Fundamentals of Automatic Control .................................................................. 26 4.1 Types of controller action ........................................................................... 26 4.2 Closed loop response................................................................................. 30 Controller Tuning ............................................................................................... 32 5.1 Performance criteria for closed-loop systems ............................................ 32 5.2 Ultimate gain (Ziegler Nichols method) ...................................................... 33 5.3 Controller synthesis.................................................................................... 34 5.4 Tuning relations.......................................................................................... 37 5.5 Guidelines for common control loops ......................................................... 39 Advanced Control Systems ............................................................................... 47 6.1 Combination of feedback and feedforward control ..................................... 47 6.2 Ratio control ............................................................................................... 49 6.3 Cascade control ......................................................................................... 50 6.4 Time delay compensation .......................................................................... 51 6.5 Nonlinear Control systems ......................................................................... 52 6.6 Adaptive Control......................................................................................... 53 6.7 Multivariable Control Problems................................................................... 53 6.8 Model predictive control ............................................................................. 55
-1-
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
u(t)
Process
y(t)
- Which is the input of the controller in conventional feedback control? Complete the picture for the stirred tank heater example! -
Simplest and despite its drawbacks widely used type of control, commonly used as thermostats in heating system (e.g. coffee machine, fridge).
Simple (inexpensive).
Oscillatory nature of controller action. - Why do we need a differential gap resp. what is the impact of the size of that gap on controller action? -
- 26 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
KP (Controller gain): Can be adjusted to make the controller output changes more or less sensitive to deviations between set point and CV. u (Bias): u(t) = u , when e = 0. u is adjusted, so that y is at its nominal steady state value, i.e. e = 0.
Inability to eliminate steady state error that occurs after set-point changes
(Offset has to be eliminated manually by resetting the bias u ). Simplicity attractive in control applications where offset can be tolerated (e.g. level control in buffer tanks). Example: Stirred tank heating process (see chapter 2.2.1 and 3.1.1): Control objective: Keep the tank temperature at its desired value set by adjusting the rate of heat input Pel.
Fin, in
U ~
Pel
TC
Presume the bias value Pel of the controller Fout, out = to be adjusted such that the process is initally at its set point s = set. (Which value has to be chosen?) What is the response of the controlled process in case of a step change in in (regulatory control, see figure below) resp. set (servo control)?
Controlled variable s,1 = set Pel Pel,1 = Pel Manipulated variable in,2 Disturbance variable in,1 t0
eS = ???
Pel = ???
in t
- 27 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Remark: Some controllers have a proportional band setting instead of a gain: 100% u PB = . KP
A narrow PB thus corresponds to a large controller gain.
umax
u
In practice, physical limits on the controller output (e.g. valve position) need to be taken into consideration.
umin e
- 28 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
- How does the derivative mode respond to a step change in the set-point? -
de 1 + dt TI
e( )d
0
The contributions of the different controller modes and the impact of the controller parameters on the closed loop response of the stirred tank heater are further investigated in exercise 8.
- 29 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
D(s) G2(s)
where G1 and G2 describe the impact of the manipulated (u) resp. the disturbance variable (d) on the controlled variable (y)
(see previous lectures).
Yset(s) -
E(s)
GC(s)
Controller
U(s)
G1(s)
Process
Y(s)
U(s) E(s)
of the controller depends on the control algorithm. How does y respond to changes in the set point yset or the disturbance variable d?
Y(s) = G1(s) U(s) + G2(s) D(s) + G2(s) D(s) = G1(s) GC(s) E(s) = G1(s) GC(s) (Yset(s)-Y(s)) + G2(s) D(s).
Y(s) =
G1(s) GC (s) G2 (s) Yset (s) + D(s) , 1 + G1(s) GC (s) 1 + G1(s) GC (s) 144 244 3 4 4 144 244 3 4 4
GCL1(s) GCL2 (s)
with GCL(s): Closed Loop transfer function The dynamics of the closed loop process thus also can be decribed by closed loop transfer functions GCL(s) indicating the impact of the inputs of the closed loop process, i.e. the set-point and disturbances.
- 30 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
The dynamic relationship between = -S and the heat input u = Pel - Pel,s can be described by the following transfer function (see chapter 3.2.3): G1(s) =
P-control:
Presume the bias value of the controller to be adjusted such that the process is initally at its set point, i.e. s = set for Pel,s = Pel . Consider deviation variable: Pel = Pel - Pel = K C e , where Pel (t = 0) = K C 0 = 0.
