You are on page 1of 5

CORNELL UNIVERSITY Mechanical and Aerospace Engineering

Feedback Control Systems M&AE 4780/5780


Prof. B. Hencey Fall 2013

Problem Set 3

Date: Due:

Sep 19 2013 3pm Oct 3 2013

Problem #1 Closed-Loop Steady-State Error For each of the open-loop transfer functions below a) G(s) = b) G(s) = c) G(s) = s2 + 9 s(s + 3)(s + 6) 3(s + 2) s(s + 1) 2(s + 1)(s 3) (s + 2)(s + 5)(s + 6)

Determine the steady-state error of the closed-loop system when the following inputs are applied: i A step input: U (s) =
1 s 1 s2 3 s2 +9

ii A ramp input: U (s) =

iii A sinusoidal input: U (s) =

Problem #2 Root Locus Sketching For each of the systems with open-loop transfer functions given below, sketch the root locus by hand. Be sure to include in your sketch the open-loop poles and zeros and the asymptotes of the root locus (you need not calculate the exact locations of break-in/break-out points, imaginary axis crossings, or angles of departure). You may assume the gain K is non-negative. Do NOT use MATLAB for this problem. a) G(s) = b) G(s) = c) G(s) = d) G(s) = K (1 s) (s + 1)(s + 2) K (s + 1) s2 + s + 1 s2 K +s+1 K s(s + 3)(s + 5)

Problem #3 System Response and Root-locus control design. Abslutely NO Matlab on this problem, work your brain! i) (Time Response and Poles ) For each step response, sketch by hand the poles of the corresponding transfer function in the complex plane.

ii) (Poles and time responses ) For each set of transfer function poles in the complex plane shown below, sketch by hand the time response to an impulse input. That is, if these are the poles of G(s) = Y (s)/R(s), sketch y (t) for r(t) = (t).

iii) (Root-locus control design ) Consider the following root loci. For each, design a controller of the form K (s) = k H (s) where H (s) = s + a or H (s) =
1 s+b

that will stabilize the system for some gain value k . In other words, you are adding either a new zero or a new pole to the system. In your solution, provide a sketch of the new root locus with your added pole/zero clearly marked, and indicate whether large or small K stabilizes the system. Do not use MATLAB.

Problem #4 Root Locus Controller Design Consider the system described by the block diagram below: + R(s) Y (s) K (s) G(s)

a) b)

c) d)

s+5 + 3s2 + 8s + 24 Plot the root locus of the system with K (s) = k (you may use MATLAB for plotting). Is the system stable for any value of k ? We wish to design a lead or lag compensator (i.e. one pole and one zero) such that the closed-loop system is stable, the peak time is less than or equal to 0.75 seconds, and the settling time is less than 2 seconds (using the 2% criterion). i. Determine the desired location of the dominant closed-loop poles based on these specications. ii. Design a compensator to place the closed-loop poles at the desired location (i.e. design K (s) as a lead or lag compensator). Avoid pole-zero cancellation in your design. Plot the root locus of the system using the compensator you have designed (you may use MATLAB for plotting). Mark the closed-loop poles and the system specications on the root locus. The specications for the system are based on second-order system dynamics, but the compensated closed-loop system is actually fourth-order. To check our approximation, use MATLAB to simulate the response of the closed-loop system to a unit step input and plot the results. Does the response satisfy the peak time and settling time specications? (If not, you do not need to redesign your compensator) G(s) = s3

Problem #5 Application Problem: Rocket Stability Control Consider again the rocket from Problem Set #3: a rocket just after lifto that has not yet reached cruising speed. The angle for the rocket centerline (t) with respect to the vertical and the angle of the thrust

nozzle (t) with respect to the centerline are functions of time. Assuming that the only forces acting on the rocket are gravity (at the center of mass) and the engine thrust F, the (highly simplied) non-linear equation of motion for this system (which you do NOT have to derive) is: 1 2 1 ml = mgl sin + aF sin 3 2

