Hebaish: Design and implementation of online identification algorithms and discrete controller on reconfigurable systems

27.08.2019, 14.00 Uhr, APB 1096

Invitation to the presentation of the bachelor thesis of Marawan Azmy Hebaish

Subject: “Design and implementation of online identification algorithms and discrete controller on reconfigurable systems”

Supervisor: Gökhan Akgün

Abstract: Modelling accuracy is the main factor in an efficient controller design. Nowadays, several systems are modeled using the conventional modelling techniques such as Lagrangian and Newtonian frameworks. Those modelling techniques are known for the high time consumption and modelling inaccuracy, this is due to the need for assuming parameters values that are hard to measure and model in an accurate way. Examples of those parameters are coefficients of friction and drag coefficients. Moreover, it is time consuming because of the known modelling complexity of those techniques. Therefore, system identification techniques started to get high attention, this is due to their ability to model the system dynamics in a very efficient manner and to capture the sudden changes in the system parameters. Moreover, they are time efficient and reliable modeling techniques. This thesis investigates online system identification for different systems types including linear and nonlinear systems. The identification algorithms include Extended Kalman Filter (EKF) for both linear and nonlinear state-space model systems. Linear identification algorithms also includes adaptive filters such as Least Mean Square (LMS) algorithm and Recursive Least Square (RLS) algorithm. Nonlinear identification algorithms includes nonlinear adaptive filters for polynomial models such as Volterra LMS algorithm, Volterra RLS algorithm and bilinear RLS algorithm. Reference systems use-cases are inverted pendulum as a nonlinear system, Direct Current (DC) motor as a linear system, also a linearized model of the inverted pendulum is identified using linear system identification algorithms. The algorithms are implemented on MATLAB/Simulink software. Discrete controllers are designed for the identified systems to investigate the identified systems responses and errors in comparison with the reference systems. Then, the identification algorithms and the controllers are implemented through FPGA in the Loop (FIL) simulation utilizing a PYNQ Z-1 board.

This is achieved by running the system reference model on MATLAB, while the identification algorithms and controllers are on the Programmable Logic (PL) of the Field-Programmable Gate Array (FPGA). The efficiency of the algorithms is tested based on the identified parameters accuracy, resources utilization and the parameters convergence rate to reach their optimum values. Moreover, for linear systems EKF was found to be the most efficient algorithm in terms of parameters convergence rate and identification accuracy in comparison with the adaptive filters. However, this response was achieved at the cost of high resources utilization. For nonlinear systems, the EKF also was found to give high accuracy parameter estimation. For the polynomial models, the used algorithms were able to identify the systems with relatively low resources utilization.