Proceeding of the 2004 American Control Conference FrM16.

4
Boston, Massachusetts June 30 - July 2, 2004

Design and Experimental Evaluation of a Nonlinear Position

Controller for a Pneumatic Actuator with Friction
Mark Karpenko and Nariman Sepehri

Abstract— This paper documents the development and deadband, tracking accuracy on the order of 2 mm and
experimental evaluation of a practical nonlinear position steady-state errors around 0.5 mm were reported. Ning
controller for a typical industrial pneumatic actuator that and Bone [3] employed a novel proportional plus velocity
gives good performance for both regulating and reference
tracking tasks. The system is comprised of a low-cost 5- plus acceleration control law with friction compensation
port proportional valve with flow deadband and a double-rod for high-accuracy point-to-point positioning of a pneumatic
actuator exhibiting significant friction. Quantitative feedback actuator.
theory is employed to design a simple fixed-gain PI control law The goal of this work is to develop a practical, yet
that minimizes the effects of the nonlinear control valve flows, accurate position controller for an experimental pneumatic
uncertainty in the physical system parameters and variations
in the plant operating point. Easy to implement nonlinear actuator. The experimental positioning system is comprised
modifications to the designed PI control law are then tuned of a low-cost 5-port proportional valve with appreciable
experimentally in a step-by-step fashion to reduce overshoot deadband and a linear actuator exhibiting significant fric-
and to negate the effects of the control valve deadband and tion. Whereas previous work seems to focus on either
actuator friction. Experimental results clearly illustrating the tracking or regulating performance, the primary objective
efficacy of the approach are presented.
here is to derive a control strategy that gives good fine
I. I NTRODUCTION and coarse positioning performance for both regulating and
reference tracking tasks. Towards this objective, quantitative
Due to their high force output to weight ratios, cleanliness feedback theory (QFT) [4] is first used to design the gains
and comparatively low cost, pneumatic actuators are well of a PI control law to satisfy pointwise tolerances on the
suited for a number of industrially relevant tasks ranging closed-loop frequency response. This enables the selection
from point-to-point positioning to high-accuracy servo po- of the PI control gains that work best with the pneumatic
sitioning and force control. However, complex nonlinear system.
dynamics, compressibility of air, and the parasitic effects Without modifying the designed PI gains, the relative
of actuator friction continue to make servo control of stability of the designed closed-loop system is improved
pneumatic actuators a difficult task. by replacing the ordinary integrator in the PI control law
The nonidealities associated with industrial pneumatic with a nonlinear reset integrator. Then, to negate the effects
actuators generally complicate the controller design to the of the control valve deadband and actuator friction, which
extent that it is difficult to achieve reasonable perfor- are not considered explicitly in the QFT synthesis, the
mance using easy to implement proportional-integral (PI) or following nonlinear modifications to the designed resetting
proportional-integral-derivative (PID) control schemes [1]. PI control strategy are implemented: (i) velocity error trig-
A number of authors have, however, proposed nonlinear gered integral augmentation, and (ii) set-point acceleration
modifications to conventional control laws that have been based overshoot reduction. These modifications, proposed
shown to dramatically improve the closed-loop performance previously by Sepehri et al. [5], were found here to be
of pneumatic servos. To name a few, Wang et al. [1] devised easy to tune and to afford significant improvements in the
a time-delay minimization algorithm aimed at reducing the closed-loop positioning performance of the experimental
dead time associated with static friction as well as a null pneumatic actuator.
offset compensation scheme for negating the effects of This paper thus makes the following contributions:
