Alternative TitleStudy on Quantification of Hammering Test Method for Concrete Structures by Machine Learning
Note (General)type:Thesis
Concrete structures constructed during Japan's high-growth period are now more than 50 years old, and the damages of aging deterioration are beginning to become apparent. The rationalization and automation of their maintenance are strongly needed, given the declining working population due to the falling birthrate and aging population, and the large socioeconomic burden that is expected to be placed on maintenance and management. On the other hand, recent years have seen the development of technologies related to machine learning and artificial intelligence, along with the improvement of computer capabilities and the rapid progress of information networks. Considering these circumstances, this study empirically examines the possibility of introducing machine learning technology into the quantification of hammering test method to rationalize and improve the accuracy of structural diagnosis technology.In this paper, I first conducted hammering tests on small models subjected to artificial defects and on a real-scale specimen subjected to internal cracks caused by electrical corrosion on the steel bars. It was confirmed that the classification accuracy by machine learning of CNN (Convolutional Neural Network) with the imaged hammering sound waveforms was equivalent to that of a skilled technician. Furthermore, we confirmed the applicability of this method to the field by appropriately classifying the deterioration state in an actual structure. I proposed a quantitative evaluation method of defects at an arbitrary site based on imaged sound waveform using a trained CNN as a feature extractor. The generalization performance of the proposed method was examined through cross-validation using hammering sound data from three different sites. As a result, it was shown that the proposed method has the potential to achieve generalization performance that can quantitatively evaluate the integrity and defects of concrete structures at arbitrary sites.
Collection (particular)国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
Date Accepted (W3CDTF)2023-07-08T03:42:31+09:00
Data Provider (Database)国立国会図書館 : 国立国会図書館デジタルコレクション