Finn Haugen. Telemark University CollegeNational Instruments Confidential 1 Learning PID control essentials with LabVIEW By Assistant Prof. Finn Haugen, Telemark University College, Norway Contents of the presentation: • Description of the case (student assignment): Temperature control of heated air tube • Block diagram of control system • Performance indexes • Control strategies (Blind; Manual feedback; Automatic feedback.) • Measurement noise • Easy controller tuning • Gain scheduling (adaptive control) • Feedforward control (added to feedback control) National Instruments Confidential 2 Case: Temperature control of air heater with LabVIEW PWM indicator Pt100 sensor (secondary) AC/DC Pulse Width Modulator (PWM) Fan Air tube Heater RS232 Serial Air Pt100-mA transducer Fan speed adjust Pt100 sensor (primary) 3 x Voltage AI (Temp 1, Temp 2, Fan indication) 1 x Voltage AO (Heating) National Instruments Confidential Laptop PC with LabVIEW Fieldpoint FieldPoint (Dual Channel Voltage I/O) 3 Block diagram of control system LabVIEW [0%, 100%] Setpoint (scaled) ySP Setpoint ySP0 [20oC, 70oC] Scaling [0%, 100%] Filtered measurement signal ymf [%] Gain scheduling Control error e Feedback uPID controller [%] Scaling Feedforward controller DAC [2V, Fan 10V] speed DisturSenbances sor Fan speed Ambient temperature Off On uff Man u0 u Scaling DAC [0V, 5V] Control signal Auto [0%, 100%] Lowpassym filter Measurement signal ym Scaling [0%, 100%] ADC Process y [oC] TransSensor [20oC, [2V, ducer [4mA, 70oC] 10V] 20mA] Effective measurement noise, n The students will implement this system from scratch in LabVIEW. National Instruments Confidential 4 Some performance indexes of control systems Maximum of absolute value of control error: | e |max Should be small or large? Mean of absolute value of control error: (Almost the same as the popular IAE index – Integral of Absolute value of control Error.) 1 | e |mean tf ti tf ti | e | dt Should be small or large? Mean of absolute value of time-derivative of control signal: (Inspired by optimal control, e.g. MPC, where the objective function includes the variation of the control signal.) | du dt mean | 1 tf ti tf ti | du dt | dt National Instruments Confidential Should be small or large? 5 Implementation of performance indexes The three performance indexes defined above can be implemented as follows. • The maximum control error index: | e |max can be implemented with the following code: National Instruments Confidential 6 Implementation of performance indexes cont. The mean of absolute error index: 1 | e |mean tf ti tf ti | e | dt can be implemented with the following code: (Alternatively, could have used the MeanPtByPt.vi.) National Instruments Confidential 7 Implementation of performance indexes cont. And the control signal time-derivative index: | du dt mean | 1 tf ti tf ti | du dt | dt can be implemented with the following code: (Alternatively, could have used the MeanPtByPt.vi.) National Instruments Confidential 8 Trying out three control strategies The three performance indexes (|e|max, |e|mean, |du/dt|mean) are recorded for each of the below control strategies: • Blind control, i.e. control with a fixed control signal • Manual feedback control, i.e. the human (student) does the control • Automatic feedback (PID) control, i.e. the computer does the control For the PID control: • PID settings: Kc = 40,8; Ti = 8.0s; Td = 2.0s. (found from the LabVIEW PID Autotuning.vi with ”fast response”). • The meas. filter is lowpass 2. order Butterworth with bandwidth 0.4Hz. National Instruments Confidential 9 Trying out three control strategies cont. The process is operated as follows: • Setpoint = 40% (fixed) • Fan speed = 60% (initial value) • A disturbance change: Increasing the fan speed for about 10 sec from 60% to 100% and then back to 60% again. • Temp1 sensor in the outmost position • Duration of experiment: 60 seconds National Instruments Confidential 10 Trying three control strategies cont. Blind control: Control Manual feedback: Automatic feedback (PID): Setpoint Filtered temp National Instruments Confidential 11 Trying three control strategies cont. Control Blind