基于曲率模态和神经网络的分步损伤识别法及其在桥梁结构中的应用

神经网络用于损伤识别遇到的最大问题就是训练样本数量的组合爆炸问题，单纯用神经网络进行损伤诊断有很大困难．提出了一种两步识别法来进行损伤诊断，即先采用结构的曲率模态，定义一个新的损伤指标，判断损伤位置，再利用BP神经网络精确识别损伤程度；运用两步识别法对一座混凝土连续刚构桥进行了损伤位置与损伤程度的识别．识别结果表明，对于2个单元和3个单元损伤的情况，分别只需16个和64个损伤样本就能取得满意的识别结果，大大减少了单纯利用神经网络进行损伤识别所需的损伤样本．

Two-step damage identification method and its application based on curvature mode and neural network

The training samples applied to damage identification of structures by adopting neural network are extremely numerous. It is difficult for the neural network method to be applied alone to structures. Hence, two-step damage identification method has been developed in this work. In the first step, the damage locations are detected by means of a new damage index in curvature mode method. Then, the damage degree can be accurately identified by applying the BP neural network. Some damaged places are assumed in a continuous rigid frame bridge. The new method can be successful to identify the damage locations as well as the damage degree. The result shows that only sixteen samples are needed when there are two damaged units and only sixty four samples are needed when there are three damaged units. Two-step method needs much less training samples than using neural net-work alone.