基于XGBoost模型的多缺陷点阵结构性能参数预测方法  

Multi-defective Lattice Structure Performance Parameter Prediction Method Based on XGBoost Model

作　　者：张芝威 张玉燕[1] 温银堂[1] 任亚雪 ZHANG Zhiwei;ZHANG Yuyan;WEN Yintang;REN Yaxue(School of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004

出　　处：《计量学报》2024年第4期559-564,共6页Acta Metrologica Sinica

基　　金：河北省科技计划(216Z1704G,20312202D,20311001D,20310401D)。

摘　　要：针对增材制造金属点阵结构缺陷多样性和随机性导致其力学性能难以准确预测问题,提出一种基于参数优化XGBoost模型的点阵结构屈服应力预测模型。基于CT测试数据,提出一种阈值判断准则,用于多种缺陷快速检测和分类。重点研究了支柱直径尺寸和角度偏差缺陷问题,分别以支柱过大、支柱过细、支柱弯曲为主的缺陷数量作为模型输入特征;考虑到缺陷位置具有随机性,将点阵结构层数作为另一种输入特征。根据实测数据建立含缺陷的点阵结构仿真模型,对其屈服应力进行仿真。利用参数优化方法优化预测模型超参数并结合屈服应力仿真结果,实现对结构件屈服应力的预测,其决定系数R 2为0.81,均方根误差为18.87。模型预测结果与实际样件实验测试结果对比,其偏差度为7.6%,表明了所建预测模型的有效性。To address the problem that the diversity and randomness of defects in additively manufactured metal lattice structures make it difficult to accurately predict their mechanical properties,a model for predicting the yield stress of lattice structures based on the parameter optimization XGBoost model was proposed.Based on CT test data,a threshold judgement criterion was proposed for the rapid detection and classification of multiple defects.Focusing on the problems of strut diameter size and angular deviation defects,the number of defects dominated by oversized struts,undersized struts,and waviness struts were used as input features of the model,respectively.The number of layers was used as another input feature,considering the randomness of defect locations.Based on the measured data,a simulation model of the defect-containing lattice structure was established,and its yield stress was simulated.Using the parameter optimization method to optimize the prediction model hyperparameters and combining the results of the yield stress simulation,the prediction of the yield stress of the structural member is achieved with an R 2 of 0.81 and an RMSE of 18.87.A comparison between the prediction results and the actual sample test results was provided in the experimental section,and the deviation is 7.6%,which shows that the prediction model can accurately predict the yield stress of the structure.

关 键 词：力学计量 屈服应力预测 XGBoost 固有缺陷 金属点阵结构 

分 类 号：TB931[一般工业技术—计量学]