control valve deadband. When used in conjunction with 1) An accurate and practical nonlinear position con-
a simple PID control law and acceleration feedback the troller is developed that overcomes many of the
velocity tracking response of the system was observed to nonidealities in a typical low-cost experimental pneu-
be much improved. Hamiti et al. [2] developed an auto- matic actuator.
tuning PI control scheme to eliminate friction induced hunt- 2) A systematic approach to the fixed-gain controller
ing. Using a high-performance servovalve with negligible design is followed: the well-established QFT design
technique is exploited for selection of the best PI con-
This work was supported by the Natural Sciences and Engineering
Research Council of Canada (NSERC). troller gains. Nonlinear modifications to the designed
M. Karpenko and N. Sepehri are with the Department of Mechanical PI controller are introduced in a step-by-step fashion
and Industrial Engineering, University of Manitoba, Winnipeg, Manitoba, to further enhance the closed-loop performance.
CANADA R3T 5V6
Correspondence should be sent to N. Sepehri: 3) Experimental results clearly illustrating the efficacy
[email protected] of the approach are presented.
0-7803-8335-4/04/$17.00 ©2004 AACC 5078
 II. E XPERIMENTAL T EST R IG In (1), xp is the actuator position and vp is the actuator
velocity. P1 , P2 , V1 , and V2 are the instantaneous actua-
The test rig, upon which all experiments were carried out, tor chamber absolute pressures and volumes, respectively.
is shown in Fig. 1. The valve is a low-cost FESTO MPYE- Parameter α is a compressibility flow correction factor,
5 series 5-port three-position solenoid driven proportional which accounts for the fact that the pressure-volume work
directional flow control valve and the actuator is a FESTO process is neither adiabatic nor isothermal but somewhere
DNC series double-rod type with a 500 mm stroke. An IBM in between [7]. Ff represents the dry friction force and FL
compatible personal computer equipped with a Metrabyte signifies the externally applied load. The magnitudes of the
M5312 quadrature incremental encoder card and a DAS- static friction and Coulomb (sliding friction) were measured
16F input/output board is used to measure the displacement experimentally and were found to be approximately 40 N
of the actuator and transmit the software generated control and 20 N, respectively.
signal to the valve. As suggested by the manufacturer, the dynamics of the
control valve spool are modelled as a first-order lag where
the displacement of the valve spool is denoted by xv and u
is the control signal. In the experimental system, the control
valve saturates at 5 V input and there exists a flow deadband
covering 12% of the range of valve spool displacement.
The nonlinear equation governing the mass flow rate of
air through each control valve orifice is [8]
 C1 C wxv Pu
 d
√
T
if PPud ≤ Pcr
ṁ =
r ³ ´2
 C1 Cd√wxv Pu 1 − Pd /Pu −Pcr if Pd > Pcr
T 1−Pcr Pu
Fig. 1. Experimental test rig. (2)
r ³ ´(γ+1)/(γ−1)
γ 2
where C1 = R γ+1 .
III. N ONLINEAR M ATHEMATIC M ODEL In (2), Pd is the absolute downstream pressure, while Pu
denotes the absolute upstream pressure. The critical pressure
A schematic of the experimental test rig for modelling ratio, Pcr , which delineates between the sonic (choked) and
is shown in Fig. 2 along with the variables of interest. subsonic flow regimes, was measured experimentally and
Assuming adiabatic charging and discharging of the actuator found to be 0.2.
chambers [6], a set of nonlinear state equations that describe A suite of experiments was carried out to identify the
the dynamic system is values of the relevant parameters of the experimental test
rig and verify the mathematical model. The remaining
ẋp = vp
model parameters and their identified nominal values are
1
v̇p = (−bvp + AP1 − AP2 − Ff − FL ) summarized in Table I.
M
γRT αγP1 A IV. L INEAR T RANSFER F UNCTION M ODEL
Ṗ1 = ṁ1 − ẋp
V1 V1 (1) To derive a transfer function representation of the pneu-
γRT αγP2 A matic actuator dynamics, (2) was linearized using a Taylor
Ṗ2 = − ṁ2 + ẋp
V2 V2 series expansion about operating point o. Neglecting the
1 kv
ẋv = − xv + u
τv τv TABLE I
I DENTIFIED NOMINAL PARAMETERS OF EXPERIMENTAL TEST RIG .

