Surrogate model based approach to predict fatigue stress field in multi-stranded cables. (November 2021)

• Record Type: Journal Article
• Authors: Belkhabbaz, A.; Gueguin, M.; Hafid, F.; Yang, C.; Allix, O.; Ghidaglia, J.-M.
• Is Part Of: International journal of solids and structures. Volume 230/231(2021)
• Publication Date: 2021-11
• Subjects: Multi-stranded cable -- Fretting-fatigue -- Contact-friction -- Surrogate models; Mechanics, Applied -- Periodicals; Structural analysis (Engineering) -- Periodicals; Elastic solids -- Periodicals; Mécanique appliquée -- Périodiques; Constructions, Théorie des -- Périodiques; Solides élastiques -- Périodiques; Elastic solids; Mechanics, Applied; Structural analysis (Engineering); Periodicals; 624.18
• DOI: 10.1016/j.ijsolstr.2021.111168
• Languages: English
• ISSNs: 0020-7683

Abstract

Highlights: Simulation of a overhead line conductor with a large number of contacts. Implementation of a regularized contact law improving convergence of the algorithm. Demonstration of a surrogate model that approximates the beam element model. Influence of sampling method allowing the best fitting surrogate model. Abstract: The mechanical behavior at contact points in high-voltage electrical cable conductors is simulated to find where damage may occur. The presented approach is based on a numerical model that is referred to as the wire model. The critical zones are detected by simulating the mechanical behavior of a cable conductor section with a large number of contacts. For the wire model, a relevant regularized contact law that improves convergence of the contact algorithm compared to the standard contact law is implemented. This model is validated by making comparisons with experimental results available in the literature for various conductors. However, this approach employs a costly finite element method. In order to reduce the evaluation cost and allow sensitivity analysis of the input parameters, a surrogate model is proposed to replace the initial wire model. The surrogate model approximates the input–output relationship of the costly wire model. The choice of the sampling method that allows selection of the best fitting surrogate model for this application is discussed, but comparisons show little influence on the obtained scores. The robustness of the obtained result depends mostly on the chosen surrogate model, where the scores obtained with random forest and neural network are better than the scores obtained with linear and polynomial regression. The application is the French high-voltage network, but the work is relevant for any other network.