Corrosion-driven section loss in steel tower structures erodes load-carrying capacity, yet field assessment still relies on subjective visual grading. This paper presents a closed-loop framework coupling quantitative image-based corrosion measurement with stochastic degradation modeling, Monte Carlo reliability simulation, and Sobol’ variance-based global sensitivity decomposition. Two complementary segmentation paths—hue–saturation–value (HSV) color-space thresholding for fleet-scale screening and DeepLabV3+ deep learning for detailed evaluation—convert imagery into calibrated section-loss estimates via nonlinear regression. Three analysis modes (single-image, multi-angle weighted-median fusion, and Oriented FAST and Rotated BRIEF (ORB) feature-matched temporal differencing) feed a Bayesian-updated power-law corrosion growth model whose outputs propagate through a time-dependent limit-state function via 106-sample Monte Carlo simulation. Sobol’ indices rank each uncertain input’s contribution to the reliability-index variance. A field demonstration on a 40-year-old galvanized lattice tower in an ISO 9223 C4 coastal environment shows that the corrosion rate constant and zinc coating thickness together govern 65% of the total reliability variance and that a risk-ranked selective maintenance strategy reduces expected life-cycle cost by 71% relative to blanket intervention.
Zhu et al. (Tue,) studied this question.