration-producing source 2 to make vibration transmit to the inside of the inspection target X. This vibration is detected by a sensor part 11 and inputted to a feature quantity extraction part 13 via a signal input part 12, and the frequency component of the vibration is extracted as feature quantity. This feature quantity is inputted to a decision part 10, and the category of the feature quantity is classified by the competitive learning type neural network 10a, provided to the determining part 10. Membership degrees are made to associate with the distance of the weight vector of the neuron of the output layer, in the competitive learning type neutral network 10a with the feature quantity. By making the competitive learning type neutral network 10a learn on the basis of the feature quantity obtained from the normal inspection target X, the inspection target X is decided as being abnormal, when the membership degrees deviate from normal. COPYRIGHT: (C)2008,JPO&INPIT
|
