構造物調査・診断(2)

[V-663] Improvement of accuracy in Impact-Echo monitoring with machine learning

〇Hiroshi Shimbo¹, Tomoko Ozeki², Toshiaki Mizobuchi¹, Junichiro Nojima³ (1.HOSEI University, 2.TOKAI University, 3.J-Power Design Co.,Ltd)

Keywords:Impact-Echo monitoring, machine learning, generalization, spectrogram, wavelet transform

In Japan, while the infrastructures developed during the high growth period is aging, appropriate maintenance are urgent social issues. There is a need for rational diagnostic techniques. The authors are developing a comprehensive evaluation method of the deterioration state of concrete structures.
In this paper, we investigated the sophistication of Impact-Echo monitoring and generalization of estimation by machine learning based on impact sound test data on actual structures. As a result, it was shown that the obtained neural network might have acquired a certain degree of generalization.