Research On Anomaly Diagnosis And Repair Methods Of Structural Health Monitoring Data

Structural health monitoring systems generally contain multiple types of sensor arrays.Through the data obtained by the sensors,the safety status of the structure can be understood in a timely manner.However,in the monitoring system,most of the sensors and monitoring substations are installed outdoors to work.Sensor failures,weak interfaces,broken transmission cables,electromagnetic interference,and equipment failures often cause abnormal monitoring data,causing data analysis and structural safety.The evaluation work is not accurate,so it is of great significance to diagnose and repair abnormal data.At present,most methods of data exception processing use fixed signal threshold to screen abnormal data or simply eliminate noise,etc.,which requires manual selection of a section of data for calculation and analysis,which is inefficient and prone to omission and misreporting of abnormal data.Therefore,it is of great significance to diagnose and repair abnormal data efficiently and accurately.Based on the processing requirements of monitoring data,this paper studies the basic problems of abnormal diagnosis and repair of health monitoring data.The main research work of this paper is as follows:(1)According to the monitoring data processing requirements of actual engineering structures,the common monitoring data formats are summarized and classified.The format of monitoring data will be standardized,based on the unified data format,conveniently data retrievaled and used.Through data visualization,the monitoring data is transformed into images,and based on data visualization,the classification standard of monitoring data images is proposed.According to the characteristics of the data and images,the monitoring data are divided into five categories: normal data,local gain data,missing data,outlier data and drift data,and the data images are manually marked for image retrieval and post-processing.(2)Aiming at the problem of abnormality in the monitoring data,using deep learning related knowledge to establish an efficient and accurate monitoring data abnormality diagnosis model;the model is based on the monitoring data image,and the image is input into the convolutional neural network for image feature extraction and image classification to realize abnormal diagnosis of the monitoring data image.By introducing transfer learning,the abnormal diagnosis model of monitoring data can be updated in real time.(3)For the abnormal data identified by the abnormal diagnosis model of monitoring data,the information of the time period of abnormal data was recorded,and the quality analysis of diagnosis results was given based on this.The identified abnormal data were repaired,and based on the linear regression model and autoregressive model,the missing data was repaired;the outlier point data were eliminated based on the triple standard deviation method and the method of eliminating the maximum value;the drift data is repaired based on the mean translation method.(4)In order to meet the needs of automatic abnormal diagnosis and repair of monitoring data,a software for abnormal diagnosis and repair of structural health monitoring data is developed based on MATLAB GUI platform.The GUI interface is designed to support users to input the basic information of monitoring data,carry out abnormal diagnosis of monitoring data,update the diagnosis model of migration learning and repair abnormal data,and finally export the quality analysis report of abnormal diagnosis result of monitoring data and abnormal data after repair.(5)According to the acceleration data of a bridge structure health monitoring,the basic information of the bridge acceleration data and the basic parameters of deep learning were imported based on the abnormal processing toolbox of monitoring data,and the abnormal diagnosis and repair of monitoring data were realized.