AUTOMATED PREPARATION GRADE JUDGEMENT SYSTEM FOR WEATHERING STEEL BRIDGES BY DEEP LEARNING

Abstract

The quality and durability of repair coatings for steel bridges are significantly affected by the surface preparation prior to repair coating. Surface preparation by blasting stipulated by ISO Sa2 1/2 or higher is required as surface treatment before repair coating. The weathering steel bridge have unevenly corroded surface (rust grade D), and it is difficult to evaluate the surface condition by comparing the photographic example. Therefore, there is a high possibility that the difference in evaluation occurs by the engineer. In this paper, the authors attempted to develop a support system to determine preparation grade from the image of blasting. The purpose of this paper is to discuss the accuracy of the trained CNN classifier by the difference in the size of the teaching image data, and to evaluate the performance of the classifier by test data. It was considered that the high value of 83% or more was able to be obtained by the result of all evaluation indices, and the classifier with high performance was constructed by this method.