actuator
xp Parameter Symbol Nominal Value
supply pressure Ps 5 bars
P1,V1, A M FL atmospheric pressure Patm 1 bars
P2,V2, A total mass of piston, rods, and load M 1.91 kg
viscous damping coefficient b 70 N·sec/m
m1 Ff m 2 piston annulus area A 10.6 cm2
xv ideal gas constant R 287 J/kg·K
temperature of air source T 300 K
ratio of specific heats γ 1.4
u pressure-volume work correction factor α 0.89
Patm Patm valve coefficient of discharge Cd 0.7
control
Ps control valve valve orifice area gradient w 22.6 mm2 /mm
signal
valve spool position gain kv 0.25 mm/V
valve first-order time constant τv 4.2 msec
Fig. 2. Schematic of experimental pneumatic actuator.

5079
 TABLE II
second and higher order terms as well as any control valve
R ANGES AND NOMINAL VALUES OF OPERATING POINT DEPENDENT
leakages, the mass flows into each actuator chamber are
AND UNCERTAIN MODEL PARAMETERS .
written as follows
∆ṁ1 = Cf 1 ∆xv − Cp1 ∆P1 Uncertain Parameter Value
(3) min nominal max
∆ṁ2 = Cf 2 ∆xv + Cp2 ∆P2
M (kg) 1.81 1.91 2.01
where ∆ denotes a perturbation from the operating point b (N·sec/m) 60 70 80
value, e.g. ∆xv = xv − xvo . Parameters Cf i and Cpi are V1o (m3 ) ×10−4 1.32 2.64 3.96
V2o (m3 ) ×10−4 1.32 2.64 3.96
known as the valve flow gain and flow-pressure coefficient, τv (msec) 3.4 4.2 5.0
respectively. Their specific values depend upon operating P1o (bars) 3.7 3.7 4.5
point pressures, P1o and P2o , as well as the operating P2o (bars) 2.3 3.7 3.7
xvo (mm) 0 0 0.125
point value of valve spool displacement, xvo . Neglecting Cf 1 (kg/sec·m) 8.0 13.6 13.6
control valve deadband and treating the effects of friction Cf 2 (kg/sec·m) 8.0 13.6 13.6
as a disturbing load, the transfer function of the open-loop Cp1 (kg/Pa·sec) ×10−10 0 0 118.6
Cp2 (kg/Pa·sec) ×10−10 0 0 51.8
system can be obtained by combining and reducing Laplace
transformations of (1) and (3). The transfer function of the
open-loop system is Fd (s)

