Loading...

Application of Kalman Filters for Dynamic Control of Beams Subjected to Moving Load and Mass

Moradi, Sarvin | 2023

84 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56151 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Mofid, Masood; Eftekhar Azam, Saeed
  7. Abstract:
  8. In this study, for the first time, a comprehensive and online framework for active control of beams subjected to the moving mass is presented. In active control problems, comprehensive knowledge of system states is required to determine the control force, but it is not possible to measure all system states in practice with a limited number of sensors. In this regard, Kalman filters are introduced to control systems to help improve the performance of control systems as observers of system states. However, in the case of beams subjected to the moving mass, due to the moving nature of the passing mass, in addition to system states, it is also important to identify the input load. In previous studies, all the characteristics of the passing mass such as mass, speed, arrival time, etc. have been assumed known, which in practice is not a logical assumption. In this research, for the first time, advanced Bayesian filters with the ability to estimate the input-state in active control systems have been used. In this regard, three advanced Bayesian filters, namely Augmented Kalman Filter (AKF), Dual Kalman Filter (DKF) and Joint Input-State estimator (JIS) with the aim of estimating input and state, have been introduced to the classical control algorithm. Thus, a comprehensive online framework for active control of beams subjected to the moving mass was formed; the only inputs of the framework are sparse noisy measurements of sensors. Extensive numerical studies have confirmed the efficiency of the proposed comprehensive framework in controlling the beams in the different sensors arrangement, variable noise levels of the sensors, as well as the different passing mass velocities. It was also shown that the application of these advanced filters in comparison with standard Kalman filters, increases the reliability of active control systems, especially in the case of beams subjected to the moving mass. In the next step, the newly introduced Physics-Informed Neural Networks (PINN) were studied to improve the performance of the proposed framework. Unlike Kalman filters, in which estimates are one-step ahead predictors, these networks can estimate system states for a longer period of time if properly trained. Thus, if the sensor measurements become unavailable, the system will continue to operate. These neural networks can also use sensor measurements to estimate the dynamic characteristics of the beam as well as the input load. In this way, more accurate models can be used in Kalman filters, which is effective in improving their performance. Finally, due to the fact that these networks have not been used in the analysis of beams before, two new architectures were introduced to increase the efficiency of the PINNs. The performance of the PINN with these two new architectures was compared with artificial neural networks and their superiority in estimating states, system characteristics and input load was confirmed. Finally, the application of these networks in the development of digital twin models throughout the life of the structure was investigated
  9. Keywords:
  10. Active Control ; Moving Mass Load ; Recursive Bayesian Filters ; Phyiscs Informed Neural Networks (PINN) ; Kalman Filters

 Digital Object List

 Bookmark

No TOC