Online Health Monitoring of Nonlinear Hysteretic Structures Using System Identification Techniques and Signal Processing Tools

Amini Tehrani, Hamed | 2020

450 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 53412 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Bakhshi, Ali
  7. Abstract:
  8. Adverse social and economic effects of earthquakes have necessitated the emergence and development of efficient methods to assess and monitor the health status of structures. Many of the structural health monitoring algorithms are based on linear models that are not able to provide sufficient dynamic information. Nonlinear models require monitoring of a larger number of structural parameters and provide a much closer to reality model of the structure. Therefore, the use of nonlinear models in the identification process provides more useful information about the safety and serviceability of post-earthquake structures. Also, most of the existing methods are not applicable for online health monitoring of structures and therefore, their results will not be available immediately after a serious event with the aim of rapid decision-making and prioritization of actions necessary to manage the crisis. During moderate to severe earthquakes, a significant number of structures will experience nonlinear deformations. In such situations, structural damping and stiffness identification before and after earthquake do not suffice for structural health assessment, and it is also necessary to consider phenomena such as reduction in energy dissipation capability, stiffness and strength degradation, pinching effect, and permanent plastic deformations to provide a comprehensive health condition assessment.
    The aim of this dissertation is to real-time health monitoring of structures with deteriorating nonlinear hysteretic behavior using Bayesian inference framework. For this purpose, joint estimation algorithms based on sigma-point Kalman filters have been developed with the aim of optimizing the unknown parameters of nonlinear models during seismic excitation. One of the most important features of the proposed hybrid methods is the ability to update the parameter noise covariance matrix at each time step, so that it prevents the occurrence of non-convergence and significant reduction in the accuracy of parameters estimation, particularly in high-dimensional identification problems. Also, Bayesian multiple-model approach, which is a suitable tool for optimal model class selection, has been used in this research mainly for improving data fitting precision, decreasing dimensions of structural unknowns vector through removing unnecessary parameters, detecting the occurrence and type of predominant phenomenon related to degradation in structural nonlinear hysteretic behavior, and finally increasing stability and convergence rate of the utilized identification algorithm. According to the obtained results, it can be concluded that in addition to state estimation, unknown parameters identification, tracing the degrading and pinched hysteresis loops, and robustness against the noisy input and measured response signals, the proposed algorithms are highly reliable in terms of estimation accuracy, convergence rate, and stability. Therefore, they can be used in the real-time health monitoring of structures after serious events such as moderate to severe earthquakes. In addition to the above, the simulation results showed that the proposed joint estimation algorithms have adequate accuracy, efficiency and robustness in identification problems with a large number of unknowns (high-dimensional augmented state vectors).
  9. Keywords:
  10. Structural Health Monitoring ; Online Monitoring ; Hysteresis Behavior ; Kalman Filters ; Bayesian Estimation ; Deteriorating Nonlinear Hysteretic Behavior ; Bayesian Multiple-Model Approach ; Joint Estimation

 Digital Object List