Laplace transform of the control law Pel (t) = K C e(t) Laplace P(s) = KCE(s) gives the following transfer function of a P-controller: G C (s) = P(s) = KC . E(s)
KK C K CL (s) = 1+ Ts = K = KK C 1 + TCL s set (s) 1 + 1+ Ts
The closed loop transfer function of the system thus is G CL (s) = (1st order transfer function),
with K CL =
KK C T and TCL = . 1+ KK C 1+ KK C
PI-control:
(2nd order transfer function). Whether or not oscillation occurs depends on the controller settings.
The impact of the controller parameters on the closed loop response (steady state error, speed of response) of this process is further investigated in exercise 8.
- 31 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
5 Controller Tuning
= adjustment of controller settings (for PID-control: KC, TI, TD), so that when coupled with the process it gives the desired closed loop response. Can be done by trial and error (see e.g. exercises 9, 10 and 12). Quite tedious and time consuming. This is why a good preliminary estimate is required. This can be obtained by experience with similar control loops, based on a model.
Obviously, these criteria involve inherent conflicts/ trade-offs, e.g. PID settings, which minimise the effect of disturbances, tend to produce large overshoots for set point changes. Controller settings have to be satisfactory for both load and set point changes! Conservative values (robustness ) create a sluggish response (performance ).
5.1.1 Stability
A process is said to be unstable, if for constrained inputs the output becomes larger and larger (positively or negatively) as time increases until a physical limit is reached/ some constraint will be met (e.g. safety valve will open). Most processes are open-loop stable (exception e.g. exothermic reactor). Processes, which are open-loop stable can be made closed-loop unstable, if the controller gain is large enough.
An unconstrained linear system is said to be stable, if the Output response is Bounded for all Bounded Inputs.
- 32 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Review example of CSTR given in exercise 6: Which of the steady states are stable?
Determined experimentally by increasing KC, until continuous cycling begins. Calculation of controller settings according to tuning rules: Controller P PI PID KC
0,5 KC,crit 0,45 KC,crit 0,6 KC,crit
TI
Tcrit /1,2 Tcrit / 2
TD
Tcrit / 8
Simple.
System is pushed to its stability limit (unstable situations can result!). Method can be time consuming (esp. if a large number of trials is required or if the process dynamics are slow) and may be expensive (due to lost productivity/ poor product quality). Not applicable to processes which are open-loop unstable. Simple processes might have no ultimate gain.
Example: See exercise 11.
- 33 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
GCL (s) =
Y(s) GCG , = Yset (s) 1 + GCG GCL (s) 1 G(s) 1 - GCL (s)
GC (s) =
Yset(s)
E(s) -
GC(s)
Controller
U(s)
G(s)*
Process
Y(s)
*Transfer functions of all loop components (e.g. process, measurement) are lumped into one single block, resp. transfer function G(s)
- 34 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
1 , 1+ TCL s
CL s
1 1 1 1 = G 1 + TCL s - 1 G TCL s
(II).
Integral action required to eliminate offset (specification of unity gain). a) First order process: G(s) = (II) then gives GC (s) = 1+ Ts 1
K
K . 1 + Ts
= T 1 1+ K TCL Ts
TCL s
T , TI = T , K TCL
i.e. set integral time TI equal to the process time constant, adjust KC to obtain the desired dynamics of the closed-loop process (i.e. TCL). b) Second order process: G(s) = K , (1+ T1s) (1+ T2s) where T1 is the larger (dominant) and T2 the smaller process time constant.
K TCL s = T1 + T2 K TCL T T 1 1 + + 1 2 s (T1 + T2 ) s T1 + T2
For T1>>T2: TI T1, TD T2, i.e. set integral time TI equal to the dominant process time constant to compensate for the dominant process lag. Tuning procedure thus is reduced to the adjustment of the controller gain KC. c) First order + dead time process: G(s) =
(II) gives GC (s) =
T 1 TDT s 1 + e K TCL Ts
K e TDT s . 1 + Ts
Unrealizable controller since it requires knowledge of the future. Specified closed-loop response must be delayed by one process dead-time.
