In current engineering practice, the optimal design of the classical viscous damping wall (VDW) model typically relies on traditional optimization algorithms, which often involve the use of a large number of VDWs. To address this problem, this study proposed a three-dimensional VDW and validated a simplified model that matched its performance under multidirectional earthquakes. Based on the proposed three-dimensional VDW model and the corresponding simplified model, a physics-informed VDW automated design framework named VDW Automated Design-DRL (VAD-DRL) was further proposed, which integrated the finite-element method with deep reinforcement learning (DRL). The framework modeled the process of optimizing VDW placement and types as a DRL process, where the agent continuously interacted with the VDW Automated Design-Env to learn the design approach and generate optimal VDW designs under multidirectional earthquakes. An engineering case was finally applied, where VAD-DRL was used for VDW design under multidirectional earthquakes. The elastic and elastoplastic responses of the structures with the two added VDW designs were validated. The results indicated that the Maxwell model was more in line with the working performance of the three-dimensional VDW under multidirectional earthquakes. The VDW design generated by VAD-DRL significantly reduced the structural responses and enhanced the structure’s seismic performance under multidirectional earthquakes.
Zhou et al. (Thu,) studied this question.