Probabilistic prediction of Load–Displacement curves of corroded strands

This study proposes a probabilistic method for predicting the non-linear mechanical behavior of corroded steel strands. The proposed method follows a four-step process: (1) Developing sophisticated finite element models that accurately represent various types of corrosion; (2) Constructing a multi-surrogate model using Gaussian process regression; (3) Predicting load–displacement curves based on a theoretical model; and (4) Implementing a probabilistic analysis using Monte Carlo simulation and kernel density estimation. Validation was performed through two approaches: (i) scenario-based synthetic simulations of 1000 corrosion cases, and (ii) experimental tensile tests on 39 real-world corroded seven-wire strand specimens. Predictions closely matched experimental results, capturing tensile strength and yield displacement within 99 % prediction bounds for 94.87 % and 89.74 % of specimens, respectively. This framework provides an effective tool for assessing corroded strands, enabling the probabilistic evaluation of prestressed concrete girders and supporting maintenance strategies for corrosion-affected infrastructure. • A probabilistic prediction framework for the structural behavior of corroded strands is developed. • Uncertainty in wire properties with corrosion is propagated to quantify impacts on structural behavior of strands. • Uncertainty is quantified by 50 %, 95 %, and 99 % prediction bounds for non-linear mechanical behavior of corroded strands. • Method verification and experimental validation confirm the effectiveness of the proposed method with in-field strand specimens.