Chapter 2. Classical Control System Design. Dutch Institute of Systems and Control

Size: px

Start display at page:

Download "Chapter 2. Classical Control System Design. Dutch Institute of Systems and Control"

Darren Grant
6 years ago
Views:

1 Chapter 2 Classical Control System Design

2 Overview Ch Classical control system design Introduction Introduction Steady-state Steady-state errors errors Type Type k k systems systems Integral Integral control control Frequency Frequency response response plots plots Bode Bode plots plots Classical Classical design design techniques techniques Classical Classical design design specifications specifications Lead, Lead, lag, lag, lead-lag lead-lag compensation compensation Guillemin-Truxal Guillemin-Truxal method method Quantitative Quantitative Feedback Feedback Theory Theory Root Root locus locus Nyquist Nyquist plots plots M- M- and and N-circles N-circles Nichols Nichols plots plots

3 Steady-state errors-1 r + F C P y Tracking behavior: Assume n t rt () = 1() t n! rs ˆ( ) = 1 n 1 s + Response ys $( ) Ls ( ) = ( )$( ) Ls ( ) Fsrs H( s) Tracking error ε ˆ() s = rˆ() s yˆ() s = [1 H()]() s rˆ s

4 Steady-state errors-2 Steady-state tracking error ( n) ε = lim ε( t) = lim sεˆ ( s) = lim t s 0 s 0 If F(s)=1 (no prefilter) then 1 1 H( s) = 1 + Ls ( ) 1 H( s) s n ( n) ε = 1 lim 0 n [1 + ( )] s s L s

5 Type k system A feedback system is of type k if Then Lo () s Ls () =, Lo (0) 0 k s ( n) ε = 1 L s lim s 0 n [1 + ( )] s 0 for 0 n< k n s = lim = 1/ Lo (0) for n k s 0 k = s + Lo () s for n> k k

6 Steady-state errors-3

7 Integral control-1 Integral control: Design the closed-loop system such that Type k control: Ls () = L o k s () s Ls () = Lo () s s Results in good steady-state behavior Also: k 1 s k Ss () = = = O( s ) for s Ls ( ) k s + L () s o

8 Integral control-2 Type k control: Hence if k Ss () = O( s ) for s 0 n t 1 vt () = 1(), t vs ˆ( ) = n! n s + 1 then the steady-state error is zero if n < k (rejection) k = 1: Integral control: Rejection of constant disturbances k = 2: Type-2 control: Rejection of ramp disturbances Etc.

9 Integral control-3 Integral control: Lo () s Ls () = = PsCs () () k s The loop has integrating action of order k Natural integrating action is present if the plant transfer function has one or several poles at 0 If no natural integrating action exists then the compensator needs to provide it

10 Integral control-4 Pure integral control: Cs () = 1 st i PI control: Cs () = g 1+ 1 st i PID control: Cs () = g std st i Ziegler-Nichols tuning rules

11 Internal model principle Asymptotic tracking if model of disturbance is included in the compensator Francis, D.A. and Wonham, W.M., (1975) The internal model principle for linear multivariable regulators, Applied Mathematics and Optimization, vol 2, pp

12 Frequency response plots Bode plots Nichols plots Nyquist plots

13 Bode plots-1 Bode plot: doubly logarithmic plot of L(jω) versus ω semi logarithmic plot of arg L(jω) versus ω L( jω ) = 2 ωo o o j + o ( jω ) 2 ζ ω ( ω) ω

14 Bode plots-2 Helpful technique: By construction of the asymptotic Bode plots of elementary first- and second-order factors of the form The shape of the Bode plot of ( jω z1)( jω z2) L( jω zm ) L( jω ) = k ( j ω p )( j ω p ) L( j ω p ) o o jω + α and ( jω) + 2 ζ ω ( jω) + ω may be sketched m

15 Nyquist plots Nyquist plot: Locus of L(jω) in the complex plane with ω as parameter Contains less information than the Bode plot if ω is not marked along the locus L( jω ) = 2 ωo o o j + o ( jω ) 2 ζ ω ( ω) ω

16 M- and N-circles-1 r + L y Closed-loop transfer function: H L = = T 1 + L M-circle: Locus of points z in the complex plane where z = M 1+ z N-circle: Locus of points z in the complex plane where arg z 1+ z = N

17 M- and N-circles-2

18 Nichols plots Nichols plot: Locus of L(jω) with ω as parameter in the log magnitude versus argument plane 2 ωo o o j + o L( jω ) = ( jω) 2 ζ ω ( ω) ω Nichols chart: Nichols plot with M- and N-loci included

19 Classical design specifications Time Rise time, delay time, overshoot, settling time, steady-state error of the response to step reference and disturbance inputs; error constants domain domain Frequency Bandwidth, resonance peak, roll-on and roll-off of the closed-loop frequency response and sensitivity functions; stability margins

20 Classical design techniques Lead, lag, and lag-lead compensation (loopshaping) (Root locus approach) (Guillemin-Truxal design procedure) Quantitative feedback theory QFT (robust loopshaping)

21 Classical design techniques Rules for loopshaping Change open-loop L(s) to achieve certain closed-loop specs first modify phase then correct gain

22 Lead compensation Lead compensation: Add extra phase in the cross-over region to improve the stability margins Typical compensator: Phase-advance network 1+ jωt C( jω) = α, 0< α < 1 1+ jωα T

23 Lead/lag compensator C( jω) = α 1+ jωt 1+ jωα T

24 Lag compensation Lag compensation: Increase the low frequency gain without affecting the phase in the cross-over region Example: PI-control: C( jω ) = k 1+ jωt jωt

25 Lead-lag compensation Lead-lag compensation: Joint use of lag compensation at low frequencies phase lead compensation at crossover Lead, lag, and lead-lag compensation are always used in combination with gain adjustment

26 Notch compensation (inverse) Notch filters: suppression of parasitic dynamics additional gain at specific frequencies Special form of general second order filter

27 Notch compensation H = u ε = s ω s ω β + 2β 1 2 s ω s ω Notch -filter :ω 1 = ω 2

28 Notch compensation ampl. β β 1 2 fase 0

29 Root locus method-1 Important stage of many designs: Fine tuning of gain compensator pole and zero locations Helpful approach: the root locus method (use rltool!)

