Using MATLAB in Feedback Systems

E102 Using MATLAB in Feedback Systems Part I Classical Design Classical Control Design with MATLAB is illustrated by means of an example of the design of a dc servomotor control system The transfer function of the dc motor is given by: H(s) = 80 s(s + 24 s + 48 ) Our task is to design a PID controller I + Ds s K (1 + j ωTD )(1 + jωTI ) C ( j ω) = jω C ( s) = P + or to achieve the following specifications for the closed loop step response: • Rise time < 0.5s • Overshoot < 10% We shall work through the following steps for the design: A Calculating the required open loop frequency response B Determining the PID parameters C Plotting the step response of the closed loop D Fine-tuning the design to achieve the specifications A Open Loop Frequency Response First, we use the design specifications to calculate the natural frequency and damping factor for the closed loop system For a rise time tr = 0.5 s, overshoot Mp = 0.1 : => ω n = 3.6 rad/s => ζ = 0.591 Now use the design relations from the lecture notes to specifiy the open loop crossover frequency and phase margin: => ω 2c = ω 2n  4ζ + − 2ζ  => ωc = 2.60 rad/s   −1  2ζωn  => φ m = tan   => φ m = 1.02 rad  ωc  Anthony Bright Page 04/ 18/ 02 B Determining the PID parameters We will use a Bode plot to help select the controller parameters for the system First we will look at the frequency response of the dc motor You can write transfer functions in MATLAB by entering the numerator and denominator polynomials of a rational transfer function as vectors of coefficients in descending powers of s For example, to enter the dc motor transfer function given in the problem statement, you would enter the numerator and denominator polynomials as two row vectors and define the system as a transfer function model: >>num=[80]; den=[1 24 48 0]; >>H=tf(num,den) Transfer function: 80 s^3 + 24 s^2 + 48 s The bode plot for the dc motor can now be plotted: >>bode(H) Anthony Bright Page 04/ 18/ 02 The bode plot shows a crossover frequency of 1.4 rad/s The crossover frequency must be raised to 2.60 rad/s to speed up the time response First, find the magnitude and phase of the dc motor at the crossover frequency : >>[magH,phaseH]=bode(H,2.60) magH = 0.4117 phaseH = -146.5225 At 2.60 rad/s, the combined phase of the controller and process must give a phase margin of 1.02 rad (59o ) The phase lead added by the controller at 2.60 rad/s must be -180 + 59 +147 = 26o ( ∠C ( jω c ) = 0.44 rad) and the gain added by the controller at 2.60 rad/s must equal the reciprocal of the dc motor gain C ( j ωc ) = = 2.43 412 We can now calculate the controller parameters, assuming that T I = 10T D by solving π ∠C ( jω c ) = tan − 1( ωc TD ) + tan −1 ( ωcTI ) − C ( jω c ) = K + ω 2c TD2 + ω 2c TI2 ωc => TD = 0.256 s ; => TI = 2.56 s => K = 0.780 So the PID controller is given by  (0.256 s + 1)( 2.56s + 1)  0.512 s + 2.20s + 0.780 C ( s) = 0.780 =  s s   >>num=[0.512 2.40 0.780];den=[1 0];C=tf(num,den) Transfer function: 0.512 s^2 + 2.4 s + 0.78 -s Now check the phase margin and crossover frequency of our design with the open loop bode plot Anthony Bright Page 04/ 18/ 02 >>margin(C*H) The phase margin is 59o at 2.6 rad/s, just as required C Plotting the step response of the closed loop Now we check the closed loop behaviour of the system First find the closed loop transfer function Q(s): >> Q=feedback(C*H,1) Transfer function: 40.99 s^2 + 175.9 s + 62.42 -s^4 + 24 s^3 + 88.99 s^2 + 175.9 s + 62.42 The syntax of the feedback function is : feedback(forward path transfer function, feedback path transfer