基于机器学习的钢-UHPC组合结构栓钉抗剪承载力研究  

Research on the shear capacity of stud in steel-UHPC composite structure based on machine learning

作　　者：戚家南 杜雨轩 李立坤[1] 韩昀芝 李旻轩 衣忠强 邹伟豪 QI Jianan;DU Yuxuan;LI Likun;HAN Yunzhi;LI Minxuan;YI Zhongqiang;ZOU Weihao(Key Laboratory of Concrete and Prestressed Concrete Structures of Ministry of Education,Southeast University,Nanjing 211189,China;National Key Laboratory of Safety,Durability and Healthy Operation of Long Span,Southeast University,Nanjing 211189,China;Bridge Engineering Research Center of Southeast University,Nanjing 211189,China;China Railway Ninth Bureau Group No.1 Construction Co.,Ltd.,Suzhou 215538,China)

机构地区：[1]东南大学混凝土及预应力混凝土结构教育部重点实验室,南京211189 [2]东南大学长大桥梁安全长寿与健康运维全国重点实验室,南京211189 [3]东南大学桥梁研究中心,南京211189 [4]中铁九局集团第一建设有限公司,苏州215538

出　　处：《东南大学学报(自然科学版)》2025年第1期146-153,共8页Journal of Southeast University:Natural Science Edition

基　　金：国家自然科学基金资助项目(52378136);中铁九局科技研发资助项目(ZTJJYGSJSZX-202405-01)。

摘　　要：组合结构良好的界面抗剪性能是充分发挥钢和超高性能混凝土(UHPC)材料性能的关键。为研究钢-UHPC组合结构栓钉剪力连接件抗剪性能,建立了UHPC中包含263个栓钉的推出试验数据库,进行了特征相关性分析,并分析了主要参数影响规律。基于机器学习算法训练并生成了10个栓钉抗剪承载力预测模型,通过8次随机抽样训练验证模型稳定性。采用确定系数(R^(2))、平均平方对数误差(RMSE)、平均绝对误差(MAE)等指标来评估预测精度,并与3种传统计算方法进行了比较。结果表明,UHPC抗压强度小于150 MPa的试验样本占比高达64.6%,而栓钉直径小于19 mm的试验样本占比达到73%,缺乏更高强度UHPC和大直径栓钉试验数据。传统公式预测精度低于机器学习模型预测精度。随机森林模型预测精度最高,R^(2)为0.98,MAE与RMSE分别为3.9和6.4;与传统模型最高精度相比,R^(2)提高了216%,MAE与RMSE分别降低了93%与90%,表明机器学习模型可大幅提升预测精度,为栓钉抗剪承载力计算提供了新的思路。Good interfacial shear resistance of composite structure is the key to making full use of the proper-ties of steel and ultra-high performance concrete(UHPC).To study the shear resistance of the stud shear con-nection of steel-UHPC composite structure,the push-out test database of 263 studs in UHPC was established,the characteristic correlation analysis was conducted,and the influence law of main parameters was analyzed.Ten prediction models of the shear capacity of screws were trained and generated based on the machine learn-ing algorithm,and the stability of the models was verified by 8 times random sampling training.The indica-tors such as the determination coefficient(R^(2)),mean square error log(RMSE),and mean absolute error(MAE)were employed to evaluate the prediction precision,and were compared with those of the three tradi-tional computing methods.The results show that the proportion of test samples with compressive strength less than 150 MPa is as high as 64.6%,while the proportion of test samples with screw diameter less than 19 mm is 73%.There is a lack of test data for UHPC with higher strength and large diameter screws.The prediction accuracy of traditional formulas is lower than that of machine learning models.The prediction accuracy of ran-dom forests model is the highest,R^(2)is 0.98,MAE and RMSE are 3.9 and 6.4,respectively.Compared with the highest accuracy of the traditional model,R^(2)is increased by 216%,MAE and RMSE are reduced by 93%and 90%,respectively,indicating that the machine learning model can greatly improve the prediction accuracy and provide a new idea for the calculation of shear capacity of stud.

关 键 词：组合结构 超高性能混凝土 栓钉 抗剪承载力 机器学习 

分 类 号：TU398.9[建筑科学—结构工程]