- 35 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
e TDT s 1 + TCL s
1 e TDT s G 1 + TCL s - e TDT s
(I) GC (s) =
a) First order + dead time process: G(s) = (III) gives GC (s) = 1 +Ts s TDT
Ke
(IV) cannot be implemented with analogue devices but needs to be converted into standard PID-form:
1) Taylor series expansion (drop all but the first two terms):
e TDT s 1 TDT s Valid, as long as dead time is small compared to the closed loop response speed. For a given dead time the error thus increases with the speed of response.
K (TCL + TDT )
, TI = T .
Set integral time TI equal to the process time constant. Increase in dead-time results in a reduction of controller gain KC. Closed loop response will deviate from the specified first order response, as the controller gain (and thus the speed of response) increases.
T . K TDT
s s
Closer to the true exponential expression Can be used for all dead time systems. Recommended for long dead-times, i.e. when TDT > T/4.
T T , TI = T, TD = DT . K (TCL + TDT ) 2
The application of the controller synthesis approach to various systems is demonstrated in exercises 9, 11 and 12.
- 36 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
yS
G(s) =
K e TDT s , 1 + Ts
with K = y/u
(where y = yS y0),
y0 t TDT T
and TDT given by the intersection of the tangent through the inflexion point with the time axis.
where T and TDT denote the time constant and dead time of a 1storder plus deadtime approximation. KC is proportional to T/TDT and 1/K, TI is proportional to TDT, TD/TI typically takes on a value of 0,1 - 0,3. A lower gain is used, when integration is included in the controller. Addition of derivative mode permits a higher gain (also valid for 0).
Only a single experimental test is necessary (no trial and error). Controller settings are easily calculated.
Settings tend to result in an oscillatory response.
- 37 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Determination of T:
yS
A2 A
y(t) y
A1
y0 0 T t 0 T t
Figure 13: Alternative ways to find the sum of time constants based on the process response curve.
The area between the response curve y(t) and its steady state value yS equals the area of the rectangle given by T and y (= K if a unit step input was introduced, i.e. u = 1), i.e.
A=
(y S - y(t)) dt = y T
A1 = A2 (both areas as given in figure 13)
Determine parameters of a 1st order + deadtime approximation as shown in figure 12 T can be obtained as T = T + TDT .
Alternative way to obtain a 1storder+deadtime approximation which is closer to the original step response: Find exponential response curve such that two points (t1, y1), (t2, y2) of the original step response are met. Step response of original system
y(t) y2
Approximation
T=
t 2 t1 , y y 1 ln y y 2
y1 y1 0 t1
y2
TDT
y = T ln 1 i + t i , where i = 1 or 2. y
t2
- 38 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Assume the system to be of 2nd order with equal time constants T1 = T2 = T/2, K . i.e. G(s) = (1 + T 2 s)2 Choose controller parameters, such that a fairly underdamped closed loop system is obtained, e.g. with a damping d = 0,7. Controller settings can be calculated according to the following tuning rules: Controller P PI PID KC
1/K 0,5/K 1/K
TI
0,5 T 0,66 T
TD
0,16 T
The application of tuning relations to various systems is demonstrated in exercises 11 and 12.
- 39 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Typical flow characteristic curves (inherent flow characteristic), where KV denotes the volumetric flow rate [m/h] of water for standard conditions ( = 0 =1000 kg/m; p = 1 bar). Quick opening flow characteristic (1) Maximum change in flow rate at low valve travels with a fairly linear relationship. Additional increases in valve travel give sharply reduced changes in flow rate. Used primarily for on-off service.
100 %
1 kv kvs
Linear flow characteristic curve (2) 3 Flow rate is proportional to the valve travel. For constant pressure drop, the valve gain s 0 100% (= ratio of an incremental change in flow rate smax to incremental change in valve plug position) will be the same at all flows. Commonly specified for liquid level control and for flow control applications requiring constant gain. Equal percentage flow characteristic (3) Equal increments of valve travel produce equal percentage changes in the existing flow. Change in flow rate is proportional to the flow rate just before the change in valve position is made. Generally used on pressure control applications and on other applications where a large percentage of the pressure drop is normally absorbed by the system itself, with only a relatively small percentage available at the control valve or highly varying pressure drop conditions can be expected.
