| 摘要 |
PROBLEM TO BE SOLVED: To highly precisely decide the status of an inspection object. SOLUTION: A frequency component extraction part 2 prepares inspection data by extracting featured values with frequency components as elements from an inspection signal based on a status detected by a signal input part 1. A cluster decision part 42 calculates a Euclidean distance between inspection data and weighting factor data to extract the minimum value for every output layer neuron on a clustering map prepared by using a teacher-less competitive learning type neural network 40 by a map preparation part 41. When the minimum value is equal to or less than the threshold of the minimum value neuron, it is decided that the featured value data are belonging to a normal category. At the same time, a reference range set by a reference range setting part 70 is compared with each element of the featured value data by a data decision part 71. When it is decided that the featured value data are belonging to the normal category, or it is decided that each of the elements of the featured value data is within the reference range, it is decided that the inspection object is put in a normal status by a control part 8. COPYRIGHT: (C)2008,JPO&INPIT
|
