A deep learning-assisted inverse design framework for nacre-like composites with excellent specific damping performance

Achieving stiffness and damping are key requirements for total mechanical energy loss, and unfortunately, these two properties are often mutually exclusive. Nacre-like composites have been experimentally proven to have both high modulus and high damping, however, there is still a lack of design tools to generate new nacreous microstructures that meet the desired requirement of specific damping performance. In this paper, we propose a deep learning-assisted inverse design framework, called the Denoising Diffusion Probabilistic Model with Frequency-aware feature Fusion and External Attention module (DDPM-FFEA), to generate new nacreous microstructures meeting excellent specific damping performance. The frequency-aware feature fusion and external attention module are integrated into DDPM to provide more accurate spatial boundary features and a larger structural design domain, making the DDPM-FFEA framework suitable for designing nacreous composites with clear feature consistency and boundary definition within a limited design domain. The DDPM-FFEA inverse model is trained to highlight nacreous composites with counterintuitive asymmetric microstructures by specifying ultrahigh specific damping property. By studying various asymmetric microstructures, the ultra-high specific damping performance is attributed to the tensile-bending coupling effect. Microstructural asymmetry becomes a key morphological feature in damping materials, which may explain why asymmetry-induced gradient materials are widely present in natural materials.