Xp (s) = G1 (s)G2 (s)U (s) − G2 (s) [Ff (s) + FL (s)] (4) U (s) −
X d ( s) Gc ( s) G1 (s) G2 (s) X p ( s)
where + +
−
γRT kv ACf 1 (γRT Cp2 + V2o s)
G1 (s) = +
(τv s + 1) (γRT Cp1 + V1o s) (γRT Cp2 + V2o s)
γRT kv ACf 2 (γRT Cp1 + V1o s) Fig. 3. Single degree-of-freedom feedback structure.
(τv s + 1) (γRT Cp1 + V1o s) (γRT Cp2 + V2o s)
(5)
and by reducing the control loop sensitivity to plant parametric
(γRT Cp1 + V1o s) (γRT Cp2 + V2o s) uncertainty. A suitable frequency domain constraint on the
G2 (s) = (6) allowable loop transmissions, L(s) = Gc (s)G(s), is written
D(s)
in logarithmic form
with ¯ ¯ ¯ ¯
D(s) = s (M s + b) (γRT Cp1 + V1o s) (γRT Cp2 + V2o s)+
¯ L(s) ¯ ¯ TU (s) ¯
∆ log ¯
¯ ¯ ≤ log ¯
¯ ¯ (9)
αγA2 s [γRT (P1o Cp2 + P2o Cp1 ) + (P1o V2o + P2o V1o ) s] 1 + L(s) ¯ TL (s) ¯
(7) where ∆ signifies the variation in the closed-loop trans-
fer function over the entire plant set, G(s). Closed-loop
The nonlinear control valve flows, changes in the system
tracking bounds TL (s) and TU (s) were derived from the
operating point, as well as uncertainties in the measurement
relevant figures of merit for the step response of a second-
of any of the physical system parameters give rise to
order system [9]:
families of representative plant transfer functions, G(s) =
G1 (s)G2 and G2 (s). The ranges and nominal values of the 22500
TL (s) =
operating point dependant and uncertain model parameters (s + 5.7)(s + 10)2 (s + 39.3)
are summarized in Table II. (10)
11.81(s + 2)(s + 20)
TU (s) =
V. QFT S YNTHESIS OF PI C ONTROLLER (s + 2.1)(s + 15)2
Fig. 3 shows a block diagram of the closed-loop feedback The time response of TL (s) is well-damped and has a 90%
control system. Gc (s) denotes the compensator, which is rise time of 0.7 sec and 2% settling time of 1.0 sec. TU (s)
restricted in this work to have a PI structure. G1 (s) and was selected to have a quicker transient response with a
G2 (s) refer to the uncertain plant transfer functions (5) and 90% rise time of 0.2 sec, 2% settling time of 0.5 sec and
(6). Defining loop transmission L(s) = Gc (s)G1 (s)G2 (s), 2 percent overshoot.
the response of the closed-loop system is written as To ensure robust stability of the closed-loop system, the
L(s) G2 (s) following constraint on the peak magnitude of the closed-
Xp (s) = Xd (s) − Fd (s) (8) loop frequency responses is imposed:
1 + L(s) 1 + L(s) ¯ ¯
¯ L(s) ¯
where disturbing force Fd (s) = L {Ff (t) + FL (t)}.
¯ 1 + L(s) ¯ ≤ 1.24 (11)
¯ ¯
Clearly, Xp (s) varies due to G1 (s) and G2 (s) uncertainty.
The objective of QFT is to synthesize control law Gc (s) The closed-loop stability specification (11) gives mini-
L(s)
to place all the closed-loop frequency responses, 1+L(s) , be- mum gain and phase margins of 5.14 dB and 45◦ , respec-
tween lower and upper tracking bounds, TL (s) and TU (s), tively. Hence, the peak overshoot in the unit step responses
5080
should not exceed 21%. In a two degree-of-freedom ¯ QFT¯ pair present in all G2 (s).
¯ L(s) ¯ To further validate the QFT design, the unit step re-
design, prefilter F (s) is available to further shape ¯ 1+L(s) ¯
to ensure the closed-loop position responses fall within sponses corresponding to the linear and Clegg-type resetting
the specified tracking envelope despite the specification of designs were simulated. The PI gains obtained from the
slightly relaxed gain and phase margins. However, in this linear QFT design were used in both control laws. The
work, replacing the ordinary integrator with a Clegg-type Clegg integrator [10] is a control element that resets the
resetting integrator [10] achieves this result and eliminates integral signal to zero whenever the input changes sign.
the need to design prefilter F (s). Steady-state errors due to Due to this resetting action, the phase lag of the Clegg
disturbances, Fd (s), are zeroed by the integral action of the integrator is approximately 52◦ less than that of an ordinary
PI controller. linear integrator [10]. Hence, the potential of the resetting
To proceed with the controller design using QFT, it was integrator to improve system performance from the point of
first necessary to select a nominal plant Gnom (s) from view of loop stability is apparent.
the set G(s) and then compute QFT bounds, B(ω), on Referring to Fig. 5a, it is observed that the unit step
the nominal loop transmission, Lnom (s) = Gc (s)Gnom (s). responses using the linear controller are all stable and
The set of all bounds delineate regions of the Nichols chart that the maximum overshoot is approximately 29%. This
where Lnom (s) should lie in order to ensure that the closed- is larger than the specified maximum peak overshoot of
loop system performs within the specified tolerances. Fig. 21% and is due to penetration of the stability bounds by
4 shows the relevant bounds and the designed nominal loop Lnom (s). The tracking bounds are clearly not satisfied by
transmission Lnom (s). the linear design. In contrast, the peak overshoot in the unit
Referring to Fig. 4, Lnom (s) was shaped by cascading step responses using the resetting control law, Fig. 5b, is less
compensator Gc (s) = Kp + Ki /s in series with Gnom (s) than 8%. This is a significant improvement over the linear
and adjusting gains Kp and Ki so that: (i) Lnom (s) lies design and justifies the use of the resetting control action.
on or above the tracking bounds that are single valued The position responses are still seen to fall just outside the
functions of the phase angle, and (ii) Lnom (s) lies exterior tracking bounds. However, as will be seen in Section VI,
to stability bounds that encircle the critical (−180◦ , 0 dB) the experimental step responses fall well within the required
point. Suitable controller gains were found to be, Kp = 10 envelope so no further design iteration was conducted.
V/m and Ki = 26 V/m·sec.
1.4
(a)
 1.2
     1.0
% UDGVHF
0.8
position

