融合机器学习的空间结构智能损伤识别理论研究

Research on the theory of intelligent damage recognition of spatial structure based on machine learning

为应对大跨空间网架损伤定位中动力指纹同损伤情况之间非线性映射关系难以求解的问题,依托实际工程模型验证了一种基于径向基函数(radial basis function,RBF)神经网络的改进空间结构损伤识别方法。该方法首先建立初始损伤工况同动力指纹数据间的映射,通过对比改进前后对验证工况下实测数据对应杆件的损伤指数识别,发现此方法能够提高网架杆件损伤预测结果10%左右的精确度,从而更快更准确地解决大跨空间结构损伤定位问题。

In order to solve the problem that it is hard to solve the nonlinear mapping relationship between the dynamic fingerprint and the damage situation in the damage localization of the large-span space grid,an improved spatial structure damage identification method based on radial basis function(RBF)neural network is verified based on the actual engineering model,which first establishes the mapping between the initial damage case and the dynamic fingerprint data.By comparing the damage index identification of the corresponding rods under the measured data under the verified working conditions before and after the improvement,it is found that this method can improve about 10%accuracy of the damage prediction results of the grid members,so as to solve the problem of damage localization of large-span spatial structures faster and more accurately.