You are on page 1of 11

Journal (~f Man~ facturing Systems , ~

Vol. 2 4 / N o . 4
20O5

TECHNICAL NOTE
Illll

A Novel, Low-Cost Pneumatic Positioning System


M. B r i a n T h o m a s , Gary P. Maul, and Enrico Jayawiyanto, M a n u f a c t u r i n g A u t o m a t i o n Laboratory,
T h e O h i o State University, C o l u m b u s , Ohio, U S A

capability is required, systems using DC servomotors or


hydraulic cylinders are usually selected over pneumatics,
as these technologies may employ linear control strategies. With a nonlinear controller, though, pneumatic actuators may replace their more expensive counterparts in
some positioning applications.
The first significant analysis of the dynamics of a pneumatic cylinder was performed by Shearer (1956). Using
principles of thermodynamics, he developed a linear
model valid for small motions about the midstroke position of a symmetric pneumatic cylinder. While this linear
model has limited utility, Shearer's methodology toward
developing the model has been used in most subsequent
works. Liu and Bobrow (1988) expand the linear model
to apply to any initial position for the cylinder. Their
linear model is used in conjunction with a PD (proportional-derivative) controller to situate the poles of the
closed-loop system. Shih and Tseng (1995) use a different approach, using system identification techniques to
develop an empirical linear model for a specific pneumatic positioning system. The authors then employ classical methods in designing a PID controller.
With the widespread availability of programmable computers, simulation and control of highly nonlinear systems is now practical. A thorough nonlinear model is
developed by Richer and Hunnuzlu (2000a), taking into
account the effects of flow through a proportional valve,
leakage, and propagation losses and delays in the air lines.
A subsequent paper from the same authors applies the
nonlinear model to a sliding mode control scheme for
force control (Richer and Hurmuzlu 2000b). Another
analysis of the nonlinear model is conducted by Kawakami
et al. (1988). In this work, the authors find little difference between models assuming adiabatic thermodynamic
processes and those assuming isothermal processes. The
nonlinear models compare favorably to the experimental
data; however, the linear dynamic model produces significant errors in simulation. Pu and Weston (1990) conduct an analysis of the steady-state speed of pneumatic
cylinders and rotary vane motors, considering the nonlinear flow of air through the valve and plumbing.

Abstract
Applications requiring accurate positioncontrol are in increasing
use in industry. Pneumatic servo systems can provide a clean,
accurate, robust positioning system; however, the proportional
valves used in such systems are relatively expensive. The use of
solenoid valves to replace a proportional flow control valve can
significantlylower the price of a positioningsystem.This substitution
is possible if the solenoid valves are operated using a pulse-widthmodulated (PWM) control scheme. This work presents a design in
which position control is realized on a single-rod, double-acting cylinder.Two solenoid valves operate conventionally to fill either chamber of the cylinder, while a third valve uses pulse-width modulation in
metering the exhaust flow.The experimental apparatus is capable of
positioninga load along a horizontal axis to within _+0.10mm.

Keywords:Pneumatic Cylinder, Solenoid Valve, Position Control

Introduction
Position control applications have typically used one
of two actuator technologies. Hydraulic actuators have
speed and force profiles compatible with many industrial processes but can present a number of workplace
hazards to personnel. Electromagnetic actuators, on the
other hand, are clean and reliable in their operation but
often require a mechanical transmission, both to convert high speed and low torque to a more useful combination and to convert rotary motion to linear motion.
While linear motors can overcome the need for a transmission, they can be expensive. Pneumatic actuators afford the opportunity to design a positioning system that
may be directly coupled with a load like a hydraulic
actuator; is clean and reliable like electric motors; and
is inexpensive. The challenge to the use of pneumatics
is the highly nonlinear dynamics that makes conventional control strategies such as PID (proportional-integral-derivative) ineffective.
This nonlinear behavior has relegated the use of pneumatic cylinders in automated equipment to applications
in which positioning accuracy is only required at the end
of the actuator's stroke. In such applications, the nonlinear dynamics of the pneumatic cylinder are not important, as positioning accuracy is obtained by moving to
and against a hard stop. When an arbitrary positioning

