Elmo Position Control

High Order Control Design Advantage Over PI and PID Controllers reference
Yaniv O.1, Theodor Y. and Safonov S.

Abstract A robotic application is used to show that advanced controllers are much capable than PI controllers. They can achieve higher bandwidth, lower settling time and better disturbance rejection. The increased performance costs little in sensor noise amplification. We show using true-life design examples that advanced control algorithms improve equally well both speed and position controllers. Introduction Consider an electrical motor with shaft angle the shaft speed,

(t ) , driven by the current i (t ) . We want

(t ) , to follow a given trajectory, T (t ) . For this purpose, we embed the

motor in a feedback structure as described schematically in Figure 1. The controller in Figure 1 generates a correcting current command, i (t ) , so as to keep the speed error,

e(t ) , minimal.

T +

Controller

Amplifier

Motor
+Load

Sensor noise
Sensor

Figure 1: Speed control feedback structure around a motor

The controller is required to minimizing the speed error and in the same time the synthesized current command must remain smooth enough so that (i) no excessive stresses will shorten the system life, and (ii) the current amplifier will be able to effectively follow the current command. The controller design must consider both small and large signals behavior. The small signal design cares for the behavior when the tracking error is small, and thus the required correction current (torque) is within the amplifier limits. Large signal (nonlinear) design must maintain good stability and performance while the current (torque) requirement goes beyond the amplifier limits. Out of range current (torque) requirements may develop due to extreme reference signal changes or due to extreme disturbances. This article focuses on the small signal (linear) design.
Address for correspondence: O. Yaniv, Elmo Position Control, Shidlovskey 1, Yavne, 81101, POB 13081, Israel and Faculty of Eng. Tel-Aviv University, Tel-Aviv 69978, Israel. oyaniv@elmo.co.il
1

-1-

Elmo Position Control

The controller must have a parameterization, so that users will be able to tune it to their specific applications. The most common controller parameterizations are P, PI and PID. A P (proportional) controller keeps the current i (t ) proportional to the speed error,

T (t ) (t ) . A PI controller generates i (t )

as the sum of two terms. The P term which is

proportional to the speed error, and the I term which is proportional to the integral of the speed. A PID controller is a PI controller plus the D term, that is, a term proportional to the speed error derivative. The worst drawback of PI and PID controllers is their poor high frequency attenuation. Some commercial motion controllers add low-pass filters to their PI controllers, to improve the high frequency attenuation. The traditional P, PI or PID controller have one big advantage they are very simple, and a technician can tune them effectively using simple "cut and try" methods. These simple controllers suffice for simple applications moderate or low performance requirements, and good enough mechanics and sensors. A very simple control problem is, however, a symptom of too generous mechanics and sensors design. More complicated controllers can push the tracking and disturbance attenuation performance to the physical limits of the system. An advanced controller can get the desired performance out of a lighter structure, or within degraded, cheaper sensors. For the same mechanics-sensors set, an advanced controller can increase the speed range in which accurate enough motions are possible. We use the term advanced-controllers for controllers of almost free structure and order. Advanced controllers do not preserve the PI simplicity. They have many parameters, and require an automated design suite for effective tuning. The decision to use advanced controllers is psychologically not easy. You have to trust the tuning suite of the Ph.D. guy better than you trust your senses. Moreover, you have to believe that the tuning suite does take appropriate design margins, so that you wont have vibrations when the load changes a bit. In this paper, we compare the performance of advanced and traditional controllers, controlling a robotic arm. The advanced controllers are shown to do much better than the P PI or PID controllers. Section 2 compares PI controllers and advanced controllers, by a laboratory test. Section 3 extends the comparison of Section 2 to frequency domain. Section 4 shows similar comparison results for cascaded position control. Embedding the speed controller of Figure 1 in an outer position feedback loop makes a cascaded position controller see Figure 2. The position controller is required to follow a trajectory PT (t ) .

d dt

Position Controller

+ + -

PT

Speed Controller

Plant

dt

Figure 2: Cascaded position control feedback loop

-2-

Elmo Position Control

Speed Control Comparison by Laboratory Tests Our design example deals with a two-join robot, see Figure 3. The robot is lightweight, and pays for the lightweight with high link compliances. An electrical motor drives each joint. For each motor, a tachometer measures the motor shaft speed and an encoder measures the motor shaft angle. The upper motor (motor 1) drives the internal link, and the lower motor (motor 2) drives the external link. For this robot, a PI speed controller proved useless, since the robot became unstable for very low gains. We helped the PI with an additional high frequency low-pass pole. The PI plus low-pass performance shown in the next figures is probably better than what an experienced technician could achieve. This is since this robot exhibits high coupling between its articulated axes. If one use traditional PI tuning methods to optimize each axis when the other axis is inactive, the integrated system may become unstable or might loose some of its gain and phase margins due to the two axes interaction. If on the other hand, one use traditional PI tuning methods to optimize each axis when the other axis is active, stability of the integrated system is guaranteed but again the closed loop might loose some of its gain and phase margins.

