Abstract
We describe the acoustic characteristic of leakage sound and the possibility to detect the gas leakage sound under noisy environment. The purpose of this study is to establish the acoustic diagnosis technique for the leakage sound. In order to realize the purpose, we examined the possibility of the adoption of neural networks to the acoustic diagnosis for actual sound involved the background noise. We applied Fast Fourier Transform (FFT) as the pre-processing method and examine features of power spectrum for the gas leakage sound. The feature is that the power spectrum for the gas leakage sound are more than those for the normal sound within the range from about 5kHz to 20kHz. We, furthermore, applied the neural network as the discrimination method. We got the result that the discrimination accuracy is about 96%.
