Research On Key Technologies Of Bridge Health Monitoring Data Analysis Based On Deep Learning

With the wide application of bridge structure health monitoring technology,the structural damage identification based on deep learning has become the focus and difficulty of the research at home and abroad.In this thesis,based on the theory of deep learning technology,the data preprocessing in the field of bridge structural health monitoring and damage identification key technologies in-depth analysis and research,in view of the bridge structure monitoring data with temporal correlation properties,network(LSTM)model was proposed based on both short-term and long-term memory model of structural damage identification for bridge structural health monitoring analysis provides a new thought and method.The main research contents are as follows:(1)In-depth analysis and research on key technologies of bridge structure health monitoring data preprocessing.This thesis designs a data pretreatment scheme for the data analysis of the problem that the output response information of the structure is inconsistent and the data volume is large.First One of discrete data-Hot Ecoding normalization processing to eliminate the dimensional inconsistency problem between various attributes,and then carried out on the monitoring data of principal component analysis(PCA)dimensionality reduction to obtain effective characterization of structure state evolution characteristics of simplified eigenvalue,to refining to remove redundant features of huge amounts of data,improve the effectiveness of the monitoring data analysis.(2)The effectiveness of the bridge structure damage identification is analyzed deeply.On the one hand,a deep learning method based on multi-layer perceptron(MLP)neural network is designed and constructed and compared with the traditional machine learning method,SVM method has higher accuracy.On the other hand,the existing bridge structure damage identification method is not fully utilized to model the structural health monitoring data.In the neural network in this thesis,with the aid of LSTM has advantages of temporal correlation time series data,innovative design and build a suitable for long time series monitoring data characteristics of bridge structure damage identification based on LSTM neural network model,compared with when dealing with discrete,the temporal correlation of data has an advantage of higher accuracy MLP neural network.Finally,based on the simulation experimental data of the beili bridge,the experimental results show that:on the one hand,structural damage identification based on MLP improved the accuracy of the SVM method by 15.75%,onthe other hand,structural damage identification based on LSTM increased by 8.6%compared with MLP.(3)In order to verify the validity of the bridge structure damage identification model proposed in this thesis,the theoretical demonstration and the data of Bookshelf frame structure are compared and analyzed.The results show that,on the one hand,compared with the traditional machine learning method SVM,the accuracy of the structure damage recognition model based on MLP neural network increases by33.7%.On the other hand,by making full use of bridge structure monitoring data sequence relation of the proposed bridge structure damage identification method based on neural network LSTM,a structural damage identification in bridge of MLP neural network model,the accuracy of 8.4%.