Abstract Abstract Many infrastructures were built intensively in Japan during the period of high economic growth Many infrastructures were built intensively in Japan during the period of high economic growth (1960s), and in recent years, the damage and deterioration of infrastructure due to high aging have (1960s), and in recent years, the damage and deterioration of infrastructure due to high aging have become apparent. Therefore, it is said that investments must be made in its repair and reconstruction become apparent. Therefore, it is said that investments must be made in its repair and reconstruction in order to continue to use the infrastructures in the future. However, local governments manage most in order to continue to use the infrastructures in the future. However, local governments manage most road bridges in Japan, and the current situation places a very high financial burden on them. In addition, road bridges in Japan, and the current situation places a very high financial burden on them. In addition, local governments face the challenge of having few engineers engaged in bridge maintenance, which local governments face the challenge of having few engineers engaged in bridge maintenance, which is an extremely serious situation in terms of the system. The universal and low-cost tool that enables is an extremely serious situation in terms of the system. The universal and low-cost tool that enables prediction of the damage degree of infrastructure deterioration and its progression is indispensable to prediction of the damage degree of infrastructure deterioration and its progression is indispensable to solve these problems. In particular, the number of cases of corrosion damage to main girders on road solve these problems. In particular, the number of cases of corrosion damage to main girders on road bridges is increasing every year, and a tool that can determine the corrosion environment of bridges is increasing every year, and a tool that can determine the corrosion environment of infrastructure in advance would be extremely useful. In this study, machine learning was used to link infrastructure in advance would be extremely useful. In this study, machine learning was used to link previously measured corrosion loss data with environmental data, and the model was constructed to previously measured corrosion loss data with environmental data, and the model was constructed to predict corrosion from environmental data. predict corrosion from environmental data. The one-year (September 2013 to August 2014) exposure test results for 150 x 70 x 2t (mm) size of The one-year (September 2013 to August 2014) exposure test results for 150 x 70 x 2t (mm) size of carbon steels at six locations in Japan were used as the corrosion data for machine learning. Exposure carbon steels at six locations in Japan were used as the corrosion data for machine learning. Exposure test specimens were sampled every month, and the corrosion loss was measured. The one-month test specimens were sampled every month, and the corrosion loss was measured. The one-month averages at each location were used for the environmental data. The investigation result of correlation averages at each location were used for the environmental data. The investigation result of correlation coefficients between each environmental factor and the corrosion loss gave that wind speed and coefficients between each environmental factor and the corrosion loss gave that wind speed and temperature showed relatively high correlations with the corrosion loss. As a result of examining each temperature showed relatively high correlations with the corrosion loss. As a result of examining each learning model for the accuracy of the correlation, it was found that the ensemble model using decision learning model for the accuracy of the correlation, it was found that the ensemble model using decision trees can estimate corrosion loss with the highest accuracy. The outdoor corrosion tests of 50 x 50 x trees can estimate corrosion loss with the highest accuracy. The outdoor corrosion tests of 50 x 50 x 2t (mm) sized carbon steel specimens on three bridges in the coastal area verified the certainty of the 2t (mm) sized carbon steel specimens on three bridges in the coastal area verified the certainty of the corrosion prediction model constructed. The corrosion losses estimated by the corrosion prediction corrosion prediction model constructed. The corrosion losses estimated by the corrosion prediction model based on environmental data at each bridge was compared with those obtained from outdoor model based on environmental data at each bridge was compared with those obtained from outdoor corrosion tests. Relatively good correlations were observed for bridges in coastal areas, whereas for corrosion tests. Relatively good correlations were observed for bridges in coastal areas, whereas for bridges in inland areas, the estimated values were larger than the measured ones. bridges in inland areas, the estimated values were larger than the measured ones. Hideki Katayama is the Director of the Materials Evaluation Field, Research Center for Structural Hideki Katayama is the Director of the Materials Evaluation Field, Research Center for Structural Materials, National Institute for Materials Science. He also concurrently serves as Group Leader of the Materials, National Institute for Materials Science. He also concurrently serves as Group Leader of the Corrosion Research Group in the same field. He holds a Ph.D. in Metallurgical Engineering from the Corrosion Research Group in the same field. He holds a Ph.D. in Metallurgical Engineering from the Tokyo Institute of Technology. He is also a visiting professor in the Department of Advanced Chemistry Tokyo Institute of Technology. He is also a visiting professor in the Department of Advanced Chemistry at Tokyo University of Science and a part-time lecturer in the Faculty of Bioscience and Applied Chemistry at Tokyo University of Science and a part-time lecturer in the Faculty of Bioscience and Applied Chemistry at Hosei University. He specializes in corrosion science and electrochemistry. at Hosei University. He specializes in corrosion science and electrochemistry. Corrosion Risk Prediction MCorrosion Risk Prediction MField Director, Research Center for Structural Materials (RCSM), Field Director, Research Center for Structural Materials (RCSM), National Institute for Materials Science (NIMS) National Institute for Materials Science (NIMS) NIMS Award Symposium 2023Hideki Katayama Hideki Katayama NIMS Award Sessionap for Map for Maintenance of Infrastructure aintenance of Infrastructure NIMS Talk NA-6 NIMS Talk NA-6 21
元のページ ../index.html#21