摘要 |
PROBLEM TO BE SOLVED: To precisely detect the abnormal status of a monitor object even when the number of learning data is not sufficient. SOLUTION: A featured value extraction part 2 inputs a measurement signal in the normal status of a monitor object as a learning signal from a signal input part 1, and extracts the distribution of frequency components from the learning signal as the featured values of the monitor object, and prepares basic learning data. When the number of learning data prepared by the featured value extraction part 2 is small, an additional featured value preparation part 4 prepares additional learning data whose each element is a value which is not less than the minimum value and not more than the maximum value of each element of the basic learning data. Afterwards, a map preparation part 50 inputs a plurality of basic learning data and the additional learning data as learning data to a competition type neural network without any teacher, and prepares a clustering map. COPYRIGHT: (C)2007,JPO&INPIT
|