抄録
The effective far-side magnetic flux leakage technique for the inspection on the bottom plates of oil tanks was proposed as a quantitative evaluation method on local corrosion. Based on the technique, the flaw characterization obtained from the inspection data under the saturation magnetization by using the neural network was presented, and its effectiveness has been proven in this paper. In addition, the practicality of the proposed neural network has been confirmed by its successful application to a large steel plate such as the bottom plates of oil tanks when original data was corrected with the result of the finite element simulation. Further improvement on the accuracy of quantitative evaluation on natural flaws can be expected by training neural network with different signals.