Which characteristic is obtained in service where the pressure drop varies (installed flow characteristic), depends on the contribution of the control valve to the overall pressure drop (characterized by ):
& V = & Vmax 1 k 2 1 + vs 1 k v (s)
where = p valve .
p
- 40 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
s smax
s smax
Rules of thumb: For < 0,2, i.e. pressure drop due to the valve is considerably smaller than that due to the plant, a valve with an equal percentage flow characteristic should be chosen. For > 0,6, i.e. pressure drop due to the valve is bigger than that due to the plant, a valve with an linear flow characteristic is more adequate.
Dynamics are very fast (order of seconds). Essentially no dead time. High frequency noise due to e.g. turbulence/ pump vibration.
transducer
actuator
orifice
control valve
- 41 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Choice of controller
No use of derivative action. (P)I Controller with a wide proportional band (i.e. low gain KC) and - in accordance to the controller synthesis approach - a low value of integral time TI. Note: Derivative mode amplifies noise while integration tends to average the response thus minimising noise. Notable exception: e.g. flow control of a condensate throttled reboiler: Vapour flow depends on the rate of condensation Flow rate can be varied by changing the area of heat transfer (i.e. by raising/ lowering the liquid level) in the reboiler. Flow control loop has much different controller tuning constants then mentioned before.
- 42 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Integrating nature (non self-regulating process!). Increase in the controller gain can bring an increase in system stability. Relatively high-gain controller can be used without concern about instability of the control system. In many control applications the liquid storage tank is used as a surge tank to damp out fluctuations in the inlet stream. Tight control is not necessary (level only to be kept between hmin and hmax).
Fin
E(s) -
GC(s)
F(s) Process*
H(s)
* Process can be described by G(s) = K/s with K depending on the cross-sectional area.
H(s) =
Controller
K K (Fin (s) F(s)) = Fin (s) GC (s) (Hset (s) - H(s)) 14 244 4 3 s s E(s) K K K H(s) 1- GC (s) = Fin (s) - GC (s) Hset (s) s s s K/s (-K)/s GC H(s) = Fin (s) + Hset (s) 1 + (-K)/s GC 1 + (-K)/s GC
How does the closed-loop process respond to a step change in the set point?
- 43 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Closed-loop response to a step change in the incoming flow (from Fin,1 to Fin,2):
In steady state: F = Fin Initial state: Choose controller bias value equal to incoming flow to eliminate offset:
F(t 0 ) = F = Fin,1
!
!
Without control
P-Control eS
F = Fin,1 t0 t
P-control: MV increases according to the error e, until F = Fin,2 - 1/K C H(s) K/s = = Fin (s) 1 + ( K)/s K C 1 1/K CK s Steady-state error eS = hset hS proportional to -1/KC. PI-control: Controller continuous to increase the value of F beyond Fin,2 to drive the level back to its set-point. Elimination of steady state error. Amplification of flow rate changes. When multiple tanks are connected in series, amplification increases.
Fin
- 44 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Choice of controller
Derivative mode normally not employed, since level measurements often contain noise due to the splashing/ turbulence of the liquid entering the tank. Integral action not necessary, if small offsets in the liquid can be tolerated. P(I) Controller where the proportional band depends e.g. on control valve sizes. Conservative controller settings to avoid large/ rapid fluctuations in the exit flow rate.
Pressure control loops can very from very tight, fast loops (almost like flow control) to slow averaging loops (almost like liquid level control). Gas pressure process is self-regulating: The vessel/ pipeline admits more feed, when the pressure is too low, reduces the intake, when the pressure becomes too high.
Choice of controller
PI Controller with the integral time TI depending on the open loops dynamics. Usually no derivative action needed (process response time small compared to other process variables.)
Figure 20: Fast pressure loop: Valve throttling the flow of vapour from a vessel.
Figure 21: Slow pressure loop: Pressure is held by throttling the water flow to a condenser.
- 45 -
Modelling of Systems and Processes: Model Based Control Systems Controller Tuning
Wide variety of processes and equipment involving heat transfer. General guidelines for controller difficult to state. Due to the presence of time delays and/or multiple thermal capacities there will usually be a stability limit of the controller gain.
Choice of controller
PID Controllers commonly employed to obtain more rapid responses. Choose integral time TI equal to dominant time constant T and derivative time TD T/4.
steam reactor
steam
heating jacket
- 46 -