刚度
联锁
反向
有限元法
非线性系统
协方差
生成语法
计算机科学
反问题
能量(信号处理)
变形(气象学)
执行机构
适应(眼睛)
能量收集
结构工程
协方差矩阵
刚度矩阵
设计策略
算法
材料科学
材料设计
基质(化学分析)
先验与后验
控制理论(社会学)
结构健康监测
随机森林
机械工程
统计物理学
随机矩阵
数学优化
粒子群优化
方位(导航)
粒子(生态学)
生成设计
遗传算法
作者
Jiawen Xu,Guohui Wang,Run Lin,Oana Ghita,Xijin Hua
标识
DOI:10.1016/j.matdes.2026.116788
摘要
Structures with tunable mechanical properties enable adaptive load bearing and energy absorption, yet their inverse design remains challenging because structural parameters, confinement conditions, and mechanical responses are coupled through nonlinear and many-to-one relationships. These difficulties are amplified in small-data settings, where deep generative models are hard to train and often lack physical interpretability. Here, we propose a random-forest-guided inverse-design framework for tunable-stiffness interlocking structures under prescribed confinement conditions. A calibrated RF forward model identifies feasible design regions for a target stiffness band and generates diverse candidate structures through likelihood-guided sampling, avoiding direct learning of a nonunique response-to-design mapping. Using fewer than 250 finite-element-generated geometry–pressure samples, the framework demonstrates improved sample efficiency, feasible-design ranking, and design-space coverage compared with genetic algorithms and covariance matrix adaptation evolution strategies. PDP, SHAP, and interaction analyses reveal how pressure and geometry jointly control stiffness accessibility and regime transitions. Additional finite element analyses show that increasing confinement activates inter-particle contacts, promotes particle rearrangement, enhances frictional dissipation, and forms connected force-transmission pathways, thereby increasing stiffness, peak force, and energy absorption. The calcaneal off-loading example further demonstrates the framework’s potential for application-oriented design of adaptive load-bearing and energy-absorbing systems.
科研通智能强力驱动
Strongly Powered by AbleSci AI