control: Manual strategy: Perform. index: feedback: Automatic feedb (PID): |e|max 0.78 % 0.86 0.44 |e|mean 0.39 0.21 0.12 |du/dt|mean 0 4.54 7.00 Observation: Automatic feedback (PID) gives smallest max and mean control error, but the control action is the most aggressive! This is general, too. National Instruments Confidential 12 The problem with measurement noise In a feedback control system measurement noise is propagated via the controller to the control signal, causing variations in the control signal. The derivative term of the controller amplifies these variations. These variations can be reduced in several ways: • Using a measurement lowpass filter, e.g. IIR filter or FIR filter. (The FIR filter on the PID Control Palette is inflexible. The Butterworth PtByPt filter on Signal Processing Palette is flexible and easy to tune.) • Setting the derivative gain to zero, i.e. using PI in stead of PID (Block diagram is repeated on next slide for easy reference.) National Instruments Confidential 13 Block diagram of control system (repeted) LabVIEW [0%, 100%] Setpoint (scaled) ySP Setpoint ySP0 [20oC, 70oC] Scaling [0%, 100%] Filtered measurement signal ymf [%] Gain scheduling Control error e Feedback uPID controller [%] Scaling Feedforward controller DAC [2V, Fan 10V] speed DisturSenbances sor Fan speed Ambient temperature Off On uff Man u0 u Scaling DAC [0V, 5V] Control signal Auto [0%, 100%] Lowpassym filter Measurement signal ym Scaling [0%, 100%] ADC Process y [oC] TransSensor [20oC, [2V, ducer [4mA, 70oC] 10V] 20mA] Effective measurement noise, n The students will implement this system from scratch in LabVIEW. National Instruments Confidential 14 Measurement noise cont. The figure below shows the PID control signal in four situations: • No measurement filter. (Max amplitude is due to the LSB of the 12 bits ADC!) • Using the 5. order FIR filter on the PID Control Palette • Using an IIR filter in the form of a 2. order Butterworth filter with bandwidth 0.4Hz (tuned by trial and error) • IIR filter, and setting derivative time to zero, i.e. PI control No filter FIR, PID contr IIR, PID contr IIR, PI contr No surprise that PI is more popular than PID in industry! National Instruments Confidential 15 Easy controller tuning Two easily available tuning tools or procedures in LabVIEW: (Tuning based on estimated process model is in advanced assignments.) • The PID Autotuning.vi, which invokes a tuning wizard. The tuning principle is to automatically change the setpoint stepwise, and to calculate the controller parameters from the response. The autotuner requires that the control loop is stable initially (with P, PI or PID controller). • Åstrøm-Hägglund’s relay-based tuning method with the PID Advanced.vi or the PID.vi. (This method is basically a practical implementation of the Ziegler-Nichols’ ultimate gain method.) National Instruments Confidential 16 Controller tuning cont. PID Autotuning.vi The wizard is opened when the autotune? input is TRUE. When the tuning is finished, the new PID settings are written to the PID_gains local variable. The FALSE case above (which is active when the tuning is finished), contains the PID Advanced.vi which is used in normal operation. National Instruments Confidential 17 Tuning cont. One of the dialog windows of the PID Autotuning.vi wizard is shown in the figure: Results: Kc = 40,8 Ti = 8.0 s Td = 2.0 s. Representative setpoint step response after tuning: Seems ok :-) National Instruments Confidential 18 Tuning cont.: Relay-based tuner In the tuning phase, the PID controller must be replaced by an On/Off-controller, creating sustained oscillations in the loop. How to turn the PID controller into an On/Off-controller: • Kc very large, e.g. 1000. • Ti = Inf • Td = 0 • The control signal amplitude, A, is set via the output range input to the LabVIEW PID functions, since A = (umax – umin)/2. Assume: • The oscillatory control error amplitude is measured as E. • The period of the oscillations is measured as Pu. By representing the square wavy controller signal by fundamental Fourier series term, the ultimate gain (relay gain) is Kcu = (Ampl out (by Fourier))/(Ampl