摘要 |
PROBLEM TO BE SOLVED: To automatically and accurately detect abnormality of an inspecting object from a physical quantity having plural frequency components. SOLUTION: A microphone 12 collects the sound of an engine 10, converts it into an electric signal, and inputs the electric signal into each of filter banks 22 which constitute a filter part 16 of a signal processing device 14. Each filter bank 22 has a difference transmitted frequency band, inputs an output signal in a feature variable extractor 24 of a feature variable extracting part 14, and the feature variable extractor 24 extracts a feature variable and inputs it in a neural network 26 of a feature variable abnormality discriminating part 20. The neural network 26 already learns a reference feature variable extracted from a normal engine, an abnormal engine, or the like, and judges whether the feature variable extracted with the feature variable extractor 24 is normal or not. A judging part 28 judges whether the engine 10 is normal or not on the basis of the judged result of each neural network 26.
|