Sensor Fault Diagnosis Of Bridge Health Monitoring System Based On Principal Component Analysis

The sensor is the most advanced device of the bridge structural health monitoring system.When the sensor fails,the returned erroneous data may cause the structural safety status assessment to be missed or misreported,which will affect the final assessment result.Therefore,it is necessary to detect and isolate potential sensor faults before conducting a safety assessment.This dissertation is based on multivariate statistical analysis of process control theory and establishes a sensor fault diagnosis method based on principal component analysis(PCA).The cumulative contribution method and weighted statistics method are proposed to improve the recognition effect of the traditional principal component analysis method.In order to use the principal component analysis for non-linear,non-Gaussian stochastic systems,a kernel principal component analysis(KPCA)method is established.These improved diagnostic methods are applied to the actual bridge structural health monitoring system,and a time-division diagnosis method is proposed to accomplish the fault diagnosis of the acceleration sensor,which provides an effective means to ensure the normal operation of the system.The main contents and main conclusions are as follows:(1)Based on the principal component analysis principle,a sensor fault diagnosis method is established.Aiming at the deficiencies of the traditional contribution graphs in the fault location,a cumulative residual contribution rate method is proposed.Through the numerical simulation of a three-span continuous beam,the results show that the principal component analysis method can identify four typical faults of the sensor,and the cumulative residual contribution rate can not only better locate the single-sensor fault,but also accurately locate the faulty position when the two sensors fail simultaneously.(2)Aiming at the weakness of the traditional PCA that is insensitive to small-amplitude sensor faults,a weighted statistics method is proposed.Through the fault-sensitive factor,the sensitivity of each PCA direction identification fault is quantified,and the weighted fault detection statistic is determined.Combined with Bayesian inference and cumulative contribution rate positioning method,weighted principal component analysis(WPCA)method is established.Numerical simulation analysis results show that WPCA successfully detects and isolates the sensor’s bias fault and constant gain fault.Compared with the traditional PCA method,WPCA’s fault detection capability is significantly improved and it can identify smaller-scale faults.(3)Aiming at the fault diagnosis of nonlinear and non-Gaussian random systems,the kernel function theory was used to establish the KPCA fault diagnosis method.A new contribution indicator was used to establish the cumulative contribution rate and applied to KPCA’s sensor fault location.A seven-variable nonlinear stochastic system including polynomial nonlinearity,exponential nonlinearity,trigonometric nonlinearity and inverse function nonlinearity is established for numerical analysis.The results show that the KPCA method has better fault diagnosis capability than PCA in nonlinear and non-Gaussian systems,so that it can identify and isolate small-scale faults in time.(4)Three fault diagnosis methods respectively based on PCA,WPCA and KPCA are applied to the Dongshuimen Bridge structural health monitoring system to achieve multi-sensor fault self-diagnosis.Considering the differences in working conditions at different times of the day,a time-division diagnosis method is proposed.The final diagnosis shows that the acceleration sensor is in good condition,which is consistent with the result of manual periodic inspection.The applicability of the sensor fault diagnosis method based on principal component analysis in the actual health monitoring system is preliminarily verified.Artificially introduced bias faults,constant gain faults and stuck faults,and used the three methods mentioned above for diagnosis.The results show that: all three diagnostic methods can identify the fault and locate the fault sensor of the bridge accurately.Both the WPCA and KPCA methods improve the fault detection effect of the traditional PCA method.