Motor 1

Link 1

Motor 2 Link 2

Figure 3: Robot for laboratory tests

a two joint robot with two motors, two tachometers and two encoders.

-3-

Elmo Position Control

The robot was tested for several speed reference commands. Figure 4 and Figure 5 show the step response of PI plus low-pass controller and of an advanced controller against the reference step, for motors 1 and 2, respectively.

1000 800 Cnt/sec. 600 400 200 0 -200 0 10 6 Am pere 2 -2 -6 -10 0 0.1 0.2 0.3 Sec. 0.4 0.5 PI+low pass Advanced 0.6 0.7 0.1 0.2 0.3 0.4 0.5 reference PI+low pass Advanced 0.6 0.7

Figure 4: Comparison between PI plus low-pass controller and advanced controller. The step command to motor 1 is 600[cnt/sec], and to motor 2 zero. Clearly for motor 1 (Figure 4), the tracking error, rise time and settling time of the advanced controller are much lower than the corresponding results of the PI plus lowpass. The rise time of the advanced controller is 0.017seconds, about 43% of the 0.04 seconds rise time of the PI plus low-pass. The same relation holds for the settling time. The prices for the higher performance of the advanced controller are twice the current peak and larger high frequency noise.

-4-

Elmo Position Control

1000 800 Cnt/sec. 600 400 200 0 -200 0 3 2 Am pere 1 0 -1 -2 0 0.1 0.2 0.3 Sec. 0.4 0.5 0.6 0.7 PI+low pass Advanced 0.1 0.2 0.3 0.4 0.5 reference PI+low pass Advanced 0.6 0.7

Figure 5: Comparison between PI plus low-pass controller and advanced controller. The step command to motor 2 is 600[cnt/sec]; and to motor 1 is zero. For motor 2, the racking error, (Figure 5), rise time and settling time of the advanced controller are much lower than that of the PI plus low-pass. The rise time of the advanced controller is 0.02 seconds, about 20% the 0.1 seconds rise time of the PI plus low-pass. The same relation holds for the settling time. The price for the higher performance of the advanced controller is again the current peak. In most real applications however, the reference command is smooth and the peak current is dictated by the reference command rather then by the controller. This is emphasized by the next comparison, which compares the controllers for acceleration-limited step response. Figure 6 and Figure 7 compare between our advanced controller and the PI plus lowpass controller for a smooth acceleration limited speed command. Again, the tracking error for the advanced controller is much lower than for the PI plus low-pass. The overshoot of the advanced controller is by far lower, and the current consumption is about the same for both controllers.

-5-

Elmo Position Control

3000 2200 C nt/sec 1400 600 -200 0 10 6 Am pere 2 -2 -6 -10 0 PI+low pass Advanced 0.1 0.2 0.3 Sec. 0.4 0.5 0.1 0.2 0.3 0.4 0.5

reference PI+low pass Advanced

0.6

0.7

0.6

0.7

Figure 6: Comparison between PI plus low-pass controller and advanced controller. Trajectory command to motor 1 is 2000[cnt/sec], acceleration limitation 20000[cnt/sec^2]. Motor 2 is commanded to stop.
3000 2200 C nt/sec 1400 600 -200 0 3 2 Am pere 1 0 -1 -2 0 0.1 0.2 0.3 Sec. 0.4 0.5 0.6 0.7 PI+low pass Advanced 0.1 0.2 0.3 0.4 0.5 0.6 0.7

reference PI+low pass Advanced

Figure 7: Comparison between PI plus low-pass controller and advanced controller. Trajectory command on motor 2 is 2000[cnt/sec], acceleration limitation 20000[cnt/sec^2]. Motor 2 is commanded to stop.

-6-

Elmo Position Control

Speed Control Comparison by Frequency Domain Analysis In section 2 the robot's tracking performance for different controllers has been studied. We subjected the controller to abrupt reference waveforms, which expose the transient behavior of the closed loop. The time domain tests of Section 2 show the final result, but they offer no explanation to the difference in the results achieved. The frequency domain analysis of this section grants insight to questions such as the feasibility of better designs, and the design margins taken. The frequency plots provide an estimate for settling time and overshoot. This estimate confirms the result of Section 2. Open Loop The robot has two motors and four sensors, two tachometers and two encoders, generating eight transfer functions from the current commands introduced into each of the motors to each of the sensors. Let pij denote the transfer function from current introduced into motor j to the integral of the angle (integral of speed) measured by the tachometer on the shaft of motor i; and rij denote the transfer function from the current command introduced into motor j to the encoder coupled to the shaft of motor i.