377

Journal ~f Mam{facturing Systems


Vol. 24/No. 4
2005

technical note
Pneumatic servo motion applications, such as that in
Liu and Bobrow (1988), typically use five-port proportional flow control valves. Wang, Pu, and Moore (1999)
use a five-port proportional valve to control a single-rod
cylinder. The control scheme uses PID control with velocity and acceleration feedforward gains. To overcome
the effects of stiction, a switching scheme saturates the
valve when motion is initiated. Drakunov et al. (1997)
apply techniques of feedback linearization toward a sliding mode controller for a rodless pneumatic cylinder. An
exception to this use of proportional valves in servo applications is the work by Kobayashi, Cotsaftis, and
Takamore (1995). Here, the authors use a single-acting
cylinder with a spring return. Two PWM-driven solenoid
valves control the fill and exhaust flows.
While servo applications typically use proportional flow
control valves, pneumatic positioning applications more
often use a small number of solenoid valves in conjunction with a PWM control scheme. Lai, Menq, and Singh
(1990) use a pair of two-position, three-way valves to
meter flow"to each chamber a pneumatic cylinder, as does
Shih and Ma (1998). A paper by van Varseveld and Bone
(1997) uses a similar physical arrangement, but applies
PWM to only one valve at a time. Chillari, Guccione,
and Muscato (2001) use a four-valve arrangement in comparing control strategies for pneumatic actuators.
Noritsugu and Takaiwa (1995) use two solenoid valves
with a pulse-code modulated (PCM) valve to meter the
exhaust. A proportional valve is used by Fok and Ong
(1999) in their study of the repeatability of a pneumatic
positioning axis.
Positioning accuracy of pneumatic positioning cylinders is typically less than 0.25 mm. Using PID control,
van Varseveld and Bone (1997) find a worst-case positioning error of 0.21 ram. Fuzzy control provides a lowload positioning accuracy of _+0.075 mm under no-load
conditions, and _+0.1 man when an external load is applied (Shih and Ma 1998). Position and pressure obsel~vers are used in the control system of Noritsugu and
Takaiwa (1995), resulting in positioning accuracies of
_+0.1 mm and _+0.2 mm under no-load and load conditions, respectively. The study of Fok and Ong (1999),
using a proportional valve, resulted in repeatabilities of
0.1 nun to 0.3 ram, depending on the starting and stopping position. Root mean square (RMS) trajectory tracking errors are calculated in Chillari, Guccione, and Muscato
(2001), which compares the ability of six control strategies to track representative command trajectories. The
RMS tracking errors fell between 0.2 mm and 2.8 mm,
depending on the strategy and trajectory. The authors

found that their fuzzy controller, with pressure feedback,


generally provided the smallest tracking error of the six
control schemes.

Apparatus and Control Philosophy


The purpose of this study is to determine the accuracy
of a pneumatic positioning system, using low-cost components, in reaching a fixed command position. A target
accuracy of _+0.10mm is established. The experimental
apparatus is shown schematically in Figure 1. The air
cylinder, mounted horizontally, has a 27.0 mm (1-1/16
in.) bore and a 101.6 mm (4 in.) stroke and is mounted
horizontally. A linear potentiometer provides position
feedback signal to the controller. An inertial load of 1.64
kg may be attached to the cylinder rod. With the hardware configuration of Figure 1, the system is realistically
limited to moving loads of 10 kg (22 lb) or smaller at
speeds of 75 mm/s (3 in./s) or less. This configuration is
scalable for larger loads or faster positioning.
Two regulators allow for independent pressure control to both chambers of the cylinder. The pressures are
set as to balance the pressure forces on the piston. Solenoid-actuated fill valves control the extension and retraction of the cylinder. These valves operate in a
conventional manner, based on the desired direction of
travel. Positioning control is obtained by metering the
exhaust through a third, PWM-driven solenoid valve.
Table l describes the motion control strategy for the
experimental apparatus.
In practice, it is preferable that the pulse-width modulator run at a moderate frequency, between 10 and 100
Hz. When the PWM control signal operates faster than
the mechanical response time of the exhaust valve, the
distinct on-off periods of the command signal are attenuated in the mechanical action of the valve, resulting in an
"average" flow rate based on the duty cycle of the command signal. For the valve used in this study, it was observed that the mechanical position of the valve did not
allow air flow when the duty cycle fell below 25% at 50
Hz. This effect is attributed to the mechanical design of
the valve. To ensure air flow through the valve, a mininmm duty cycle is enforced in the command signal.
The position signal from the linear potentiometer is sent
to an analog-to-digital signal conditioner. PD control is
realized using LabView software, which converts the continuous command signal to a PWM signal operating at 50
Hz. While the dynamics of the pneumatic cylinder are nonlinear and vary with the cylinder's position, a lxajectoryfollowing capability is not required in this application. PD

378

Journal of Manufacturing Systems


Wol. 24/No. 4
2005

technical note
y

Table 2

SUPPLY

Control P a r a m e t e r s

Parameter
PWM drive frequency
Controller update rate
Proportional gain
Differential gain
Toleranceband
Minimum duty cycle

GU~]'OeS

JST
!

Symbol
-

k~,
k.
-

DCMLv

%Salue
50 Hz
12 Hz
12.5% per mm
0.8% per mm.sec-~
__0.1mm
Varies by pressure
and load

FILL
VALVE~

k duty cycle

C~'L~NDER

100% -

a~dLOAD
LINEAR
POTENTIOMETER

error

Figure 1

Experimental A p p a r a t u s

Table 1

~ - - tolerance bund

Control Strategy

Desired
Motion
Extend cylinder
Retract cylinder
Stop cylinder

Extend
Fill Valve
On
Off
On

Retract
Fill Valve
Off
On
On

Figure 2

EMaaust
Valve
PWM
PWM
OFF

Exhaust Valve Duty Cycle as a Function of the Error Signal

control is judged to be suitable for this position-control


application. Table 2 lists the parameters used in the PD
controller. Only one parameter, the minimum exhaust
valve's duty cycle, DCMm, is adjusted in the LabView program to account for changes in pressure and load.
Figure 2 shows the exhaust valve duty cycle as a function of the error signal. The enforcement of a minimum
duty cycle has the effect of increasing the effective gain
for small displacements outside the positioning tolerance
band. This allows the system to more quickly overcome
static friction for short moves.
The dynamics of a pneumatic cylinder may be described
by a fourth-order set of nonlinear differential equations.
The governing equations were first derived by Shearer
(1956); later authors have applied his methodology in
deriving variations of these equations. Assuming an adiabatic process, the differential equations describing the
dynamics of a pneumatic cylinder can be expressed as

kRT rh, kPl 2


= Alx
-7

kRT
+ kP2 Jc
P2 =a2-~-x)rh2 ( L - x )

