Specialized course Process Control at Norwegian University of Life Sciences (NMBU), Spring 2018


Compulsory exercise to lessons in Lecture 4

 

Systems theory

1.      Learning Matlab Control System Toolbox (“CST”): Work through the following sections of this tutorial to the Matlab Control System Toolbox: 1, 2, 3 - intro, 3.1, 3.2, 5. (You are not expected to report anything from this task.)

2.      Using CST: Stability analysis and simulation of transfer functions: See Exercise 4.2 in the exercise book. For each of the transfer functions: Calculate the poles, plot the poles, determine the stability property (“manually”), and simulate both the impulse response and the step response. [Abhilash]

3.      Using CST: Stability analysis and simulation of a state space model: See Exercise 4.3 in the exercise book. [Alexander]

a.       Determine the stability property of the system from its eigenvalues. Simulate an initial state response of the state space model, i.e. assume some non-zero initial state, and simulate the responses in the states.

b.      Find the transfer function from u to y, and calculate its poles. Are these poles the same as the eigenvalues of the state space model?

4.      Using CST: Tuning and simulation of a feedback control system: See Example 5.1 in the text book. Assume the controller is a PI controller. Tune the controller using the Ziegler-Nichols’ method. Plot the closed loop poles (i.e. the poles of the closed loop, i.e. the feedback control system), and simulate the setpoint-step response after the tuning. (Tip: Create a transfer function model of the control system using the feedback function of CST.) [Duo]

5.      Using CST: Discretization of a continuous-time model: See Exercise 4.3 in the exercise book. Discretize the given continuous-time state space model using the ZOH method with time-step 0.1 s. Calculate the eigenvalues of the resulting discrete-time model, and conclude about its stability property. Is the stability property the same as for the original continuous-time model ? (Cf. Problem 3 above.) [Karla]

6.      Symbolic linearization with Matlab Symbolic Toolbox: Linearize the nonlinear model (1.2) - (1.3) shown in the exercise book. (Tip: Use the jacobian function.) Compare with (your) manual result. [Xiaodong]

 

Optimization

The Rosenbrock optimization problem (“ROP”) is a “standard” optimization problem, cf. https://se.mathworks.com/help/optim/examples/banana-function-minimization.html.

1.      Grid search: Solve the ROP using the grid search method. The initial guess can be set to [-1.9,2] as in the above reference, and you may allow for a search of the optimal solution within -5 and 5 for both x1 and x2. (Template for solution, similar to script presented in lecture 22 Feb.) [Abhilash]

2.      fmincon: The same as the above problem, but now by using fmincon. (Template for solution, similar to script presented in lecture 22 Feb.) [Alexander]

3.      Newton search: The same as the above problem, but now by implementing the Newton search method, from scratch. You may derive the gradient function and the Hessian function symbolically with the gradient() and the hessian() functions in the Symbolic Toolbox. (Template for solution, similar to script presented in lecture 22 Feb.)
Newton search: x_kp1 = x_k - inv[Hessian_f(x_k)]*Gradient_f(x_k), where “kp1” means “k plus 1”. [Duo]

4.      Grid search, with constraints: Same as the grid problem above, but now include the constraint x2 >= x1+1. [Karla]

5.      fmincon, with constraints: Same as the fmincon problem above, but now include the constraint x2 >= x1+1. [Xiaodong]


Updated 11 March 2018 by Finn Aakre Haugen, course teacher.