组合神经网络在结构损伤检测中的应用

子结构的动态响应变化与整体结构相比，对结构内部损伤反应更为敏感。组合神经网络可以克服单个神经网络功能的单一局限性，实现更加全面综合的仿真识别功能。本文首先运用双协调自由界面模态综合法对结构进行模态分析，获取各子结构及整体结构的模态信息。然后，通过组合BP神经网络将损伤子结构与整体结构的模态频率变化率组合起来进行结构损伤检测。该方法在改善网络训练性能的同时，提高了检测结果的准确性和可靠性。文章最后通过数值算例验证了该方法的可行性和有效性。

Application of the combined neural network in structural damage detection

Comparing with the entire structure system, the dynamic property of the substructure is more sensitive to the structural damage. The combined neural network may overcome the limitation of the single neural network in simulation and identification performances, and has the advantage of implementing more comprehensive function. The author first used the bi-compatible free-interface component mode synthesis (CMS) method to do modal analysis. After obtaining the modal data of each substructure and the entire structure, the frequency change rates of the damaged substructure were used together with that of the entire structure system to participate in the designing of combined BP neural network system for damage detection. This scheme can effectively increase the detection precision and improve the network performance. In the end, numerical examples demonstrate the feasibility and effectiveness of the method.