In the last assignment you calculated the linearized system about operating point 0 = 0 and 0 = 0 : 1 2 1 ml = mgl + aF 3 2 You determined through analysis and simulation that the open loop transfer function G(s) = (s)/(s) was unstable: (s) aF = 1 2 2 (s) [( 3 ml )s ( 1 2 mgl)] a) Now lets consider implementing closed-loop control on the rocket, as shown below. For now, take d(s) = 0. Calculate the closed loop system and tell whether it can be stable for any values of the

parameters. Do not use MATLAB. Note: in the above diagram, the input is zero (straight up ight). i. Proportional control K1 (s) = kp . ii. Derivative control K2 (s) = kd s. ki iii. Integral control K3 (s) = . s iv. PD control K4 (s) = kp + kd s. What eects will kp and kd have on the rocket if you vary them? b) Make a list of two or three potential disturbances or errors in the system. How might you take these into account with your model? (Note: feel free to use pictures, words, or equations to describe how you model the errors, but you dont have to re-derive the equation of motion or transfer function.) c) Consider a constant oset disturbance d(t) = 1(t) in the block diagram. This could be, for example, the result of an error in mounting the thrust nozzle so that is not measured exactly from the centerline. How would this aect the rockets performance for each of the (stabilizing) controllers from part (a)? Problem #6 Internal stability (5780) Consider the following SISO negative feedback system with controller K (s) and plant G(s):
D1 (s) R (s) E (s) K (s) U (s) G( s ) D2 (s) Y (s)

The signals D1 (s) and D2 (s) are actuator and plant/sensor noise. The controller and plant are governed by the following transfer functions. Notice that the plant contains a pole in the right-hand complex plane that is cancelled by a zero in the controller. K (s) = s8 (s + 3)2 G(s) = s+4 . s(s 8)

Pole-zero cancellation should never be applied to unstable poles because such cancellation is impossible to achieve in practice for the following reasons:

The plants pole will always be slightly dierent from the pole in the dynamical model. Noise injected into the system may excite unstable operation, even if the closed-loop transfer function suggests otherwise. Here we will explore why unstable pole-zero cancellation is problematic. a) Initially assume noiseless operation, i.e. D1 (s) = D2 (s) = 0. Compute the transfer function it stable? Hint: First compute 1 + K (s)G(s) and save it for later. b) Use the nal value theorem to show that limt y (t) = 1 for a step input. c) Construct and simulate the system in Matlab using the following commands. Does the system behave the way you expect? Why or why not? s = tf ( s ); K = ( s - 8) / ( s + 3)^2; G = ( s + 4) / ( s * ( s - 8)); step ( *** PUT Y ( s )/ R ( s ) HERE ***
Y (s) R(s) .

Is

);

d) Now let us consider the internal stability of the loop when noise inputs are considered. For a system to be stable, all signal paths in the loopnot just the measured system outputmust be stable. Compute the following transfer functions and determine, by inspecting the poles, whether each is stable. Hint: To save time, rst compute 1+1 GK . Do not forget the sign inversion from Y (s) E (s). i) Problem #7 (5780) Frequently, in physical systems, energy can be expressed in terms of squares of physical states. Given a dynamical system x (t) = Ax(t), the generalized instantaneous energy of the system can be expressed in matrix form as V (x(t)) = x (t)P x(t) P = P > 0. For a system to be asymptotically stable, it must dissipate energy so that it eventually reaches a zero-energy state. The generalized dissipation, or rate of change of energy, can be expressed as (x(t)) = x (t)Qx(t) V Q = Q > 0,
U (s) D1 (s)

ii)

U (s) D2 (s)

iii)

Y (s) D1 (s)

iv)

Y (s) D2 (s)

meaning the system energy should always trend toward zero. a) Show that A P + P A + Q = 0. Hint: Take the derivative of V (x(t)) and set it equal to the denition (x(t)). of V b) Show that the equation derived in part A implies that the eigenvalues i (A) < 0. i. Start with the equation A P + P A = Q. Multiply the entire equation on the left and right by x and x, respectively. ii. Let x be an eigenvector of A, and use Ax = x to show that 2 () = Because Q > 0 and P > 0, this shows that x Qx x Px

() < 0.

You might also like