抄録
Purpose of this study is to trail on defect detection using strain checker and neural network. Training data and input data in neural network were used both strain distributions obtained by finite element analysis (FEA), and measured with the strain checkers, respectively. Neural network program employed in this study is statistical analysis software "R". Stress distributions on plate specimen with an artificial defect subjected to 3-points bending moment were investigated from analytical and experimental methods. Neural network was carried out to estimate for a position and a depth of the defect. Furthermore, effects of distance from load point and geometry variation of defect on detection accuracy were simulated. As a result, it was confirmed that this approach is applicable in the defect detection.
