Fatigue and fretting-fatigue in overhead line conductors are of major concern for Transmission Systems Operators (TSO) in charge of electrical power grids. These phenomena are caused by Aeolian vibrations and may dramatically reduce the lifetime to failure of the corresponding assets. As the French grid is ageing, RTE (Réseau de Transport d’Electricité) is addressing this issue through its R&D department and the OLLA (Overhead Line Lifetime Assessment) project. The purpose of OLLA is to estimate the ultimate lifetime for virtually every span of the grid relying on historical environmental data, physical understanding and modelling, and Artificial Intelligence (AI). In that context, the proposed work focuses on the multiscale modelling of a portion of overhead conductor using Finite Element Analysis (FEA) modelling, experimental data and Artificial Neural Networks (ANN). Overhead conductors in powerlines are, just like every other wire ropes, composed of multi-layered helical strands that can be made of quasi-pure aluminium, steel or aluminium alloys. All these strands are subjected to a large number of contacts with their respective neighbours, where fretting may occur. Fretting is a contact loading happening when two contacting bodies suffer a small oscillatory relative displacement. It can nucleate cracks that may ultimately lead to wire failures thus damaging the conductor itself. The multiscale approach developed in OLLA uses a macroscopic FEA model that can estimate fretting loadings within the assembly, while a local FEA model account for a single fretting-fatigue contact between two wires. This model is suitable with refined tools to estimate fretting-fatigue lifetimes, through the application of multiaxial fatigue criteria, crack propagation analyses and complex plastic laws. All this numerical modelling has been supported by various experimental studies that produced valuable experimental data on this very specific configuration. The last step of this strategy is to apply ANNs that take as inputs both FEA and experimental results to produce a hybrid FEA/AI numerical twin of a span. This tool offers much more promising results than models relying solely on FEA or AI algorithms trained on the experimental data available. Finally, ANNs are also considered for establishing efficient metamodels for the FEA models described earlier in order to achieve the fastest calculation chain possible.