【摘要】 建立钢桁梁桥的ANSYS有限元模型,将钢桁梁桥按几何位置划分为4个子结构,通过概率神经网络将螺栓脱落损伤定位在子结构内。通过降低杆端单元的抗弯刚度来模拟螺栓的脱落损伤,并选择曲率模态变化率作为概率神经网络的输入参数。通过数值模拟,分析不同训练样本下概率神经网络对螺栓脱落损伤的定位识别结果。结果表明:Spread值对概率神经网络的识别正确率影响很大,须通过试算合理取值;训练样本数量相同时其损伤程度对概率神经网络识别结果的影响很小;使用足够多的单损伤样本对概率神经网络进行训练后,对单损伤和多损伤定位识别的正确率均高达90%以上。更多还原

【Abstract】 The ANSYS finite element model of the steel truss bridge was established,which is divided into four substructures according to the geometric position.The bolt looseness damage was located in the substructure through the probabilistic neural network.The bolt looseness damage was simulated by reducing the bending stiffness of rod end element,and the curvature mode change rate was selected as the input parameter of probabilistic neural network.Through numerical simulation,the location and identification results of bolt looseness damage based on probabilistic neural network with different training samples were analyzed.The results show that the Spread value has a great influence on the recognition accuracy of probabilistic neural network,and it must be reasonably selected through trial calculation.When the number of training samples is the same,the damage degree has little effect on the identification result of probabilistic neural network.When enough single damage samples are used to train probabilistic neural network,the accuracy of single-damage and multi-damage location and recognition can reach more than 90%.更多还原