Research On The Characteristic Analysis And Identification Method Of Structural Health Monitoring Outliers

Structural health monitoring is essential for structural security,and it is also an important basis for the assessment and maintenance management of large structural state.The analysis of the abnormal value of monitoring data is one of the important tasks in the field of structural health monitoring.The distribution of abnormal data will not only reduce the efficiency of data analysis and cleaning,but also affect the accuracy of structural state assessment.At present,the identification and cleaning of abnormal values have received extensive research and attention in all major fields.This article focuses on the field of structural health monitoring,analyzes and combines the characteristics of structural monitoring data.From the three-way pair of three-way pairing of time,frequency domains,and space-time correlation The abnormal value recognition of structural monitoring data has been studied.The main research content is as follows:(1)Summarize the causes of the abnormal values of structural health monitoring,mathematical expression,time-frequency domain statistics,time and space-related characteristics,and commonly used identification and cleaning methods of outliers.The improvement of the existing abnormal value recognition methods was used to achieve preliminary cleaning of noise,jumping,drift,missing,and gain data,and summarized the various abnormal cleaning methods.Four types of abnormalities that use the existing method to clean the effect and can reflect the health state of the structure:jumping points,overall drift,gradual drift,and gain.(2)For the above four types of abnormalities,design the abnormal value recognition method based on the frequency-frequency domain feature extraction,and the time-frequency analysis has depicts the various types of abnormalities.Mathematical expression and time domain issues are mostly stayed;using features to train the SVM machine learning model with the Gauss nuclear function training to realize the identification and classification of the above four types of abnormalities.(3)Pay more attention to a single sensor for the existing abnormal identification methods,and ignore the correlation between monitoring data time and space;and the problem of the combination of monitoring data independence and space interoperability.Related features combine the time and space correlation of monitoring data,and realize the recognition and classification of autocorrelation abnormalities and interoperability abnormalities by improving the SVM model.(4)Putting a structural hierarchical warning method based on abnormal identification and monitoring data time and space correlation.This method uses the space-time correlation between monitoring data to make structural abnormal level determinations to make up for the lack of early warning based on a single sensor.The results of the eight comprehensive judgments are divided into three categories according to the abnormal space-time related characteristics.The results of each judgment were further explained and explained by the numerical examples,and the effectiveness of the warning method was verified by engineering instances.