Probabilistic prediction of mechanical characteristics of corroded strands

Abstract

Steel strands are widely used as important structural members of bridges. Their failure can be detrimental to the structure; therefore, various studies on predicting their mechanical characteristics have been conducted. However, explaining the mechanical characteristics of steel strands is difficult because of geometric complexity, difficulty in corrosion modeling, and various uncertain factors. This paper proposes a new method for the probabilistic prediction of the mechanical characteristics of corroded steel strands. First, finite element (FE) models are built for several types of corroded wires. Second, based on the FE analysis results, a nonparametric surrogate model is constructed using Gaussian process regression. Third, the ultimate strength and strain of the corroded steel strands are predicted probabilistically by conducting a Monte Carlo simulation with a theoretical strand model. As a result, the probabilistic ranges of 50% and 95% are estimated. Based on the prediction results, appropriate probabilistic distributions for the ultimate strength and strain are studied. The proposed method is applied to several specimens of corroded seven-wire strands. The prediction results are in good agreement with the test results. Additionally, a failure probability assessment is conducted as an application example based on the goodness-of-fit test.