function) In our case we have unity feedback Anthony Bright Page 04/ 18/ 02 Plot the step response of the closed loop system for < t < s >>step(Q,4) The step response meets the rise time requirement (< 0.5 s) but not the overshoot specification (> PID_tune • Tune the controller parameters to achieve the desired specification Anthony Bright Page 04/ 18/ 02 E102 Using MATLAB in Feedback Systems Part II State Space Design We will carry out the same design as in Part I using State Space Methods The state space equations describing the dc motor are: dθ =ω dt dω = 4i dt di = −24i − 12ω + 20v dt Our task is to design a state space control scheme to achieve the following specifications for the closed loop step response : • Rise time < 0.5s • Overshoot < 10% We shall work through the following steps for the design: A Entering the state space representation of the system B Implementing full state feedback with pole placement C Implementing full state feedback with optimal control D Implementing feedback with a state observer A State Space Representation of the System  θ Enter the state space matrices, with x = ω , u = v, y = θ  i  >>A=[0 0;0 4;0 –12 –24];B=[0 20]’;C=[1 0];D=0; >>H=ss(A,B,C,D) a = x1 x2 x3 x1 0 x1 x2 x3 u1 0 20 x2 -12 x3 -24 b = Anthony Bright Page 04/ 18/ 02 c = y1 x1 y1 u1 x2 x3 d = Continuous-time system The command ss created a state space model of the system We can now look at the frequency response bode(H), step response step(H), and so forth, just as we did for the transfer function models in Part I With full state feedback, the feedback gain can be evaluated using pole placement or the LQR method B Pole Placement First determine the dominant closed loop pole positions • What values of the dominant poles will give the appropriate time response ( >p=[-2.2 + 2.8j, -2.2 – 2.8j, -22]; Find the full state feedback gain to place these poles >>K=place(A,B,p) place: ndigits= 15 K = 3.6437 0.7756 0.1200 The command place matches the eigenvalues of the closed loop system matrix Af = A-BK with the chosen poles It also measures the number of digits accuracy in the position of the closed loop poles Next, the reference gain Kr is determined >>N=inv([A B;C D])*[zeros(length(B),1);1]; >>Nx=N(1:length(B));Nu=N(length(B)+1); Kr= Nu + K*Nx Kr = 3.6437 The closed loop system can now be defined in state space form >>Q=ss(A-B*K,B*Kr,C-D*K,D*Kr); Anthony Bright Page 04/ 18/ 02 and the closed loop step response plotted >>step(Q) Rise time = 0.53 s Overshoot 9% Close enough!! C Optimal Control: LQR Method First choose the weighting parameters of the performance index >>Q=[10 0;0 0;0 0];R=.1; This choice weights the motor angle with a factor of ten and the control input with a factor of 0.1 Solve the linear quadratic regulator with Matlab function lqr >>[K,S,e]=lqr(A,B,Q,R) K = 10.0000 2.2273 S = 2.8273 0.3817 0.0500 0.3817 0.0774 0.0111 0.3267 0.0500 0.0111 0.0016 e = -21.8628 -4.3360 + 4.2180i -4.3360 - 4.2180i Anthony Bright Page 04/ 18/ 02 The values of the feedback gain vector K are the optimum values to minimize the performance index The matrix S is the solution of the associated Riccati equation and the e vector contains the closed loop eigenvalues (poles) The closed loop response follows as before >>Kr= Nu + K*Nx >>Q=ss(A-B*K,B*Kr,C-D*K,D*Kr); step(Q) Rise time 0.38 s, Overshoot 4% This looks too easy!! D Observer Design Full state feedback of the servo motor requires simultaneous measurement of motor angular position, angular velocity and motor current Typically all of these are not measured So we need to include an observer to estimate