in) = (4*A/π)/E The PID setting can now be found from the Ziegler-Nichols’ formulas. National Instruments Confidential 19 Relay-based tuning cont. A = 20% Pu = 12 sec Result from an experiment: A = 20 %. E = 0.4 %. Pu = 12 sec. Thus, Ku = 4*A/(pi*E) = 63.7. PID setting: Kc = 0.6* Kcu = 38.2. Ti = Pu/2 = 6 s. Td = Pu/8 = 1.5 s. 2E = 0.8% (The PID Autotune.vi gave Kc = 40,8; Ti = 8.0 s; Td = 2.0 s – not so different.) National Instruments Confidential 20 Gain scheduling (adaptive control) The problem: It can be shown both experimentally and mathematically (using a simplified model) that the gain and the transport delay of a flow process increases as the flow descreases. If the (temperature) controller is tuned at a high flow rate, the control system may get poor stability if the flow rate decreases. The figures to the right illustrate this for the air heater. The PID controller was tuned at flow rate 100%: Kc = 42.0; Ti = 5.0s; Td = 1.25s. This control system becomes unstable at the minimum flow rate (3.2%). National Instruments Confidential 21 Gain scheduling cont. One simple solution: Since the stability of the control system depends on the flow rate, let us try varying the controller parameter settings as functions of the flow rate. This is implemented using the PID Gain Schedule.vi. The scheduling is based on three PID settings each found by using relay-based tuning: Flow 67%: Kc = 24.1; Ti = 8.0s; Td = 2.00s. Flow 33%: Kc = 30.6; Ti = 7.0s; Td = 1.75s. Flow 3%: Kc = 34.7; Ti = 5.5s; Td = 1.38s. National Instruments Confidential 22 Gain scheduling cont. The result: The figures to the right illustrates that the control system now has good stability for the minimum flow (and for the maximum flow). An alternative solution: Conservative tuning Tune the controller at one specific flow rate, and keep the controller settings fixed for all flow rates. For which flow rate? Any drawback? (This solution is not demonstrated here.) National Instruments Confidential 23 Feedforward control (+ feedback control) Variations of the air flow act as disturbances to the process. The feedback controller tries to compensate for such variations using the temperature measurement. Can we obtain improved control by also basing the control signal on measured air flow, which is here available as the fan speed indication? Let us first try without feedforward. The figure shows ordinary PID control as the fan speed was changed from minimum to maximum, and back again. Performance indexes: |e|max = 1.01. |e|mean = 0.36. National Instruments Confidential 24 Feedforward cont. Now, let us try feedforward from fan speed (air flow). (Block diagram is repeated on next slide for easy ref.) A number of corresponding values of fan speed and control signal was found experimentally. Temperature setpoint was 40 deg C. The feedforward control signal, u_ff, was calculated by linear interpolation with Interpolate 1D Array.vi, and was added to the PID control signal to make up the total control signal: u = u_PID + u_ff. Performance indexes: |e|max = 0.27 (vs 1.01). Much better! |e|mean = 0.073 (vs 0.36). Much better! National Instruments Confidential 25 Block diagram of control system (repeted) LabVIEW [0%, 100%] Setpoint (scaled) ySP Setpoint ySP0 [20oC, 70oC] Scaling [0%, 100%] Filtered measurement signal ymf [%] Gain scheduling Control error e Feedback uPID controller [%] Scaling Feedforward controller DAC [2V, Fan 10V] speed DisturSenbances sor Fan speed Ambient temperature Off On uff Man u0 u Scaling DAC [0V, 5V] Control signal Auto [0%, 100%] Lowpassym filter Measurement signal ym Scaling [0%, 100%] ADC Process y [oC] TransSensor [20oC, [2V, ducer [4mA, 70oC] 10V] 20mA] Effective measurement noise, n The students will implement this system from scratch in LabVIEW. National Instruments Confidential 26 Conclusions • LabVIEW with PID Toolkit offers a flexible and user-friendly environment for students to learn practical PID control. • Practical control is best learned in (practical) labs because the students will then experience important realistic problems and phenomena related to e.g. noise. National Instruments Confidential 27