首页|期刊导航|四川轻化工大学学报(自然科学版)|基于纳米压痕和LSTM-GA预测材料力学性能

基于纳米压痕和LSTM-GA预测材料力学性能OA

Prediction of Mechanical Properties of Material Based on Nanoindentation and LSTM-GA

中文摘要英文摘要

基于低阶应变梯度塑性理论(CMSG)和纳米压痕有限元模拟,结合量纲分析、长短期记忆(LSTM)人工神经网络及遗传算法(GA),提出了一种预测材料等双轴残余应力及本构参数的方法.通过对AA 7050 T451铝合金的纳米压痕载荷-位移曲线数据进行反演分析,验证了该方法的准确性.结果表明,所推导的无量纲函数具有良好的拟合效果,决定系数(R²)均大于0.994.经过训练和验证的人工神经网络在预测纳米压痕载荷-位移曲线形貌参数时,预测误差(MAPE)均不超过5%.此外,AA 7050 T451铝合金纳米压痕数据的反演分析表明,该方法对等双轴残余应力的预测误差为7.28%,同时反演的本构参数与实验数据高度一致.

Based on the conventional theory of mechanism-based strain gradient plasticity(CMSG)and nanoindentation finite element simulation,this study presents a methodology for predicting equi-biaxial residual stress and the constitutive parameters of materials.The approach integrates dimensional analysis,long short-term memory(LSTM)artificial neural network,and genetic algorithm(GA).The accuracy of the method is validated through the reverse analysis of the load-displacement curve data for the AA 7050 T451 aluminum alloy.The results indicate that the dimensionless functions exhibit strong fitting,with a determination coefficient(R²)exceeding 0.994.Furthermore,the mean absolute percentage error(MAPE)of the trained and validated artificial neural network is less than 5%when predicting the shape parameters of the load-displacement curve from nanoindentation.Additionally,the reverse analysis of the nanoindentation data for the AA 7050 T451 aluminum alloy reveals a prediction error of 7.28%associated with the proposed methodology.Furthermore,the constitutive parameters obtained from the reverse analysis demonstrate a strong correlation with the experimental data.

朱达炫;张波;孙占坤;刘明

西南交通大学 力学与航空航天学院,四川 成都 611756西南交通大学 力学与航空航天学院,四川 成都 611756内蒙古工业大学 理学院,内蒙古 呼和浩特 010051福州大学 机械工程及自动化学院,福建 福州 350108

数理科学

纳米压痕人工神经网络遗传算法残余应力力学性能

nanoindentationartificial neural networkgenetic algorithmresidual stressmechanical properties

《四川轻化工大学学报(自然科学版)》 2026 (4)

14-28,15

国家自然科学基金项目(11872321)

10.11863/j.suse.2026.04.02

评论