(2)

~ ( M g + P j A _PzA2_ParMARoD_b2_Fmc)

(3)

where x, 2 , P], and P2 are the system state variables-position, velocity, pressure in the blind end, and pressure
in the rod end, respectively.
This formulation uses state variables that are easy to
measure experimentally, but the nonlinear terms--particularly the i/x and 1/(L - x) terms--prohibit an analytical
solution. In addition, the gas mass flow inputs to the system, rh 1 and rh z , are nonlinear functions of the command
signal and the pressttre in each side of the cylinder, and
could be considered as additional states of the system.
Recognizing the mass flow inputs, rn I and rh 2 , are not
strictly independent of each other in this system, a reducedorder model of the system may be developed. Moving the
pneumatic cylinder of Figure 1 from one position to another requires the removal of gas from one chamber of the
cylinder and addition of gas to the other. "Hae ideal gas law

(1)

379

Journalo["Mant(facturingSystems
Vol. 24/No. 4
2005

technical note
shows that for an ideal, frictionless cylinder, the mass of
gas removed from one side of the cylinder is equal to the
gas added to the second, regardless of the design of the
cylinder. In effect, the function of the valves is to change
the equilibrium position of the system by means of the
flow of gas. This may be expressed mathematically as

i][
0:
0

2co = kVALvEU

(7)

(4)

LkvALwkeJ

In this expression, u is the command signal sent to


PWM-controlled exhaust valve. The sign of the control
signal determines which fill valves in F~'gure 1 are open
or closed (Table 1). The control gain, kVALVE,in Eq. (4)
is not strictly constant, but can be considered the mathematical representation of the pressure dynamics in Eqs.
(1) and (2). Considering it a bounded constant enables
treatment of the system dynamics using linear analysis
tools, with the understanding that some detail of the systems dynamics has been lost.
Similarly, Eq. (3) may be expressed in terms of the
cylinder's equilibrium position, XEQ,instead of pressure.
The mass of gas in the chambers determines the equilibrium position of the cylinder. Displacement from equilibrium changes the pressures in the cylinder, creating a
restorative force imbalance on the piston. In the reducedorder model, the restorative force may be treated as an
equivalent spring. Equation (3) is now
1

)~='~(Mg-fCcrL (X-XeQ)-b.t-Fve,c )

m2

L o I

This expression has the appearance of a linear, thirdorder differential equation with a command input, Xcog,
and disturbance term, FFR1OThe terms kcrL and kVALV
f
are time-variant and nonlinear functions of the state variables, though, and prevent a purely linear analysis of the
system. Having bounds of the nonlinear terms, though,
allows for some understanding of the system dynanfics.
The characteristic equation, X, of the closed-loop "linear" system may be expressed as

(~l(lq-kvaLvEkD)S-l-(kcYLk-K~vEkp)

(8)

For the pneumatic cylinder in this experiment, Table 3


lists the nominal values of the system variables for the
cylinder at the mid-stroke position, having a mass load
and a pressure of 414 kPa in the blind end. Under these
conditions, the roots of the closed-loop system are s = 5.88 and s = -3.96 + 95.11i. The real root has a time
constant of 0.17 s, and the lightly damped complex roots
have a natural frequency of 15.1 Hz.
Moving the cylinder away from its mid-stroke position increases the effective stiffness, /~crL, which in turn
moves the complex roots of Eq. (8) away from the real
axis without significantly changing the real root. Uncertainty in the control gain, kVALVE,m o v e s all three roots,
with an increase in kVALV
E moving the real root away
from, and the complex roots toward, the imaginary axis.
Friction, though, is the strongest nonlinear term in Eq.
(7), having two significant effects. First, it is expected to
diminish high-frequency oscillation associated with the
complex roots of Eq. (8), allowing the single real root to
dominate the system response. Second, friction will cause
the c~ hnder s position to lag behind the equilibrium position. Both phenomena are observed in comparing Figure

(5)

where kCyL is the cylinder's spring stiffness coefficient.


Like the control gain, kvacw, this coefficient is not constant but varies as a function of both stroke and displacement from equilibrium. Assuming displacement from
equilibrium is sufficiently small and ignoring the volume
of gas in the supply lines leading to the cylinder, the stiffness coefficient is expressed as

lCcyL= RT I ~

, 71ijkxEo/1

[--[,VALVEk,--kvALVEAD

(6)

This equation is derived in Thomas (2003).


Equations (4) and (5) describe the open-loop dynamics of the reduced-order system model. Assuming the experimental apparatus is operating in the horizontal plane,
the closed-loop system dynamics with a PD controller
may be expressed in matrix form as

380

Journal of ManuJ~wturing Systems


Vol. 24/No. 4
2005

technical note
IIIII

Table 3

0.06

Nominal Values of System Parameters

Method of
Value
Determination
1.64 kg
Measurement
0.02 N-sec/mm
Empirical

Variable
Meaning
M
Mass load
b
Viscousdmaaping
coefficient
]~crL Equivalent spring
9.32 N/ram
stift'ness
kVALVE Control valve
0.75 mm/%-sec
gain
k
Proportional
12.5 %/mm
feedback gain
kD
Derivative
0.8 %-sec/mm
feedback gain
F~.~c Coulomb friction
45 N

i
i

I
i

i
I

O.[E; -

0.04"

0.03 -

- - m

. . . . .

