广西大学学报（自然科学版）

为了提升简支梁桥状态评估效率和评估智能化水平，提出基于准静态响应的桥梁正常使用状态的快速智能评估方法。首先，以车辆信息监测和关键截面的挠度监测为基础，建立轻量化监测体系；其次，建立以车辆信息和截面响应为输入、准静态挠度响应为输出的3种机器学习模型，构建桥梁响应有限元代理模型；再次，提出基于逐次变分模态分解(SVMD)的准静态挠度曲线自适应提取方法，通过迭代分解、残差评估，自动确定最优模态分解个数；最后，通过损伤桥梁和未损桥梁的准静态响应曲线面积比，构建评估系数β和桥梁状态分级表，并通过简支箱梁桥的数值算例进行验证。结果表明：基于反向传播(BP)神经网络构建的有限元代理模型训练结果最优，决定系数(R²)为98%、均方根误差(RMSE)为0.006 9;通过SVMD提取出的准静态挠度响应整体相对误差均小于2%,峰值相对误差均小于1%;将模型刚度折减成设计值的88%时，计算简支箱梁桥的β值为0.88,通过桥梁状态分级表评估桥梁的状态为一般，与设置的损伤状况一致，可初步判断桥梁的正常使用状态。

To enhance the efficiency and intelligence of the condition assessment for simply supported beam bridges, a rapid and intelligent evaluation method for the normal service condition of bridges based on quasi-static responses was proposed. First, a lightweight monitoring system was established based on vehicle information monitoring and deflection monitoring of key sections. Second, three machine learning models with vehicle information and section responses as inputs and quasi-static deflection responses as outputs were constructed to build a finite element surrogate model for bridge responses. Third, a quasi-static deflection curve adaptive extraction method based on SVMD(sequential variational mode decomposition) was proposed, which automatically determined the optimal number of modal decomposition through iterative decomposition and residual evaluation. Finally, an assessment coefficient β and a bridge condition classification table were constructed based on the area ratio of quasi-static response curves of damaged and undamaged bridges, and verified through a numerical example of a simple supported box girder bridge. The results show that the finite element surrogate model constructed based on BP(back propagation) neural network has the optimal training results, with an R²(coefficient of determination)of 98% and an RMSE(root mean square error)of 0.006 9; the overall relative error of the quasi-static deflection response extracted by SVMD is less than 2%, and the peak relative error is less than 1%; the β value of the simple supported box girder bridge with a stiffness reduction of 88% of the design value is 0.88, and the bridge condition is assessed as general through the bridge condition classification table, which is consistent with the set damage condition, and can preliminarily determine the normal use condition of the bridge.