ANN-Based Surrogate Modeling for Seismic Fragility Assessment of Double-Layer Barrel Vault Roofs Supported by Double-Layer Latticed Walls

Abstract

Double-layer barrel vault roofs with double-layer vertical walls are widely used in important public buildings because of their high structural efficiency, favorable stiffness-to-weight ratio, and architectural versatility. Although incremental dynamic analysis (IDA) is a widely accepted approach for seismic assessment, it requires numerous nonlinear time-history analyses (NTHAs), resulting in high computational cost. This study presents an artificial neural network (ANN)-based surrogate modeling framework to accurately predict the seismic response of these structural systems, reducing the need for repeated NTHAs, enabling rapid estimation of structural dynamic responses, and facilitating direct development of seismic fragility curves. The proposed framework substantially decreases computational effort while maintaining an effective balance between accuracy and efficiency. A comprehensive seismic damage database is first generated using finite element (FE) models developed in OpenSees. Fragility curves are then obtained using both the conventional IDA procedure and the proposed ANN-based surrogate approach. Results show that the ANN surrogate accurately predicts the responses of structures subjected to scaled ground motions and effectively captures their nonlinear seismic behavior. Furthermore, the resulting fragility curves closely match those from the conventional IDA method, demonstrating the accuracy, reliability, and efficiency of the proposed framework for rapid seismic assessment of double-layer barrel vault structures with double-layer walls.