基于双层深度置信网络的梁桥结构损伤识别方法研究OACSTPCD

To accurately and efficiently identify structural damage in bridges,we propose a meth-od based on a double-layer deep belief network(DBN).This approach combines deep learning with structural dynamic characteristics of structural engineering.First,the initial three vertical vibration frequencies of the structure,along with the first three vertical vibration modal displace-ments of midspan nodes,are taken as parameters.These parameters serve as the input data for the first-layer DBN to identify the damage location of the structure.Following this,the differ-ences in the modal displacement of the first-order vertical vibration are taken as parameters.These are then used in the second-layer DBN to predict the extent of the structure damage.As a case study,we applied this method to the Zhengzhou-Xuchang suburban railway bridge.The calculation results show that when the error is not considered,the results of the structural dam-age identification method based on the double-layer DBN are precise.When the noise level does not exceed 10％,the accuracy of the location identification results is 100％.Even when the noise level does not exceed 15％,the maximum absolute error of quantitative identification results is not larger than-1.15％.Compared with the traditional BP neural network method,the proposed method demonstrates higher recognition accuracy and a stronger capability to resist noise.