the state of the system We will choose the observer poles to be twice as fast as the system poles: >>pe=2*e pe = -43.7256 -8.6720 + 8.4359i -8.6720 - 8.4359i Now match the eigenvalues of Ao = A-LC with the chosen poles >>L=place(A’,C’,pe)’ Anthony Bright Page 10 04/ 18/ 02 L = 1.0e+003 * 0.0371 -0.0329 1.3527 Note that the place function solves A-BK, so we need to invert the A and C matrices to find the column vector L The observer gain L is now incorporated into an augmented state space description of both the system state and the observer state: >>Ae=[A -B*K;L*C A-L*C-B*K];Be=[B*Kr;B*Kr]; >>Ce=[C -D*K; 0 C-D*K];De=[D*Kr;D*Kr]; The closed loop step response follows as before: >>Q=ss(Ae,Be,Ce,De); >>step(Q);title('LQR Design with observer') The upper reponse curve corresponds to the measured output y = θ ; the lower curve to the estimated output yˆ = θˆ Overshoot 4%, rise time 0.38s, the observer does an excellent job! Anthony Bright Page 11 04/ 18/ 02 Before you come to the conclusion that the problem is solved and you can take the rest of the evening off, you should just check the closed loop response of the other two state variables, the motor speed ω and the motor current i For example to find the closed loop step response of the motor current, we change the output matrix: >>C=[0 1];Ce=[C -D*K2; 0 C-D*K2]; >>Qi=ss(Ae,Be,Ce,De);step(Qi) The motor current exceeds amps for a short duration This current level could burn out the motor windings We need to relax some of the design specifications and penalize large values of current in our weighting matrix Q • Fine tune the observer-based state space control scheme for rise time < 1.5 s, overshoot < 10%, motor speed < rad/s and motor current < amps The M-file stspace_tune.m given on the next page may be helpful: Anthony Bright Page 12 04/ 18/ 02 %stspace_tune M file to tune a state space controller % %Enter state space matrices A=[0 0;0 4;0 -12 -24];B=[0 20]';C=[1 0];D=0; H=ss(A,B,C,D); % LQR design Q=[10 0;0 0;0 0];R=0.1; [K,S,e]=lqr(A,B,Q,R); % Reference Gain N=inv([A B;C D])*[zeros(length(B),1);1]; Nx=N(1:length(B));Nu=N(length(B)+1); Kr= Nu + K*Nx; %Observer Design % place observer poles p=2*e; L=place(A',C',p)'; % closed loop state space form Ae=[A -B*K;L*C A-L*C-B*K];Be=[B*Kr;B*Kr]; Ce=[C -D*K; 0 C-D*K];De=[D*Kr;D*Kr]; % Step Response Q=ss(Ae,Be,Ce,De);step(Q) Anthony Bright Page 13 04/ 18/ 02 E102 Using MATLAB in Feedback Systems Part III Digital Control Design We continue the servomotor design with digital control The transfer function of the dc motor is given by: 80 H ( s) = s( s + 24 s + 48) We shall work through the following steps for the design: A Emulating the continuous design with a discrete PID controller IT D C (z) = P + + (1 − z −1 ) −1 1− z T B Implementing discrete state space feedback A Emulation of Continuous Control Choose a sampling frequency ≅ 20 times the closed loop bandwidth (20Hz) >>fs=20;T=1/fs; Convert the analog PID settings to discrete form: >>P=2.44;I=0.864;D=0.567; >>Cz=tf([(P+I*T+D/T) –(P+2*D/T) D/T],[1 -1 0],T) Transfer function: 13.82 z^2 - 25.12 z + 11.34 z^2 - z Sampling time: 0.05 Notice that the addition of the sample time T to the transfer function command tf creates a discrete time model of the PID controller Anthony Bright Page 14 04/ 18/ 02 Now we convert the continuous process transfer function to the equivalent discrete transfer function using the c2d (continuous-to-discrete) command applying a zero order hold to the input: >>Hs=tf(80,[1 24 48 0]);Hz=c2d(Hs,T,'zoh') Transfer function: 0.001259 z^2 + 0.003814 z + 0.0006928 z^3 - 2.232 z^2 + 1.533 z - 0.3012 Sampling time: 0.05 Now we find the step response The output of the system is continuous, but we can find the sampled output from the discrete closed loop transfer function: >>Qz=feedback(Cz*Hz,1);step(Qz); If we compare this response to the results for the continuous control system (Part I), the rise time is reduced slightly but the overshoot is increased Anthony Bright Page 15 04/ 18/ 02 B Discrete State Space Design The continuous state space equations for the system can be converted to discrete form and state feedback strategies applied to the discrete system First find the equivalent discrete state model for the process: >>A=[0 0;0 4;0 -12 -24];B=[0 20]';C=[1 0];D=0; >>Hc=ss(A,B,C,D); >>Hd=c2d(Hc,T,'zoh'); >>Ad=Hd.a >>Bd=Hd.b Ad = 1.0000 0 0.0492 0.9586 -0.3426 0.0034 0.1142 0.2734 Bd = 0.0013 0.0690 0.5710 To apply the optimal control LQR method for the state feedback gain, choose the weighting parameters of the objective function >>Q=[10 0;0 0;0 ];R=0.1; Now solve the discrete linear quadratic regulator with Matlab function dlqr >>[K,S,e]=dlqr(Ad,Bd,Q,R) K = 8.4950 2.0222 0.3007 S = 61.7350 7.6247 1.0043 7.6247 1.5544 0.2236 1.0043 0.2236 0.0328 e = 0.7875 + 0.1689i 0.7875 - 0.1689i 0.3350 Anthony Bright Page 16 04/ 18/ 02 Now the reference gain is found: >>N=inv([Ad-eye(3) Bd;C D])*[zeros(length(Bd),1);1]; >>Nx=N(1:length(Bd));Nu=N(length(Bd)+1); Kr= Nu + K*Nx Kr = 8.4950 The observer is designed by pole placement Choose estimator poles to be one half the value of the system poles (making them about twice as fast in the discrete domain), and solve for the estimator gain: >>pe=e/2, L=place(Ad',C',pe)' pe = 0.3938 + 0.0845i 0.3938 - 0.0845i 0.1675 place: ndigits= 15 L = 1.2769 7.0198 -3.3579 Write the augmented discrete state space equations for both the state variables and the estimates of the state variables and plot the step response: >>Ae=[Ad -Bd*K;L*C Ad-L*C-Bd*K];Be=[Bd*Kr;Bd*Kr];Ce=[C -D*K];De=D*Kr >>Q=ss(Ae,Be,Ce,De,T);step(Q) Anthony Bright Page 17 04/ 18/ 02 The response is practically identical to the results from the continuous state space system and leaves open the same questions about the behavior of the other state variables • Fine tune the observer-based digital state space control scheme for rise time < 1.5 s, overshoot < 10%, motor speed < rad/s and motor current < amps The following M-file may be helpful: %digi_tune M file to tune a digital state space controller % % Enter sample time fs=20;T=1/fs; % LQR design A=[0 0;0 4;0 -12 -24];B=[0 20]';C=[1 0];D=0; Hc=ss(A,B,C,D);Hd=c2d(Hc,T,'zoh'); Ad=Hd.a;Bd=Hd.b; Q=[10 0;0 0;0 0];R=0.1; [K,S,e]=dlqr(Ad,Bd,Q,R) % Reference Gain N=inv([Ad-eye(3) Bd;C D])*[zeros(length(Bd),1);1]; Nx=N(1:length(Bd));Nu=N(length(Bd)+1); Kr= Nu + K*Nx; % place estimator poles pe=e/2; L=place(Ad',C',pe)'; % closed loop state space form Ae=[Ad -Bd*K;L*C Ad-L*C-Bd*K];Be=[Bd*Kr;Bd*Kr];Ce=[C -D*K];De=D*Kr % Step Response Q=ss(Ae,Be,Ce,De,T);step(Q) Anthony Bright Page 18 04/ 18/ 02 ... transfer function models in Part I With full state feedback, the feedback gain can be evaluated using pole placement or the LQR method B Pole Placement First determine the dominant closed loop pole... Q =feedback( C*H,1); step(Q,4) Run the M- file >> PID_tune • Tune the controller parameters to achieve the desired specification Anthony Bright Page 04/ 18/ 02 E102 Using MATLAB in Feedback Systems. .. Q=ss(Ae,Be,Ce,De);step(Q) Anthony Bright Page 13 04/ 18/ 02 E102 Using MATLAB in Feedback Systems Part III Digital Control Design We continue the servomotor design with digital control The transfer