Eq. (8)

1--

. . . . . . .

- -

i
i
i

~
i
i
I

- - - [

P
E

i
i

i
i

- -

. . . . . . . .
i
f
i

0.02-

Empirical
0.01 -

Control
parameter
Control
parameter
Empirical

0.00
0.5

0.0

1.0

time
l--

1.5

2.0

(s)

cylinder position - -

equilibrium I

Figure 3
Linear Simulation without Friction, Extension to 50 mm

4 with Figure 3. Figure 3 shows the linear response of the


system, without friction, for extension to 50ram using
the values listed in Table 3. Figure. 4 includes the effect of
Coulomb friction in the response, and clearly exhibits
both the lack of oscillation and the lag between the equilibrium position and the cylinder's position.

0,06

i
I

i
I

i
I

0.05

0.04

Experimental Conditions
The PD control algorithm was tested using different
configurations of the experimental apparatus to evaluate the positioning accuracy. Three nominal gauge pressure settings were applied to the blind end of the cylinder:
207 kPa (30 psi), 310 kPa (45 psi), and 414 kPa (60
psi). The pressure in the rod end of the cylinder was set
proportionately higher (227 kPa, 339 kPa, and 453 kPa,
respectively) to account for the difference in piston areas to achieve balance in the pressure forces. Tests were
conducted with no external load on the cylinder, and
with a mass load rigidly attached to the cylinder. When
the load is attached to the cylinder, an unknown frictional force is generated, acting between the mass and
its supporting surface. The mass load is supported in a
manner that does not generate side loads on the cylinder
rod. With LabView, the dynamic responses of the system were recorded for position step commands to the 10
ram, 50 mm, and 90 lnm positions, in both extension
and retraction. Positioning error was calculated by averaging the last 12 data points in a record, ending approximately 8 seconds after the step command. The
positioning en'or in a single record was typically either
always positive or always negative.

0.03

0,02

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

0.01

0.00

0.0

03

1.0

1.5

2.0

time (s)
[ ~ c y l i n d e r position - -

equilibrium

Figure 4
Linear Simulation ~ t h

Friction, Extension to 50 m m

Results
No-Load

Conditions

Figures 5, 6, and 7 show the transient response of the


positioning system in extension. Table 4 shows the steadystate positioning errors, calculated using the last 12 data
points in each data record. Figure 8 shows the transient
response for positioning from 100 mna to 10 ram, with
Table 5 showing the positioning errors in retraction. The
maximum positioning error in Tables 4 and 5 is 0.19 ram,
though most positioning errors fall within the _+0.10 mm
en'or tolerance band.

381

Journal o[ Manufacturing Systems


Vol. 24/No. 4
20O5

technical note

12

-I-

/ - -

.I

T
I

10

I~_

4r ["

. . . .

I. . . .

--

--

--

"
0.0

'_,__ :__
o

. . . .

0.5

1.0

1.5

90

i _ /~ - ~ _ ~~_

80

. . . .

I-

50

. . . .

T -~--

40

. . . .

..~t#

:
2.0

3.0

3,5

~-

I
-

-- -- -- 4 . . . .

]
I

- - ~ - - - - ~

- - ~

- - - T - - - q

.
0.0

4.0

. . . .

0.5

I. . . .

I. . . .

I. . . .

I-- . . . .

I. . . .

I. . . .

I. . . .

I. . . . .

I
-I . . . .

. . . .

,
I. . . .

I
I. . . .

,
E. . . . .

I. . . .

J. . . .

I. . . .

~- . . . .

I
I.

. . . .

I. . . .

I. . . .

[. . . .

. . .

. . . .

]. . . . .

.
2.0

2.5

3.0

3.5

4.0

(see)

I 2 0 7 k P a - - 3 1 0 k P a

--

-414kPa

Figure 7
No-Load Extension to 90 mm

Figure 5
No-Load Extension to 10 mm

~01

I.

I. . . .

1.5

Command

-414kPa

I. . . .

I. . . .

time

....

...... ~ ' - - I

(see)

- 2 0 7 k P a ~ 3 1 0 k P a

. . . .

1.0

~
"

_ _1

~1

10-

.
I ------q

~ -~-

----

2.5

_
I

' -?-:::i-

. . . .

i
i

I~

60

30

-- -- -- /

:___',___

time

Command

....

100

110
I

100

- - 7 - - - 5 -

- q -

~. . . .

~. . . .

t . . . .

L. . . .

F---F---

90

40 [

jl~[-i

!3 - I-i
"~

20

"~- -- ~

-E

_ _ ~ _

70

. . . .

_ ~ - - - J ~

_I . . . .

t. . . .

7 -

_ _ ~ _ _ _ L _ _ _
I
I

[-

60
.

. . . . . .

T .
I

I~

80
. . . . . . . .

50

- - - Y
__

40

--I

I
3-

J_

30

I
q

20

----

0.4~.d

I-

C - - - F - -

L
I

. . . .