Figure 8 and

Figure 9 depict these eight discrete Bode plots.

p -20 -40 -60 -80 -100 -120 -20 -40 -60 -80 -100 -120 1 10 dB

11

p -20 -40 -60 -80 -100 -120

12

21

22

-20 -40 -60 -80 -100 10 10 [rad/sec]

pij
2 3

-120 1 10

10 10 [rad/sec]

Figure 8: Bode plot of

, from input j to integral of the tachometer on link i

-7-

Elmo Position Control

r -20 -40 -60 -80 -100 -120 -20 -40 -60 -80 -100 -120 1 10
Figure 9: Bode plot of

11

r -20 -40 -60 -80 -100 -120

12

dB

21

22

-20 -40 -60 -80 -100 10 10 [rad/sec]

2 3

-120 1 10

10 10 [rad/sec]

rij , from input j to the encoder located on link i

Figure 8 shows that link 1 has four dominant resonances; with frequencies ranging from 200 to 3000 rad/sec. Link 2 has a dominant resonance at about 100 rad/sec. These resonance frequencies would limit the performance of any controller, but their effect on PI controllers is most marked. Advanced controllers can attenuate the resonant frequencies using notch or low-pass filters; or they can actively damp some resonant modes. The transfer functions from current commands to encoders, Figure 9, differ from the transfer function of Figure 8 since the encoders are mounted on flexible couplings, whereas the tachometers are mounted rigidly on the motor shaft. In some frequencies the coupling between the axes is so large that the reaction to current injected to motor 1 on its shaft, is much lower than the reaction of the shaft of motor 2. The ratio between these reactions is depicted in Figure 10. For currents injected to motor 1, whose spectral densities are mainly around 200Hz and/or 300Hz, the tachometer located on motor 2 reads a signal up to 5 times larger than the tachometer located on motor 1. A similar phenomenon, but much lower in size, happens when motor 2 is driven.

-8-

Elmo Position Control

p /p
21

11

30 15 0 -15 -30 -45 1 10 30 15 0 -15 -30 -45 1 10 10

2

dB

2 10 p 12 /p 22

10

dB

10

[rad/sec]
Figure 10: Bode plot relative cross talk. The upper plot is the ratio of tachometer 2 to tachometer 1 due to current injected to motor 1. Lower plot is the ratio of tachometer 1 to tachometer 2 due to current injected to motor 2. Controller Design Advanced control design techniques are based on the controlled plant transfer function; the process of achieving this transfer function is called identification. Following the plant identification process, the control engineer designs a controller using his own experience, knowledge and skills. The major properties of a closed loop feedback system can be concluded from the open loop transfer function, for example, rise time, settling time, robustness to plant changes, amplification of sensor noise, and if it is possible to improve the closed loop performance. Bode plots of the open loop for motor 1 and motor 2 for the PI plus low-pass controllers are shown in Figure 11 and Figure 12, respectively. Bode plots of the open loop for motor 1 and motor 2 for the advanced controllers are shown in Figure 13 and Figure 14, respectively.

-9-

Elmo Position Control

30 10 dB -10 -30 -50 1 10 90 Phase[deg] 0 -90 -180 -270 -360 1 10 10

2

10

10

10

[rad/sec] Figure 11: Open loop Bode plot of motor 1 (PI and low-pass)

30 10 dB -10 -30 -50 1 10 90 Phase[deg] 0 -90 -180 -270 -360 1 10 10

2

10

10

10

Figure 12: Open loop Bode plot of motor 2 (PI and low-pass)

[rad/sec]

-10-

Elmo Position Control

30 15 0 -15 -30 -45 1 10 dB

10

10

60 0 -60 -120 -180 -240 -300 -360 1 10

Phase[deg]

10

10

Figure 13: Open loop Bode plot of motor 1 (advanced controller)

[rad/sec]

30 15 0 -15 -30 -45 1 10 dB

10

10

60 0 -60 -120 -180 -240 -300 -360 1 10

Phase[deg]

10

10

[rad/sec]

-11-

Elmo Position Control

30 15 0 -15 -30 -45 1 10 dB

10

10

60 0 -60 -120 -180 -240 -300 -360 1 10

Phase[deg]

10

10

Figure 14: Open loop Bode plot of motor 2 (advanced controller)

[rad/sec]

