Loading...

PID Auto-Tuning and Flight Trimming for the Quadcopters by Deep Reinforcement Learning and Digital Twin Technology

Taheri, Sara | 2023

66 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56012 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Rezaeizadeh, Amin
  7. Abstract:
  8. In recent years with the development of industrial automation, artificial intelligence, and the coming of the fourth-generation industrial revolution, making robots intelligent and solving old challenges in non-linear systems in order to make human life easier has always been the point of view of researchers. Flight systems are always considered to be one of the most nonlinear systems. In linear systems, a PID controller can be easily designed and implemented; But in non-linear systems, because the set point is constantly changing, the PID gain must be manually adjusted by the operator, which is almost impossible in practice; Therefore, finding a fast and efficient way to solve this challenge is an important issue in the control engineering of complex systems. Several pieces of research have been conducted for online and automatic PID gain adjustment of flying robots such as quadcopters, but they are mostly not applicable in the real world. In addition, in these systems, one of the most common errors is the lack of balance at the start of the flight. In this case, pilots should manually trim the system based on trial and error at the beginning of the flight and adjust them until the outputs are completely balanced. However, manual trimming can always be a challenging problem. Accordingly, recent attempts have been made to automate it by removing unbalanced variables with autopilot systems. Another challenging problem in nonlinear systems is the design and tuning of the PID controller. In this research project, using deep reinforcement learning and digital twins, two methods for real-time and automatic adjustment of PID controllers and trimming of remote control quadcopters have been developed
  9. Keywords:
  10. Deep Learning ; Reinforcement Learning ; Quadcopter ; Proportional-Integral-Derivative (PID)Controller ; Autopilot ; Remote Control ; Flight Trim ; Digital Twin

 Digital Object List

 Bookmark

No TOC