基于灰度分析的桥梁健康监测系统传感器异常检测

Sensor Anomaly Detection in Bridge Health Monitoring System Based on Gray Correlation Analysis

为了能够及时检测出传感器的异常状况,设计了一种基于灰色关联度分析的桥梁健康监测系统传感器异常检测模型.首先,分别对多个应变传感器在正常工作和单传感器异常时采集的数据进行灰色关联度分析,得到表征每一列传感器数据与其余列数据在序列几何形状上相似程度的最不相关次数序列.通过对比发现无异常和有异常时最不相关次数的分布呈现明显差异,从而验证了此方法的可行性;然后,设计了一种权重计算策略,将最不相关次数序列转化为归一化的值,并将其作为评价指标,以此来量化每一列传感器数据与其他列数据之间的相关程度;最后,通过分析多组应变数据的评价指标,设置了多阈值预警机制,以实现对不同程度的异常情况做出相应的判定.在另一组加速度监测数据上模拟多种异常程度不同的异常状况并进行检测,结果显示整体的异常识别率在90%以上.

In order to be able to detect the abnormal condition of sensors in time,a sensor abnormality detection model for bridge health monitoring system based on gray correlation analysis is designed.First,the datas collected from multiple strain sensors in normal operation and single sensor abnormality are analyzed by gray correlation analysis,respectively,and the least correlated count columns that characterize the degree of geometric similarity between each column of sensor data and the rest of the columns of data are obtained.The comparison revealed that the distribution of the least correlated counts when there was no abnormality and when there was an abnormality showed marked differences,thus verifying the feasibility of this method;then,a weight calculation strategy was designed to transform the least correlated count columns into a normalized value and use it as an evaluation index,which quantified the degree of correlation between each column of sensor datas and the rest of the columns of datas;finally,through the analysis of the evaluation indexes of multiple sets of strain datas,the multi-threshold warning mechanism was set up to realize the corresponding determination of different degrees of sensor anomalies.On another set of acceleration monitoring datas to simulate a variety of degrees of anomalies and detection,the results show an overall anomaly recognition rate of over 90%.