Alessandro Puglisi

Alessandro Puglisi

PhD Candidate in Underwater Robotics. Enhancing underwater systems' perception for exploration and environmental monitoring

Centralized, Decentralized and Distributed Control of a Three-Tanks System

Completed - 2023 | MATLAB - Simulation - Decentralized Control - Control tuning and optimization

GitHub Repo


Final group project assignment of a course on Networked Control systems.

In this project, we solve a benchmark problem in control theory, the (linearized) modeling and control of a system of three tanks in series. In particular, coherently with the course topics, we implement different control schemes (centralized, decentralized, distributed). Simulation and performance analysis are done both in continuous-time and discrete-time domains.

Summary:

  • Modeled and Simulated a Three Tanks system dynamics in MATLAB.
  • Implemented and Tuned Multi‐Variable Control structures, using Linear Matrix Inequalities solved with YALMIP optimization toolbox, making the system five times faster and eliminating water oscillations.
  • Improved controller performance with a rate limiter, reducing control action peak by a factor of ten, without losing system speed.

Given the state-space model of a three-tank system:

Objective:

  • Model the system in MATLAB and analyze its stability both in continuous and discrete time.
  • Use the MATLAB optimization toolbox ‘yalmip’ to solve linear matrix inequality (LMI), and tune different controller structure (in continuous and discrete time):
    • centralized
    • decentralized
    • different distributed schemes
  • Simulate the system. Analyze the performance and provide graphical results and quantitative proof of the results’ quality.

Original Contribution: Because the control action often performs unrealistic steps, control rate limitations have been included (slides 69-82). To do so, a novel idea has been implemented: Defining an extended system model and new LMIs, with a cost function to limit control input magnitude and rate. With this approach, I obtain a 100% more realistic control action.

Final Results:

  • Speed up the system, making it two times faster than the open loop dynamic.
  • Removed any waves (oscillations and overshoots) during settling time.
  • Found the best choice of sampling time that doesn’t degrade performances.
  • Provided a realistic interpretation of the results, considering the physical limitations of a real system.