为对框架结构柱损伤位置和程度进行识别,应用均匀设计方法构建可用于结构损伤正分析档案库的试验表,按此试验表对抽样出的损伤位置和程度进行BP神经网络训练. 训练后的神经网络根据分步式识别方法,以固有频率对结构出现的损伤进行层定位,再以位移变化率对损伤进行构件定位,同时反馈获得构件的损伤程度. 通过2个模型结构损伤识别的算例,分别验证了所构建神经网络和识别参数的正确性,以及基于均匀设计试验分步式损伤识别方法的有效性. 计算结果表明均匀设计试验可在保证识别效果的同时,大幅减少神经网络的训练次数;分步识别方法可对框架结构损伤位置进行精确判定,克服了一步识别框架损伤或单独使用某种参数进行损伤识别的缺陷.%In order to detect damage locations and damage extent of the frame column accurately, the paper introduces a uniform design method to establish the experiment table, which can be used in the forward analysis about damage identification, and then trains BP neural network by the damage locations and extent sampled from the experiment table. According to the concept of the multi-hierarchical damage identification method, the neural network could identify the damage floors by natural frequency. Then, the variation rule of displacement is applied to detect the damage members and their damage extent. Through two examples of structural damage identification and the correctness of the neutral networks, the identification parameter is proved, and the effectiveness of the uniform design testing multi-hierarchical damage identification methods are validated as well. The results show that the uniform design experiment uses less training time of the neural networks to obtain the similar detecting performance; the multi-hierarchical identification method can accurately capture the damage member of the frame structures, which avoids defects of the single step or single parameter identification method.
展开▼
