Integrating lumped damage mechanics and LSTM networks for the time-dependent reliability assessment of tubular offshore trusses under stochastic corrosion.2025. Montilla, S.; Bosse, R.; Gidr˜ao, G.; Picón, R.; Beck, A.; Wei Li.; Flórez López, J.

A probabilistic framework is presented for the time-dependent reliability of tubular truss structures under stochastic corrosion and deterministic wind lateral loads. The chemo-mechanical Lumped Damage Mechanics (LDM) element treats corrosion as a nodal degree of freedom and captures spatial–temporal degradation, local buckling under compression, plastic hardening in tension and counter-buckling under cycles. This physics-based model is coupled with data-driven tools: a Random Forest classifier is used to estimate limit-state exceedance in order to generate fragility curves, and an LSTM network is used to forecast node-level damage trajectories. An offshore truss case study involving 8800 simulations yields 666,578 sequences for training purposes, demonstrating widespread moderate damage without collapse and identifying critical members. It also shows that corrosion amplifies both capacity loss and uncertainty over service life. This hybrid framework offers a computationally efficient and scalable lifecycle assessment, supporting risk-informed inspection and maintenance in challenging marine environments.