I. . . .

I. . . .

--

i
I--

I
I

F------F-----L
[
1-7

I
I

2.0

2.5

3.0

3.5

~I
--

Ilgll

10

I
J

------T------q--

J
I. . . .
_~_

0.5

1.0

1.5

0.0

1.0

2.0

3.0

time

Conl~nd

4.0

5.0

6.0

0.0

(see)

time

, 2 0 7 k P a ~ 3 1 0 k P a

-414kPa

Command

10mm
50mm
90mm

Error

i414kPa

Table 5

Table 4
Positioning

- 2 0 7 k F a ~ 3 1 0 1 d z a

Figure 8
No-Load Retraction to 10 mm

Figure 6
No-Load Extension to 50 min

No-Load

4.0

(see)

No-Load

on Extension

Positioning

Error

on Retraction

103 kPa

207 kPa

310 kPa

103 kPa

207 kPa

3111 kPa

-0.01mm

0.04mm

0.10mm

10 m m

-0.05 m m

0.00 m m

0.15 mm

-0.09mm
0.01mm

0.03mm
0.11mm

-0.12mm
-0.19mm

50 mm
90 mm

-0.04 nma
-0.07 mm

0.06 m m

-0.03 m m

0.03 mm

0.05 mm

Inertial Load

Gravitational (Constant) Load

An external m a s s adds a 1.64 kg external load with an


u n k n o w n fiicfion c o m p o n e n t to the actuator. Figures 9,
10, and 11 s h o w the s y s t e m response to step c o m m a n d
inputs in extension. Table 6 g i v e s the positioning errors
for the inertial load in both e x t e n s i o n and retraction. T h e
range o f positioning errors in Table 6 is comparable to
those in Tables 4 a n d 5.

The experimental apparatus w a s reoriented for vertical motion. The m a s s load, suspended b e l o w the pneumatic cylinder, applies a constant force to the cylinder.
The pressure in the blind end o f the cylinder w a s set at
2 0 7 kPa. The pressure in the rod end w a s increased from
3 4 0 kPa to 370 k_Pa to account for the added w e i g h t o f
the payload.

382

Journal

o.I"

Systems

Manufacturing
Vol.

24/No.

2005

technical note
4 <
2

i
I

. . . .

i
I

in-'~-

_1 . . . .

,
P

.~
~

g
6

.
I
~1
I

i
I

'

+.

. . .

L
I

i
I

i
I

i
I

,. . . .
i

T - - - 5
I

I
I . . . .
I
I

I
I---]
]

I
J

i
4-I
I

-i

L _

I
I

i
I

----f

80

. . . .

70

60

,. . . .

T- - - T

. . . . .

-I

I. . . .

I-

. . . .

I. . . .

I-

. . . .

I. . . .

r-

- + - -

-I
I

40 T - [

a -

r
I

I
I

I
I

. . . .

-I

3l

i
i

. _ - J _ _ I

-I

10

--

J.

"1- -

I
- -

I
-

r - - - T .....
I

. . . .

~. . . .

F - - -

. . . .

I. . . .

J.

I. . . .

2.0

2.5

3.0

3.5

I
-I

--

; - 5 - - -

I
--

i. . . .

-I . . . .
r=

20

04

. . . .

1.0

1.5

--

--

--

. . . . .
----

_-I_

I
--

--

--

--

-----4

--

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

0.0

0.5

time (sec)
Command i

"207kPa

Load

,310kPa

Extension

4.0

time (sec)
~

-414kPa

.....

Corm~nd

-207kPa

,.310kPa~-

.414kPa

Figure 11

bYgure 9
Mass

Mass

to 10 mm

Load

Extension

to 90 mm

1~ble 6

60

Mass

50
I

40

-I . . . .

(!

3o

r - -

I . . . .

I. . . .

I
I

I
I

I . . . .

I. . . .

. . . .

I
I
-I . . . .

-I . . . .

10

-, . . . .

PI

---J

_ _ _ a . _ _ _ J

------F------J
I

I
I

90 I

--4---

100

. . . . .I . . . . . I. .

,-

I
I

i
I
I

r---

r'--

--

. . . .

-T------~

I. . . .

t--I

2.5

3.0

3.5

"

0.0

0.5

1.0

1.5

2.0

207 kPa ~

Load

Extension

103 kPa
10 mm, extending 0.06 mm
50 ram, extending -0.10 mm
911ntm, extending -0.02 mm

207 kPa
-0.14 mm
-0.06 mm
-0.10 mm

310 kPa
-0.18 mm
-0.02 mm
-0.07 mm

10 mm, retracting
50 rnm, retracting
90 nun, retracting

-0.17 mm
0.03 wan
0.00 mm

0.05 mm
0.05 mm
-0.08 mm

310 kPa --

-0.09 mm
0.09 mm
0.02 mm

vertically-loaded system moving to 50 mm, from both


the fully extended and fully retracted positions. The vertical scale in Figure 12 is reversed so that the ordinate
matches the physical arrangement of the apparatus. Table
7 shows the positioning errors for the vertical load arrangement. Except for one datum, the positioning errors
in the vertical orientation are similar to those found in
the horizontal.
Increasing the sampling period of the controller, so
that the cylinder could not cross the tolerance band between samples, would allow the controller to stop the
cylinder while in the control band. This approach was
not pursued due to time constraints.

