论文部分内容阅读
通过理论推导得到了模型参数误差对损伤引起模态参数改变的贡献的表达式，用该式可指导神经网络输入参数的选择和输入向量的构造。理论分析表明，适当地构造输入向量，可以减小模型参数误差对结构损伤识别的影响。在采用BP网络和合适的输入向量后，还用数值模拟的方式对一榀六层框架的损伤识别进行了确定性研究和概率分析，结果表明，用神经网络进行结构损伤识别，受模型参数误差的影响很小，在训练神经网络时，10％的模型参数误差是可以接受的。最后，用一个两层钢框架的实验数据验证了神经网络在有模型误差时的识别能力。
The expression of the contribution of model parameter error to modal parameter changes caused by damage is derived theoretically. This formula can guide the selection of input parameters and the construction of input vector. Theoretical analysis shows that properly constructing input vectors can reduce the impact of model parameter errors on structural damage identification. After adopting BP network and appropriate input vector, the damage identification of one-six-frame frame is also confirmed by numerical simulation and probability analysis. The results show that the structural damage identification by neural network is affected by model parameter error Of the impact is very small, 10% of the model parameter error is acceptable when training the neural network. Finally, the experimental data of a two-story steel frame is used to verify the recognition ability of the neural network when there is model error.