Numerical and Neural Network Investigation of a New Dual Self-centering Energy Dissipation Device

Abstract

Proposed self-centering (SC) seismic systems can reduce residual drift, but their SCand energy-dissipating (ED) solutions often face limitations, including reliance onyielding components, full-bay tendon layouts, and detailing complexity in beam-column connections. This paper investigates a compact SC-ED sub-assembly com-prising a high-strength post-tensioned multi-wire steel tendon routed in a 180° patharound a roller bearing, with both tendon ends aligned. A three-dimensional nonlin-ear finite element (FE) workflow, building on earlier numerical work and selectivelyrefined here, is used to evaluate response under a displacement-controlled cyclicprotocol representative of seismic demands. Results show stable flag-shaped hys-teresis without residual displacement and elastic behaviour of the primary compo-nents. ED is interpreted as arising from tendon-bearing interface friction and curva-ture-activated internal mechanisms in the multi-wire tendon. An artificial neuralnetwork (ANN) surrogate model trained on force-displacement data reproduces thenonlinear response on unseen data, enabling rapid prediction. The study demon-strates the feasibility of a tribology-driven curved tendon-bearing SC-ED sub-as-sembly for seismic-resilient structural connections.