Structural Damage Identification Based On Filter Technique

Because the design life of engineering structure is generally long,it will inevitably lead to local damage of structures under the influences of random disasters such as environmental erosion,vibration and fatigue.Damages within a small range can not cause the collapse of structure usually.However,with the accumulation of damages,the structural resistance and the life expectancy declines,leaving a hidden danger to the safety and the stability of engineering structures.So the Structural Health Mointoring(SHM)has been paid more and more attention in recent years.Damage identification is considered to be the core of the SHM,so researchers have focused a lot on the real-time online structural damage identification algorithms based on filter technique,such as the Extended Kalman Filter and the Particle Filter.Kalman Filter(KF)has achieved a great success in the field of aerospace applications since it was first published in the early 1960s.With further research,the KF technology applies more and more to various fields,such as the navigation guidance,the industrial controlling,the target tracking,the geodesy an finance,etc.Based on the traditional Extended Kalman Filter(EKF),this paper firstly studies its parameter identifications in different structures(single/multiple-Degree-Of-Freedom(DOF)structures or linear/nonlinear structures)and under different noise Gaussian or non-Gaussian noise)to understand its performance in civil structures.The results show that EKF identifies quite well in linear structures under the Gaussian noise,but not very well in nonlinear structures under the non-Gaussian noise.Although EKF can still be used in this situation,its indetification accuracy,will decrease.Especially when it comes to nonlinear structures and the non-Gauss noise,identification results by the EKF tend to be really terrible.For the study purpose of the structural nonlinearity,a bilinear model is used as an example to verify the general applicability of the Bouc-Wen model to the nonlinear model.The linear acceleration theory is introduced into the EKF process in this paper and an Improved EKF is put forward.Based on this new algorithm,parameters identification of a single-DOF linear structures under the Gaussian noise is finished.It is shown that the improved algorithm not only has similar recognition accuracy as the original algorithm,but also can reduce the complex derivation of the state transition matrix.To overcome some limitations of the EKF in time-varying systems,this paper introduces the Memory Attenuation EKF and performs parameter identifications in a single DOF linear structure under the Gaussian noise,which verifies the effectiveness of the Memory Attenuation EKF in time-varying system.Considering the poor performance of the EKF in the non-Gaussian case,this paper continues to study the Particle Filter(PF).The steady-state distribution likelihood value is introduced on the basis of predecessors' research to extend the Gaussian likelihood value.The validity and the applicability of the steady-state distribution likelihood value is verified by the parameter identification of single degree linear structures.At the same time,the recognition behavior of PF in different structures(single or multiple DOF linear structures)under different noise(the Gaussian noise or the non-Gaussian noise)is studied,and the identification results are compared with the traditional one.The simulation results show that the PF has obvious advantages in the non-Gaussian noise system.Finally,the EKF and the PF are applied to a five-layer steel frame parameter identification experiment,and the superiority of PF compared with the EKF is verified from the experimental point of view.