Comparing Figure 11, Figure 12, Figure 13 and Figure 14, we have the following conclusion: 1. The bandwidth of the advanced controller for motor 1 is about 15Hz, almost 2.5 larger than the 6.2Hz of the PI controller. 2. The bandwidth of the advanced controller for motor 2 is about 13Hz, almost 2.5 larger than the 5.7Hz of the PI controller. 3. The low frequency disturbance attenuation of the advanced controller for motor 1 is 5 times better than that of the PI controller. 4. The low frequency disturbance attenuation of the advanced controller for motor 2 is 10 times better than that of the PI controller. We present a Nichols chart in order to convince the reader that a fair comparison was made, in the sense that similar gain and phase margins were taken for the PI plus lowpass and the advanced controllers. Figure 16 compares the open loop on motor 2 for advanced controller (left) and PI plus low-pass (right). Clearly both have the same phase and gain margins, about 8dB and 35deg. Figure 15 is the same comparison for motor 1. The margins of the advanced controller are similar. It is impossible to increase the gain of the PI (right) and maintain the same margins since: (i) the phase margin will be less than the required 35deg and (ii) the resonance whose gain is about 9dB is highly phase uncertain.

-12-

Elmo Position Control

30 20 10 dB
co

30 20 10 dB

0 -10 -20

=14

0 -10 -20

co

=6.2

-30 -360 -270 -180 -90 Phase[Deg]

-30 -360 -270 -180 -90 Phase[Deg]

Figure 15: Comparison by open loop Nichols plot of speed controller of motor 1, left advanced, right PI plus low-pass. Crossover requencies are 14[Hz] and 6.2[Hz], respectively.

30 20 10 dB
co

30 20 10 dB

0 -10 -20

=16

0 -10 -20

co

=5.2

-30 -360 -270 -180 -90 Phase[Deg]

-30 -360 -270 -180 -90 Phase[Deg]

Figure 16: Comparison by open loop Nichols plot of speed controller of motor 2, left advanced, right PI plus low-pass. Gain and phase margins are about the same, crossover frequencies are 16[Hz] and 5.2[Hz], respectively.

-13-

Elmo Position Control

Cascaded Position Control Comparison Laboratory Tests measures the motor shaft speed and an encoder measuring the motor shaft angle. A cascaded position controller has been designed where the speed loop is the PI plus lowpass or the advanced controllers of sections 2 and 3 and the position controller is a simple gain. Figure 17 and Figure 18 show test results for that cascaded position controller. The comparison shows that the advanced controller tracks the reference command much better than the PI plus low-pass. The current consumed by both controllers is about the same with about the same peak value.

600 450 C nt 300 150 0 0 2 1.5 Am pere 1 0.5 0 -0.5 -1 0 0.1 0.2 0.3 Sec.
Figure 17: Comparison between PI plus low-pass controller and advanced controller.

reference PI+low pass Advanced 0.1 0.2 0.3 0.4 0.5 0.6 0.7

PI+low pass Advanced

0.4

0.5

0.6

0.7

The trajectory command for motor 1 is 500[cnt] with speed and acceleration limitation of 2000[cnt/sec] and 20000[cnt/sec^2], respectively. Motor 2 commanded to stop.

-14-

Elmo Position Control

600 450 C nt 300 150 0 0 2 1.5 Am pere 1 0.5 0 -0.5 -1 0 0.1 0.2 0.3 Sec.
Figure 18: Comparison between PI plus low-pass controller and advanced controller.

reference PI+low pass Advanced 0.1 0.2 0.3 0.4 0.5 0.6 0.7

PI+low pass Advanced

0.4

0.5

0.6

0.7

The trajectory command on motor 2 is 500[cnt] with speed and acceleration limitation of 2000[cnt/sec] and 20000[cnt/sec^2], respectively. Motor 1 commanded to stop. Conclusions We used a robotic application to compare the performance of a traditional PI-PID controller versus more advanced controllers. For this compliant mechanic system, the PI controllers were left behind the more advanced controllers, in the criteria of bandwidth, settling time and low frequency disturbance rejection. This is just another case, where to get the most out a mechanical system, PI controllers are not enough. Complex control problems deserve an advanced controller. To effectively design an advanced controller, we need a frequency domain system model, and an automated controller design system. We developed an identification & design environment that can identify the dynamics of complex mechanical systems, including inter-axis coupling. The identification results are directly fed to an automatic controller design environment, and the results of the controller design are directly fed to program the motion controller. The same environment also designs automatically the large signal control policy. Large signal advanced controllers are out of the scope of this paper they deserve their own paper.

Copyright 2001 Elmo Position Control. All rights reserved.

-15-