     

upper tracking
0.6 bound
    
0.4
 /QRP MZ lower tracking
0.2
bound
RSHQORRSJDLQ G%

    

%  0
 0 0.5 1.0 1.5 2.0
     1.4
(b)
1.2
      % 
1.0
%
0.8
position

    

 0.6
% 
    
0.4
% 
 0.2
    
%
% 0
0 0.5 1.0 1.5 2.0

       time (sec)
RSHQORRSSKDVH GHJ
Fig. 5. Simulated closed-loop unit step responses: (a) linear PI control
Fig. 4. QFT bounds, B(ω), and nominal loop transmission, Lnom (jω) law; (b) resetting PI control law.
on the Nichols chart.

To ¯verify ¯the design, the closed-loop gain variation, VI. E XPERIMENTATION

¯ L(s) ¯
∆ log ¯ 1+L(s) ¯, was calculated at various frequencies. The Typical experimental step responses with the resetting PI
actual closed-loop gain variation was found to be less than controller are shown in Fig. 6. To prevent friction induced
the required tolerance at all frequencies. Thus, the pointwise hunting, the rule for resetting the integrator was modified
closed-loop tracking specification (9) is satisfied. However, slightly so that the integral was continuously reset whenever
with reference to Fig. 4, the closed stability bounds are the position error, e, fell below a certain threshold, ε = 1
slightly violated. This was a necessary design tradeoff mm. The control strategy was implemented as follows:
required to properly attenuate resonance in the closed-loop
step responses caused by an under-damped complex pole u(t) = Kp e(t) + Ki I(t) (12)
5081
 500
where 450
(a) actual
desired
( 400
I(t − ∆t) + e(t)∆t if |e(t)| > ε

position (mm)
350
I(t) = (13) 300
0 if |e(t)| ≤ ε
250
200
In (12) and (13), u is the control signal, I is the integral 150
signal and parameters Kp = 10 V/m and Ki = 26 V/m·sec 100
0 4 8 12 16 20 24
are the same proportional and integral gains obtained from 1.0
the preceding QFT design. (b)
0.5

control signal (V)

1.2
0.0
1.0
normalized position

0.8 step size (mm) -0.5

upper tracking 120
0.6 bound 80 -1.0
0.4 40 0 4 8 12 16 20 24
lower tracking
0.2 bound time (sec)
0
0 0.5 1.0 1.5 2.0 Fig. 7. Benchmark experimental reference tracking performance: (a)
time (sec) position; (b) control signal.

