Surrogate-Based Probabilistic Fatigue Assessment of Transmission Lines with Neural Network: From Wind Data to Inspection Planning

Transmission line towers experience frequent fluctuations in wind loads, which can cause fatigue damage in bolted connections, thereby risking their long-term reliability. Direct probabilistic fatigue assessment of transmission towers using nonlinear time-domain simulations under stochastic wind loading is computationally infeasible for long-term reliability and inspection planning, motivating the use of surrogate models to enable efficient uncertainty propagation and decision analysis. This study introduces a surrogate-based probabilistic framework to evaluate fatigue damage and inform inspection scheduling. High-fidelity finite-element models simulate joint responses under both preloaded and loosened bolt conditions, with fatigue assessed through rainflow cycle counting and Miner’s rule. The resulting simulation database is used to train neural network surrogates, enabling rapid prediction of fatigue accumulation across wind speeds and directions. These surrogates are integrated with joint wind probability distributions and a Poisson-based extreme-event model to estimate annual damage and account for uncertainty through sampling. Results identify critical fatigue-prone connections and examine how stress factors and bolt preloading influence vulnerability. Using these findings, inspection strategies are optimized by considering detection probabilities and cost-risk trade-offs. Overall, this approach offers a fast and reliability-based tool for evaluating fatigue risk and planning inspections of transmission systems.