- 414 kPa

Figure 10
Mass

Errors

4.0

time ( s e e }
C ,o m m a n d

Positioning

. . . .

-I . . . .

Load

to 50 mm

It was observed that the system response to a step input would oscillate under the command position for some
period of time before motion stopped. Closer inspection
of the data revealed that the cylinder moves sufficiently
fast as to cross the tolerance band around the conm]and
position in less than one sampling period. During upward
motion (reta'action), the weight of the external payload
adds to the restorative force in the actuator. The combined acceleration from the air pressure and gravity forces
the payload down faster than when only an inertial load
is present. The system will oscillate indefinitely until a
sample happens to fall within the tolerance band.
To prevent this undesirable behavior, the minimum
duty cycle function of Figure 2 was made asymmetric. A
lower minimmn duty cycle for downward motion compensates for the additional gravitational load acting on
the cylinder. Figure 12 shows the step response of the

Discussion
The second section of this paper discussed the control
philosophy for the pneumatic positioning system. A minimum duty cycle (DCMIN)in the exhaust valve is enforced,
as the average mechanical position valve spool does not
allow flow at duty cycles of less than 25%. Without a
minimum control effort, the system would not be able to
compensate for small positioning errors. The width of

383

Journal c~f Manufacturing Systems


Vol. 24/No. 4
2005

technical note
Table 8
Minimum Duty Cycles

time (see)
0.0

0.5

0~
20~
40

|
!

.
. . . . . .

~ . . . . .

. . . . . . . . .

~ - - j

1,0

1.5

i
T . . . . . .

r
r . . . . . . . . . . . .

. . . . . .

2.5

2.0

Cylinder Blind
End Pressure
207 kPa
310 kPa
414 kPa
207 kPa
310 kPa
414 kPa
310 kPa

i
r- . . . . . . . . . . . .

60

1 t ......... : S
80

. . . . . . . . . .

100

I
- -I . . . . . .
I

120

Command

. . . . . . . .
,

'

T . . . . . .
r
I
- . . . . . .
I

~- . . . . . .
!
I
F . . . . . .
r
I

o Extending

p. . . . . .
I
I
I. . . . . .
I
I

Retracting

DCw~,

Payload
None
None
None
1.64 kg
1.64 kg
1.64 kg
1.64 kg
(vertical)

50%
37.5%
30%
50%
41%
32t,
36% (up)
25% (down)

Figure 12
"Vertical Mass Load Motion to

50

50,2

mm

i
i
J

50.1

Table 7
Vertical Load Positioning Errors

10 mm
5(1 mm
90 nun

Extending (down)
-0.10 mm
-0.36 mm
0.10 mm

.-,=, ~-

EE
50,0

; __

Retracting(up)
-0.15 mm
-0.00 mm
0.03 mm

'~

i
i
i

i
i
i

D- =-,-

i
i
i

i
i
i

i
F
t

,~=- - - L - - - -

t
,
i

I
,
I

I
J
I

i
,
~I

4.5

5.0

5.5

i
i
J

~ , i

7.0

7.5

,
1I

49.9

o.
49.8

49.7
4.0

this dead band, without DCMIN,may be estimated. By dividing the 25% minimum duty cycle for air flow by the
12.5% per mm proportional error gain, the width of the
dead band is +2 m m about the command position.
Through the course of this work, the minimum duty
cycle, DCMm, is the controller parameter from Table 2
that is adjusted for variations in pressure and external
loading. This suggests that, once the proportional and
derivative gains are determined for a particular hardware
configuration, adjustment of DC~,N alone is sufficient to
compensate for variations in the supply pressures, loads,
or the plant. For a vertical load, two values o f DCM1Nare
required. Proper adjustment o f DC,~IN is necessary for
good performance of the controller. If this parameter is
too small, the response time and accuracy of the controller will suffer. If it is too large, the system can limit cycle
about the desired position. Table 8 gives the minimum
duty cycles used in this paper.
In each experiment, the pressure in both chambers of
the cylinder was adjusted to establish force equilibrium
over both faces of the cylinder piston. In practice, a perfect force balance is not obtainable. Measurement errors
exist in pressure gauges and sensors, and the internal
mechanisms of regulators suffer from drift and resolution limitations. As a result, small force imbalances will
induce drift in the cylinder away from the command po-

6.0

time

6.5

8.0

(see)

310 kPa]

Figure 13
Position ltoiding Detail

sition. Evidence for this was observed in the experimental data, as most experiments had a positioning error that
was either always positive or always negative. The force
imbalance is responsible for many of the results outside
the _+0.10 mm tolerance band in Tables 4, 5, and 6. The
minimum duty cycle acts to increase the effective feedback gain for small displacements. With this, the controller is able to return the cylinder to within the tolerance
band in one sample period.
Figure 13 shows the positioning detail for extension to
50 m m with 414 kPa applied to the blind end of the cylinder and an external mass load. In this test, the pressure
force imbalance induces a drift toward the lower tolerance
bound (retraction). From Figure 13, it is apparent that increased resolution in the analog-to-digital conversion of
the sensor and a faster sampling rate would allow a smaller
positioning tolerance to be applied to the system.
The stiffness of the pneumatic cylinder may be calculated using the ideal gas law. This analysis, supported by