Fig. 6. Typical experimental step responses with resetting control law

(12).
employs a nonlinear filter that estimates the velocity error,
ėdb , caused by valve deadband and actuator friction. When
Referring to Fig. 6, it is observed that the responses fall ėdb exceeds experimental threshold ėmin = 0.005 m/sec,
within the specified lower and upper tracking bounds. The the output of the controller is brought to a level necessary
slight discrepancy between the responses of Fig. 5b and to overcome the deadband instead of waiting for the error
those of Fig. 6 is due to the combined effects of the control to accumulate. The nonlinear filter is
valve deadband and actuator static friction. Reasonable
ẋ2
speed of response (90% rise times of 0.68 sec) and steady- ėdb = (ẋp − ẋd ) 2 d 2 (14)
state positioning errors less than 1 mm were consistently ẋd + β ẋp
obtained in experiments using the proposed resetting PI Constant β = 50 was found experimentally through trial
control strategy. and error. The resulting position control algorithm was
The performance of the control system for a typical implemented as follows:
reference tracking task was examined next. The test signal
consisted of a number of S-curve trajectories covering  −u
lower −Kp e(t)
if ėdb > ėmin & u(t) > −ulower
 Ki
60% of the actuator stroke. Peak actuator velocities were


I(t − ∆t) + e(t)∆t if |ėdb | ≤ ėmin
purposefully kept small in an effort to observe the tracking I(t) = uupper −Kp e(t)
 Ki
if ėdb > ėmin & u(t) < uupper
limitations imposed by friction and control valve deadband.


0 if ẋd (t) = 0 & |e(t)| ≤ ε

The benchmark performance of the positioning system (15)
with resetting PI controller is illustrated in Fig. 7. The
integral was reset according to (13), where ε = 1.0 mm In (15), thresholds ulower = uupper = 0.65 V were selected
only when time derivative of the reference position ẋd = 0. experimentally.
Fig. 7a shows the position response. As expected, significant The performance of the positioning system with velocity
deadtime (on the order of 0.75 seconds) results from the error triggered integral augmentation is illustrated in Fig. 8.
effects of valve deadband and actuator friction. The peak As is seen, implementing velocity error triggered integral
position error is approximately 50 mm and stick-slip motion augmentation significantly reduces the response deadtime.
is apparent in the response to the small low velocity ramp However, overshoot observed during reference tracking
trajectories. Overshoot is also observed when the rate of tasks remains. To alleviate this, a braking acceleration term
change of the reference position, ẋd , is brought to zero. was added in the computation of the integral signal. For
The following modifications to the designed PI control |ėdb | ≤ ėmin , the integral is now computed as follows:
strategy were implemented next to improve the closed-loop
tracking performance: (i) velocity error triggered integral I(t − ∆t) + e(t)∆t + Ka ẍbraking ∆t (16)
augmentation, and (ii) set-point acceleration based over- where parameter Ka = 0.15 sec2 is a fixed gain tuned
shoot reduction. Velocity error triggered integral augmen- by trial and error. The braking acceleration, ẍbraking , is
tation reduces the response deadtime resulting from valve calculated from
deadband and actuator friction via nonlinear conditioning of (
the integral signal. This scheme, applied initially to improve ẍd if ẍd ẋd < 0
ẍbraking = (17)
the position response of a Unimate hydraulic robot in [5], 0 if ẍd ẋd ≥ 0
5082
Equation (17) limits the integrated accelerations to braking VII. C ONCLUSIONS
accelerations. This reduces the strength of the integral signal A practical, accurate and easy to implement nonlinear po-
when the desired velocity is approaching zero, which is sition controller for a typical industrial pneumatic actuator
when the problems with overshoot are observed [5]. with control valve deadband and significant actuator friction
has been developed and evaluated experimentally. Design of
500 a fixed-gain PI control law via quantitative feedback theory
(a) actual
450 desired minimized the effects of the nonlinear control valve flows,
400
changes in the system operating point and uncertainties in
position (mm)

350
300 the measured plant parameters.
250 Use of a Clegg-type reset integrator in the designed PI
200
150
control law was found to improve the relative stability
100 of the closed-loop system. Velocity error triggered inte-
0 4 8 12 16 20 24
1.0
gral augmentation and set-point acceleration based over-
(b) shoot reduction algorithms were implemented in a step-
0.5 by-step fashion to further enhance the reference tracking
control signal (V)