30 Root locus method-2 Ls () N() s ( s z1)( s z2) L( s zm ) = = k D() s ( s p )( s p ) L( s p ) 1 2 n L Closed-loop characteristic polynomial χ () s = D() s + N() s = ( s p )( s p ) L( s p ) + k( s z )( s z ) L( s z ) 1 2 n 1 2 Root locus method: Determine the loci of the roots of χ as the gain k varies m

31 Root locus method-3 χ () s = ( s p )( s p ) L( s p ) + k( s z )( s z ) L( s z ) 1 2 n 1 2 Rules: For k = 0 the roots are the open-loop poles p i For k a number m of the roots approach the open-loop zeros z i. The remaining roots approach The directions of the asymptotes of those roots that approach are given by the angles 2i + 1, i 0,1,, n m 1 n m π = L m

32 Root locus method-4 The asymptotes intersect on the real axis in the point (sum of open-loop poles) (sum of open-loop zeros) n m Those sections of the real axis located to the left of an odd total number of open-loop poles and zeros on this axis belong to a locus The loci are symmetric with respect to the real axis...

33 Root locus method-5 Ls () = k ss ( + 2) Ls () = ks ( + 2) ss ( + 1) Ls () = k ss ( + 1)( s+ 2)

34 Guillemin-Truxal method-1 r + C P y Closed-loop transfer function: PC H = 1 + PC Procedure: Specify H Solve the compensator from C 1 H = P 1 H

35 Guillemin-Truxal method-2 Example: Choose H() s = m m 1 ams + am 1s + L+ a0 n n 1 m m 1 + n 1 + L+ m + m 1 + L+ 0 s a s a s a s a This guarantees the system to be of type m + 1 How to choose the denominator polynomial? Well-known options: Butterworth polynomials Optimal ITAE polynomials

36 Butterworth and ITAE polynomials Butterworth polynomials Choose the n left-half plane poles on the unit circle so that together with their right-half plane mirror images they are uniformly distributed along the unit circle ITAE polynomials Place the poles so that () tet dt 0 is minimal, where e is the tracking error for a step input

37 Butterworth and ITAE m = 0

38 Guillemin-Truxal method-3 Disadvantages of the method: Difficult to translate the specs into an unambiguous choice of H. Often experimentation with other design methods is needed to establish what may be achieved. In any case preparatory analysis is required to determine the order of the compensator and to make sure that it is proper The method often results in undesired pole-zero cancellation between the plant and the compensator

39 Quantitative feedback theory QFT-1 Ingredients of QFT: For a number of selected frequencies, represent the uncertainty regions of the plant frequency response in the Nichols chart Specify tolerance bounds on the magnitude of T Shape the loop gain so that the tolerance bounds are never violated

40 QFT-2 Example: Plant Ps () = s 2 g (1 + sθ ) Nominal parameter values: g = 1, θ = 0 Parameter uncertainties: 0.5 g 2, 0 θ 0.2 Tentative compensator: k + std Cs () =, k= 1, Td = 1.414, To = st o

41 QFT-3 Responses of the nominal design Specs on T Frequency [rad/s] Tolerance band [db]

42 Uncertainty regions Uncertainty regions for the nominal design The specs are not satisfied Additional requirement: The critical area may not be entered

43 QFT-4 Design method: Manipulate the compensator frequency reponse so that the loop gain satisfies the tolerance bounds avoids the critical region Preparatory step 1: For each selected frequency, determine the performance boundary Preparatory step 2: For each selectedfrequency, determine the robustness boundary

44 Performance and robustness boundaries Nominal plant frequency response Robustness boundaries Performance boundaries

45 QFT-5 Design step: Modify the loop gain such that for each selected frequency the corresponding point on the loop gain plot lies above and to the right of the corresponding boundary For the case at hand this may be accomplished by a lead compensator of the form 1+ st Cs () = 1 + st Step 1: Set T 2 = 0, vary T 1 Step 2: Keep T 1 fixed, vary T 2 1 2

46 QFT-6 Eventual design: T 1 = 3 T 2 = 0.02

47 QFT-7 Responses of the redesigned system

48 Prefilter design-1 2½-degree-of-freedom configuration Closed-loop transfer function H = NF D cl F o r F o e C o F X Y X + + u P z For the present case: Dcl ( s) = 0.02 ( s ) ( s )( s ) N() s = 1

49 Prefilter design-2 Use the polynomial F to cancel the (slow) pole at , and let 2 ωo Fo ( s) =, ωo = 1, ζo = s + 2ζ ω s+ ω 2 2 Perturbed responses o o o

Automatic Control 2. Loop shaping. Prof. Alberto Bemporad. University of Trento. Academic year

Automatic Control 2. Loop shaping. Prof. Alberto Bemporad. University of Trento. Academic year Automatic Control 2 Loop shaping Prof. Alberto Bemporad University of Trento Academic year 21-211 Prof. Alberto Bemporad (University of Trento) Automatic Control 2 Academic year 21-211 1 / 39 Feedback