| 摘要 |
PROBLEM TO BE SOLVED: To provide an abnormality monitoring device which does not deliver incorrect determination on whether amount of features of an object signal is normal or abnormal, even if used for equipment which changes with lapse of time. SOLUTION: Sounds and vibration produced from equipment X are detected by a sensor section 2, and the amount of features is extracted from object signals outputted from a sensor section 2 in a feature extraction section 3. The amount of features extracted in the feature extraction section 3 is inputted in two units of neural network 1a and 1b. One of the neural networks 1a or 1b operates in an inspection mode and classifies a category of the amount of features. Another side of the neural networks 1a or 1b operates in a learning mode, and learns by using the amount of features as learning data. A switching determination section 4 evaluates the reliability of a decision result on the neural networks 1a and 1b which operate in the inspection mode. If the reliability of the decision result falls off, operations of both neural networks 1a and 1b are replaced. COPYRIGHT: (C)2008,JPO&INPIT
|