0.0
performance of the experimental pneumatic actuator de-
spite the parasitic effects of control valve deadband and
-0.5 actuator friction. The experimental results clearly illustrate
-1.0
the efficacy of the proposed fixed-gain nonlinear controller,
0 4 8 12 16 20 24 indicating that experimental regulating errors less than 1
time (sec) mm could be achieved consistently. Maximum steady errors
were observed to increase only slightly to 4 mm for more
Fig. 8. Experimental reference tracking performance with velocity error
triggered integral augmentation: (a) position; (b) control signal. demanding reference tracking tasks covering 60 percent of
the actuator stroke.
R EFERENCES
Fig. 9 shows the experimental responses of the position-
ing system using the proposed modified Clegg integrator [1] J. Wang, J. Pu, and P. Moore, “A Practical Control Strategy
with velocity error triggered integral augmentation and set- for Servo-Pneumatic Actuator Systems,” Control Engineering
Practice, vol. 7, pp. 1483-1488, 1999.
point acceleration based overshoot reduction. Parameter [2] K. Hamiti, A. Voda-Besancon, and Roux-Buisson, “Position
ε had to be increased from slightly 1.0 to 4.0 mm to Control of a Pneumatic Actuator Under the Influence of
prevent hunting when the actuator was stopped after the Stiction,” Control Engineering Practice, vol. 4, no. 8, pp.
S-curve input. Referring to Fig. 9, inclusion of the braking 1079-1088, 1996.
acceleration term in the calculation of the integral signal [3] S. Ning and G.M. Bone, “High Steady-State Accuracy Pneu-
matic Positioning System with PVA/PV Control and Friction
has removed the overshoot and has greatly improved the Compensation,” IEEE International Conference on Robotics
reference tracking ability of the servopneumatic actuator. and Automation, Washington, DC, vol. 3, pp. 2824-2829,
2002.
[4] I.M. Horowitz, Quantitative Feedback Design Theory - QFT,
450
(a) actual vol. 1, Boulder, CO: QFT Publications, 1993.
400 desired [5] N. Sepehri, A.A. Khayyat, and B. Heinrichs, “Development
350 of a Nonlinear PI Controller for Accurate Positioning of an
position (mm)

300 Industrial Hydraulic Manipulator,” Mechatronics, vol. 7, no.

250 8, pp. 683-700, 1997.
200 [6] S. Liu and J.E. Bobrow, “An Analysis of a Pneumatic
150 Servo System and its Application to a Computer-Controlled
100 Robot,” ASME Journal of Dynamic Systems, Measurement
0 4 8 12 16 20 24
1.0 and Control, vol. 110, no. 3, pp. 228-235, 1988.
(b) [7] E. Richer and Y. Hurmuzlu, “A High Performance Pneumatic
0.5 Force Actuator System: Part I - Nonlinear Mathematical
control signal (V)

Model,” ASME Journal of Dynamic Systems, Measurement

0.0 and Control, vol. 122, no. 3, pp. 416-425, 2000.
[8] F.E. Sanville, “A New Method of Specifying the Flow
-0.5 Capacity of Pneumatic Fluid Power Valves,” BHRA 2nd
International Fluid Power Symposium, Guildford, England,
-1.0
0 4 8 12 16 20 24 pp. D3-37-D3-47, 1971.
time (sec) [9] C.H. Houpis and S.J. Rasmussen, Quantitative Feedback
Theory Fundamentals and Applications, New York: Marcel
Fig. 9. Experimental reference tracking performance of designed nonlin- Dekker, 1999.
ear PI controller: (a) position; (b) control signal. [10] J.C. Clegg, “A Nonlinear Integrator for Servo-Mechanisms,”
Transactions of the American Institute of Electircal Engineers
Part II - Applications and Industry, vol. 77, pp. 41-42, 1958.
5083