采用机器学习方法评估流线型箱梁颤振临界风速  

Predicting critical flutter wind speed of a streamlined box girder by using machine learning

作　　者：梅瀚雨 王骑[1,2] 廖海黎[1,2] 刘珉巍 MEI Hanyu;WANG Qi;LIAO Haili;LIU Minwei(Department of Bridge Engineering,Southwest Jiaotong University,Chengdu 610036,China;Wind Engineering Province Key Laboratory,Southwest Jiaotong University,Chengdu 610036,China)

机构地区：[1]西南交通大学桥梁工程系,成都610036 [2]西南交通大学风工程四川省重点实验室,成都610036

出　　处：《振动与冲击》2021年第14期195-202,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金(51778547;51678508)。

摘　　要：为在工程设计初步阶段快速评价流线型箱梁颤振性能,基于4种机器学习算法(支持向量机回归、神经网络、随机森林回归和高斯过程回归),利用自由振动风洞试验获取的15种不同流线型箱梁断面在5种不同风攻角下的颤振临界风速,建立了从气动外形与动力参数到颤振临界风速的颤振临界风速预测模型。结果表明:支持向量机回归模型预测结果整体精度最高,且在实际桥梁上表现性能较好;神经网络模型较差,但其相对误差仍远低于JTG/T 3360-01-2018《公路桥梁抗风设计规范》中提供的3种不同颤振临界风速简化计算公式的结果。该研究的结果符合预期要求,未来可进一步扩充数据集,以期成为对抗风设计工作者提供高精度颤振性能评估的有力工具。To quickly evaluate the flutter performance of a streamlined box girder in the preliminary stage of bridge design,critical flutter wind speeds of 15 different sectional models at 5 different angles of attack were tested by free vibration tests in wind tunnel and utilized to build flutter prediction models by inputting dimensional information and dynamic parameters based on four different machine learning algorithms,including support vector regression,neural network,random forest regression and Gaussian process regression.The results show that the support vector regression model has the highest prediction accuracy and get the best effect performance in the prediction of some practical bridges while the neural network model is poor,but its relative error is still far lower than that of the calculation results of three different simplified critical flutter wind speed formulas provided in the present JTG/T 3360-01-2018 Wind-resistant Design Specification for Highway Bridges.The results of the paper meet the expected requirements and the dataset can be further expanded in the future so as to provide a powerful tool for wind designers to evaluate the flutter performance of streamlined box girders with high-precision.

关 键 词：流线型箱梁 机器学习 颤振 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3