Surface preparation grades evaluation system for spliced joints of steel bridges using YOLO and deep metric learning

Abstract

For the longevity of steel bridges, thermal spraying is performed as a corrosion protection measure. Before thermal spraying, it is necessary to apply surface preparation as specified by ISO standards, which involves removing coating and rust on the steel surface by blasting. Currently, the preparation grades on the steel surfaces are evaluated by human visual inspection. In order to prevent bias in human judgments, a quantitative evaluation method is necessary. In this paper, we developed a system for automatically evaluating the surface preparation grades for spliced joints with high-strength bolts on steel bridges. Spliced joints serve to connect structural members using high-strength bolts. However, due to the complex geometry around the bolts, surface preparation tends to vary in quality. In our system, we used YOLO to detect high-strength bolts and obtain images of splice joints with the high-strength bolts masked and evaluated the surface preparation grades of the splice plates using deep metric learning. We applied the contrast limited adaptive histogram equalization (CLAHE) algorithm to the captured images to ensure stable judgments in various imaging conditions. In operational tests using real girders, we confirmed that the system accurately detected the surface preparation grades. This reduces the burden of management tasks by recording inspection results automatically.