Bridge SHM Data Anomaly Detection And Recovery Based On Deep Learning

Through the real-time monitoring data of structural health detection system(SHM),the safety status of the structure can be detected in time.However,sensor failure,loose interface and electromagnetic interference will cause abnormal data collected by the structural health system,and the abnormal data will seriously interfere with the analysis and evaluation of the structure.Therefore,it is of great significance to explore a method that can accurately diagnose and repair abnormal data.Starting from the actual needs,this paper studies the basic problems of abnormal diagnosis and repair of monitoring data.The main research work of this paper is as follows:(1)Based on the actual monitoring data,the common monitoring data types are classified and analyzed.On the basis of wavelet transform and Hilbert transform,this paper studies how to extract the accurate features of wavelet transform and Fourier transform,respectively.(2)Aiming at the problem of bridge monitoring data anomaly detection,a data anomaly detection algorithm based on dual channel convolution neural network is proposed in this paper.A two-channel convolutional neural network model is designed and built.The convolutional neural network model has a two-dimensional convolutional neural network(2d-cnn)channel for convoluting the time-frequency map and a one-dimensional convolutional neural network(1d-cnn)channel for convoluting the original one-dimensional vibration time series data.It can input the twodimensional time-frequency map of acceleration data and the original one-dimensional vibration time series data in parallel.Finally,an accurate classification model is obtained by training a large number of data sets.In the process,six training conditions are designed to study and analyze the imbalance of training data.Among them,the accuracy of the optimal model is 95.10%.(3)On the basis of correctly identifying data anomalies,aiming at the repair problem of bridge monitoring data anomaly detection,this paper adopts a data anomaly repair method based on long short term memory(LSTM).This method uses the time series prediction function of long short term memory network to transform the complex and changeable data anomaly repair problem into time series data prediction problem.The results show that under the appropriate repair length,the trend of the measured data and the repaired data is the same,the oscillation is the same,and there is only a slight difference in the amplitude.In addition,the random subspace method is used to identify and compare the vibration modes and modal frequencies of the original data and the repaired data.The results show that the identification results of the repaired data and the original data are basically consistent under the appropriate length.It is proved that this method has certain data repair ability.