Inverse design of energy-dissipating steel plate walls based on self-supervised diffusion models

Abstract

This research proposes a novel energy-dissipating system integrating non-structural infill walls and energy-dissipating steel plate walls (EDWs) to improve the seismic performance of frame-infill wall structures. Based on the diffusion model, a self-supervised inverse design model (SSIDM) is developed to enhance the design accuracy of complex EDWs. SSIDM is developed by constructing a pre-trained hysteresis performance prediction network (HPPN) and a microstructure generation network (DiffEDW). During the microstructure generation process, a discriminator is employed to predict the overall mechanical performance of the EDW. In cases where the mechanical performance does not meet the expected demands, additional Gaussian random noise is introduced, and the generation process is repeated until an EDW structure that satisfies the hysteresis performance target is obtained. SSIDM learns the conditional distribution of microstructures corresponding to a given complete hysteresis performance, enabling a one-to-many mapping from attributes to geometry. Ablation experiments demonstrate that the self-supervised diffusion model with the HPPN generates EDWs with smaller errors in the finite element analysis (FEA) results and mechanical performance targets compared to the model without the HPPN. This study reveals that the self-supervised method offers significant advantages in generating periodic large-scale complex topologies and demonstrates potential for accelerating multi-scale structure generation. The dataset and related code are available at https://github.com/AshenOneme/SSIDM.