1996 年 62 巻 598 号 p. 1368-1375
A new method for estimating pre-crack length is developed. First, the wavelet transform is applied to a pixel distribution which is obtained from a digital photoimage of a fracture surface. However, the result of the wavelet transform is usually ambiguous. Then the neural network is applied to the result, and the cumulative frequency distribution of the output of the neural network is used to indicate the boundary between crack surface and pre-crack surface. A fracture surface of an Al-Al2O3 composite specimen is analyzed to demonstrate the applicability of the proposed method.