High-dimensional reliability analysis of guyed transmission towers via deep learning and progressive collapse simulation

This paper presents a reliability assessment of a guyed transmission line tower (TLT) using a high-fidelity progressive collapse model. The simulation captures critical nonlinearities, including geometric and material effects, eccentricities, bolt slippage, and stiffness degradation. Since each evaluation can require up to five days, an active learning framework (AL-SNDGPR) is proposed to make this high-fidelity assessment computationally feasible. This methodology utilises spectral normalisation (SN) to enhance surrogate stability, effectively overcoming the ‘curse of dimensionality’ encountered by conventional methods like AK-MCS. Verified against an 80-variable analytical benchmark, the framework is applied to a Brazilian guyed tower with 30 random variables. Results show that the integration of SN significantly reduces prediction uncertainty. Furthermore, global sensitivity analysis quantitatively proves wind speed dominance while revealing that structural parameters with low coefficients of variation contribute significantly to the failure variance. This study demonstrates that advanced surrogate modelling is essential for evaluating the true safety margins of complex infrastructure using computationally intensive models.