DEVELOPMENT OF ATMOSPHERIC CORROSION MODEL OF BRIDGES BASED ON WEATHER INFORMATION BY COMPUTATIONAL FLUID DYNAMICS AND MACHINE LEARNING

Abstract

 The atmospheric corrosion prediction model for maintenance of bridges based on weather monitoring and high-resolution weather simulation was investigated. First, a method combining a computational fluid dynamics model and a statistical procedure proposed by CRIEPI had a capability to estimate the amounts at locations of surveyed bridges and also wide-area distributions of the cumulative amount of airborne sea salt by considering the local topography. Second, a machine-learning approach to corrosion and meteorological data extracted several feature parameters. The generalized predictive model for atmospheric corrosion rate was built on the basis of Random Forest algorism. The accuracy of the prediction model was verified using data obtained from the exposure test at Choshi, and at the actual three bridges located in Choshi area. Based on this prediction model, the corrosion map with 1 km mesh spatial resolution was created by open source GIS software.