384

Journal o f Manufacturing Systems


Vol. 2 4 / N o . 4
2005

technical note
previous works (Fok and Ong 1999), assumes displacements are small enough to allow ibr an isothermal analysis. Neglecting the volume of air in the external tubing,
the cylinder stiffness is expressed as

60.,

,,

,,

1.0

1.5

40

g
g 30
o. 20

Note that Eq. (6) could be manipulated, using the ideal


gas law equations, to find Eq. (9). Both equations are
singular at the limits of travel, when x = 0 or x = L. In the
real system, the compliance of the air in the air lines
connected to the cylinder will prevent an infinite stiffness. The minimum stiffness, however, is of more interest to the design engineer. Differentiation shows Eq. (9)
has a minimum value at the mid-stroke position, when x
= L/2. Here, the stiffness is

kcrL MIN -- 4PjAI


'

l0
0
0.0

0.5

2.0

Time (see)
-o--Data ~Simulation

Figure 14
M a s s Load E x t e n s i o n to 50 m m

actual position, as seen in Figure 4. When enough lead


exists, the restorative force generated by the difference
between the equilibrium and actual position is enough to
break the stiction that might otherwise prevent the system
from moving away from its command position. A contributing factor to this is resonance in the system. At 90 nun
and 414 kPa with no load, the resonant frequency of the
cylinder is estimated to be 54 Hz, very close to the 50 Hz
PWM frequency of the exahaustvalve. This same frequency
match occurs around 10 nun for the same pressure and
loading conditions, but does not appear elsewhere in the
experimental conditions. Figure 5 shows a similar anomaly
for the movement to 10 iron, but the scale of Figure 5
exaggerates the magnitude of this dynanfic.
A statistical analysis of the data in Tables 4 through 7
was performed to determine if any correlation between
the positioning error and the control parameters exists. In
testing for differences in the mean, no significant relationships were found to exist in the positioning error with
respect to the cylinder pressures, loading conditions, direction of travel, or command position. Further analysis
examined a set of 25 repeated runs under the same nominal operating conditions. The repeated positioning error
behaves like a random variable with a normal distribution. A Fourier analysis of the error as a function of time
shows no significant frequency information.
The cost of the hardware used in the experimental apparatus (Figure 1), discounting the controller, is approximately
$530. For comparison, a similar pneumatic system using a
proportional air valve wotfld cost about $760, due to the
higher cost of the valve. Replicating the experimental apparatus with a hydraulic system would cost $1,300, not
including the reservoir and pump unit. A linear motor sys-

(10)

For the system used in this paper, the stiffness at midstroke is 4.66 N/mm (26.6 lb/in.) for the system with a
blind-end pressure of 207 kPa (30 psi). At 310 kPa (45
psi), the stiffness is 6.98 N/ram (39.9 lb/in.), and at 414
kPa (60 psi), the stiffness is 9.32 N/mm (53.2 lb/in.).
Qualitatively, the experimental data agree with the estimated pole locations in the linearized analysis of Eq.
(8). The time rise is consistent with a first-order linear
system with a time constant of 0.17 s, and the oscillatory
behavior generated by the complex roots is suppressed by
the Coulomb friction in the cylinder. Figure 14 shows the
experimental data from Figure 10 overlaid with the matching simulation data from Eq. (7) and Figure 4. As with
previous works such as Kawakami et al. (1988), discrepancies exist between the simulated and actual dynamic
response in Figure 14. Many of these discrepancies can
be attributed to the simplifying assumptions made for the
simulation. The simulation model treats many of the
nonlinear elements in Eq_(7) as linear functions, such
as the cylinder stiffness, kcrL, and the flow valve coefficient, kj. The model also ignores the effects of stiction.
Stiction accounts for the delay in the experimental response seen in Figure 14, and plays a significant role in
the final positioning due to stick-slip motion over small
motion increments.
Figure 7 shows an apparent anomaly in its 414 kPa noload response. Immediately after the cylinder passes its
command position, the cylinder jumps 5 nun away from
its command position. This behavior is attributed to the
fact that the equilibrium position, xEQ,leads the cylinder's

385

Journal o f Mam(facturing Systems


Vol. 24/No. 4
2005

technical note

Illl

L
M

tern, with integated position feedback, power supply, and


servo controller, costs more than $2,200. An open-loop
stepper motor-timing belt system can be realized for less
than $530, but such a system would be susceptible to positioning errors from missed steps. System costs were obtained with reference to McMaster-Carr (2004).
While the system has been successfully demonstrated
in both the vertical and horizontal configurations, it is
expected that a similar pneumatic positioning system would
find more practical application in the horizontal configuration. In the vertical configuration, the rod end and blind
end pressures are set for a specific loading condition. Any
changes in the external load would upset the force balance created by these specific pressures. Therefore, the
practical application of this positioning system is restricted
to those applications in which external loads act perpendicularly to the cylinder's axis of motion. An example of
such an application is a gantry-style robot, in which the
positioning cylinders move the payload in the horizontal
plmae. A pick-and-place end effector, realized with conventional pneumatic cylinders, can act in the vertical axis
to place or retrieve parts fi'om multiple positions in the
working space of the robot.

