【摘要】 桥梁结构的健康监测中,由于材料参数、建模误差及环境噪声等不确定性的影响,传统的损伤识别方法存在损伤识别精度低和鲁棒性不足等问题。针对这一问题,提出基于数据与模型双驱动的两阶段损伤识别方法,旨在提高损伤识别的精度和噪声鲁棒性。该方法依托物理引导神经网络(PGNN)对初始有限元模型进行修正,通过引入基于模态灵敏度分析的物理损失函数,指导神经网络优化模型修正过程,从而使修正后的模型能更准确地反映结构的动力响应。进一步,基于修正后的模型,结合卷积神经网络(CNN)和交叉模型交叉模态(CMCM)算法进行结构损伤识别。通过I40钢-混组合桥梁验证数据与模型双驱动的钢-混组合桥梁的两阶段损伤识别方法的有效性,结果表明:基于建立的精准有限元模型,在不同工况下损伤识别结果误差最大不超过7%。更多还原

【Abstract】 Due to the existence of the uncertainties,such as material parameters,modeling errors and environmental noise,traditional damage identification methods often suffer from insufficient accuracy and robustness.To address this issue,a two-stage damage identification method based on ’ data and model’ dualdrive was proposed,aiming to improve the accuracy and noise robustness of damage identification.In the first stage,a Physics-Guided Neural Network(PGNN) was employed to update the initial FEM(finite element model).By incorporating a physics-based loss function derived from modal sensitivity analysis,the network was guided to optimize the model updating process,leading to a more accurate dynamic response of the structure.In the second stage,based on the modified model,structural damage identification was performed using a combination of convolutional neural networks(CNN) and cross-model cross-mode(CMCM) algorithms.The effectiveness of the proposed method was validated through a case study on an 140 steel-concrete composite bridge.The results demonstrate that,based on the accurate FEM,the maximum error in damage identification under various conditions remains within 7 %.更多还原