| Because of the increasing construction of bridge engineering,there were unpredictable hidden dangers during the active service time,which led to the frequent occurrence of bridge safety accidents that constantly heave into people’s view.In order to ensure the safe service of bridge engineering,the research and application of damage identification method of bridge was the most important research for scientists and engineers at present.Aiming at the problems which current bridge damage identification method was difficult to extract the key features of damage information from the monitoring bridge response data,this paper presents a damage identification method for bridge based on CNN-LSTM architecture neural network to verify its effectiveness.Firstly,the paper analyzes and studies the theory of convolutional neural network(CNN)and short-term memory neural network(LSTM)in detail.Combining their strongpoints which they can extract features from data,and integrate the excellent characteristics of features,and apply them to damage recognition methods of bridge.Secondly,the model of equivalent element to reduce stiffness was used to simulate the damage crack of simply supported bridge with different damage degree,and the experimental data of acceleration signal trained by network were obtained by numerical simulation.Then the CNN and LSTM networks are designed and hybridized,and the network training and network testing of CNN-LSTM、CNN、LSTM were carried out by using the segmented acceleration signal data sets.CNN-LSTM damage identification of simply supported bridges was better than the other two models under various damage conditions.Then the damage recognition performance of the method was verified by simple-supported bridge experiment under moving load.At the end of this paper,the next research work of the damage identification method of bridge is planned to further verify its application performance in practical bridge engineering in the future. |
PDF Full Text Request