P
R
s

T
x

Subscripts
1
2
ATM
CYL

EQ
FRIC
ROD
E4LVE

Chillari, S.; Guccione, S.; and Muscato, G. (2001). "An experimental


comparison between several pneumatic control methods." Proc.
of 40th IEEE Conf. on Decision and Control, Orlando, FL, Dec.
2001, pp1168-1173.
Draknnov, S.; Hanchin, G.D.: Su, W.C.; and Ozguner, U. (1997).
"Nonlinear control of a rodless pneumatic servoactuator, or sliding modes versus Coulomb friction." Automatica (v33, n7),
pp1401-1408.
Fok, S.C. and Ong, E.K. (1999). "Position control and repeatability
of a pneumatic rodless cylinder system for continuous positioning." Robotics and Computer Integrated Mfg. (v15), pp365-371.
Kawakami, Y.; ,~kao, J.; Kawai, S.; and Machiyama, T. (1988). "Some
considerations on the dynamic characteristics of pneumatic cylinders." Journal o f Fluid Control (v19, n2), pp22-36.
Kobayashi, Shigeru; Cotsaftis, Michel; and Takamori, Toshi (1995).
"Robust control of pneumatic actuators based on dynamic impedance matching." Ptvc. of IEEE Int'l Conf. on Systems, Man, and
Cybernetics, pp983-987.
Lai, Jiing-Yih; Menq, Chia-Hsiang; and Singh, Rajendra (1990). "Accurate position control of a pneumatic actuator." "l~ans. of ASME
(v112, Dec. 1990L pp734-739.
Liu, S. and Bobrow, J.E. (1988). "An analysis of a pneumatic servo
system and its application to a computer-controlled robot." Journal of Dynamic Systems, Measurement, and Control (Sept. 1988),
pp228-235.
McMaster-Carr (2004). McMaster-Carr Catalog 110. McMaster-Carr
Supply Co.
Noritsugu, T. and Takaiwa, M. (1995). "Robust positioning control
of pneumatic servo system with pressure control loop." Proc. of
IEEE Int'l Conf. on Robotics and Automation, pp2613-2618.
Richer, Edmond and Hurmnzlu, Yildirim (2000a). "A high performance pneumatic force actuator system: part I - nonlinear mathematical model." Journal (~[ Dynamic Systems, Measurement, and
Control (Sept. 2000), pp416-425.
Richer, Edmond and Hurmuzlu, Yildirim (2000b). "A high performance pneumatic force actuator system: part 11 - nonlinear controller design." Journal of Dynamic Systems, Measurement, and
Control (Sept. 2000), pp426-434.

A low-cost position control system for a pneumatic


actuator is presented. The system can achieve positioning
accuracies of under 0.2 ram, similar to pneumatic positioning systems presented in previous works. The control
system uses a modified PD controller to determine the
duty cycle on a PWM-driven solenoid valve controlling
the exhaust flow from the cylinder. The control algorithm has been shown to be effective in both horizontal
and vertical configurations, though it is expected to be
applied in horizontal applications where there are no significant variations to the load applied to the cylinder. This
system provides a significant reduction in the cost of the
hardware over competitive technologies.

Nomenclature
Symbols

DCMl,v
F
g
k
k

blind end of cylinder


rod end of cylinder
atmospheric
cylinder
equilibrium
friction
rod
valve

References

Conclusions

A
b

cylinder stroke
payload mass
mass flow rate of gas
pressure
gas constant
LaPlace operator
temperature
position

area

coefficient of viscous friction


minimum duty cycle of exhaust valve
force
acceleration of gravity
ratio of specific heats
gain (/~ indicates a time-variant gain)

386

Journal of Manufacturing Systems


Vol. 24/No. 4
2005

technical note
Authors' Biographies

Pu, J. and Weston, R.H. (1990). "Steady state analysis of pneumatic


servo drives." Ptvc. of the h~stitution of Mechanical Engineers
(v204), pp377-387.
Shearer, J.L. (1956). "Study of pneumatic processes in the continuous control of motion with compressed air, I, II." Trans. of ASME
(Feb. 1956), pp233-249.
Shih, Ming-Chang and Ma, Ming-An (1998). "Position control of a
pneumatic cylinder using fuzzy PWM control method."
Mechatronics (v8), pp241-253.
Shih, Ming-Chaug and Tseng, Shy-I (1995). "Identification and position control of a servo pneumatic cylinder." Control Engg. Practice (v3, n9), pp1285-1290.
Thomas. M.B. (2003). "Advanced servo control of a pneumatic actuator." PhD dissertation. Columbus. OH: The Ohio State Univ.
van Varseveld, Robert B. and Bone, Gary M. (1997). "Accurate position control of a pneumatic actuator using on/off solenoid valves."
1EEE/ASME Trans. on Mechatronics (v2, n3), pp195-204.
Wang, Jihong; Pu, Junsheng: and Moore, Philip (1999). "A practical
control strategy for servo-pneumatic actuator systems." Control
Engg. Practice (n7), pp1483-1488.

Dr. M. Brian Thomas is an assistant professor of industrial engineering at Cleveland State University and is a 2003 PhD graduate of
The Ohio State University.
Dr. Gary P. Maul is an associate professor of industrial engineering at The Ohio State University. He has more than 30 years of
experience in industrial automation and material handling,
Mr. Enrico Jayawiyanto earned his master's degree in industrial
engineering from The Ohio State University in 2003. He is working
as an automated systems engineer.

387

You might also like