Mechanical Systems and Signal Processing 91 (2017) 93–110

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Dynamic contraction behaviour of pneumatic artificial muscle

Marc D. Doumit ⇑, Scott Pardoel
Department of Mechanical Engineering, University of Ottawa, Ottawa, Ontario, Canada

a r t i c l e i n f o a b s t r a c t

Article history: The development of a dynamic model for the Pneumatic Artificial Muscle (PAM) is an
Received 27 June 2016 imperative undertaking for understanding and analyzing the behaviour of the PAM as a
Received in revised form 1 January 2017 function of time. This paper proposes a Newtonian based dynamic PAM model that
Accepted 2 January 2017
includes the modeling of the muscle geometry, force, inertia, fluid dynamic, static and
dynamic friction, heat transfer and valve flow while ignoring the effect of bladder elasticity.
This modeling contribution allows the designer to predict, analyze and optimize PAM per-
Keywords:
formance prior to its development. Thus advancing successful implementations of PAM
Pneumatic artificial muscle
Soft
based powered exoskeletons and medical systems. To date, most muscle dynamic proper-
Linear and pneumatic actuator ties are determined experimentally, furthermore, no analytical models that can accurately
predict the muscle’s dynamic behaviour are found in the literature. Most developed analyt-
ical models adequately predict the muscle force in static cases but neglect the behaviour of
the system in the transient response. This could be attributed to the highly challenging task
of deriving such a dynamic model given the number of system elements that need to be
identified and the system’s highly non-linear properties. The proposed dynamic model in
this paper is successfully simulated through MATLAB programing and validated the pres-
sure, contraction distance and muscle temperature with experimental testing that is con-
ducted with in-house built prototype PAM’s.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction

The PAM is a pneumatically powered unidirectional actuator whose structure and functionality differs significantly from
traditional pneumatic actuators (i.e. cylinders). In its most common design, the PAM is made of an elastic bladder that is
enclosed in a double helically braided sleeve. Both bladder and sleeve are held airtight at their ends using mechanical fix-
tures. Fig. 1 shows an example of the PAM prototype that was previously designed and developed by Doumit et al. [1]
and used for this research.
With reference to Fig. 1, h is the braid angle of the sleeve, b is the sleeve inclination angle at muscle ends, DL is the muscle
contraction distance, and F is the muscle force. The muscle operation is simple, as the bladder is inflated, using an external
fluid flow through the end fixture, the muscle pressure increases and results in two types of stresses acting on the muscle’s
thin inner walls, namely, hoop and longitudinal stresses [2]. The hoop stresses will yield to a force that would radially
expand the muscle. Subsequently due to the sleeves structural properties this will longitudinally contract the muscle. The
longitudinal stresses will yield to a force that will longitudinally stretch (relax) the muscle. An unbalance of these two forces
(radial and longitudinal) yields to a muscle motion such as contraction or relaxation. Neglecting all other forces such as fric-

⇑ Corresponding author.
E-mail address: marc.doumit@uottawa.ca (M.D. Doumit).

http://dx.doi.org/10.1016/j.ymssp.2017.01.001
0888-3270/Ó 2017 Elsevier Ltd. All rights reserved.
94 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

Fig. 1. PAM prototype: (a) pressurized state (b) deflated state.

tion, the maximum contraction distance of the muscle coincides with the equilibrium state of both radial and longitudinal
forces. If the muscle contraction is externally resisted, the muscle produces a substantial contraction force that is called the
muscle force. However, the muscle does not produce any expansion force if the relaxation motion is resisted, and thus the
PAM is a unidirectional actuator.
The earliest reference of the PAM dates back to 1941, Johnson and Pierce patented the ‘‘Expansible Cover” which is also
referred to as a mining tube or a cartridge [3]. This was used in coal mines where the device was inserted in a rock seam and
upon inflation, its radial expansion would cause the rock to crack open. Later in 1949, De Haven revealed the longitudinal
contraction ability of the device by patenting the tensioning device for producing a linear pull [4]. This was applied to a crash
belt apparatus and was operated by a sudden release of compressed air that was triggered by a charge of gun powder. How-
ever, it was not until 1955 that the device was proposed as an actuator by Gaylord who patented the fluid actuated motor
system and stroking device [5]. The PAM is also referred to as the McKibben muscle after the physicist Joseph L. McKibben. In
1957, McKibben introduced the PAM to the field of orthotic medicine by using the PAM to control a Wrist-Driven Wrist-
Hand Orthosis [6]. Whereas, the PAM was perceived as a great innovation in this era, it was not a flawless success. This
was attributed to the PAM’s structural reliability, pneumatic power storage system, controls as well as other factors. After
the letdown of the PAM, there was a period of approximately two decades before it was reintroduced by the industry and
researchers. Among the pioneers, in the mid 1980’s, Bridgestone Rubber introduced a PAM version called Rubbertuators
for robotic applications [7]. Bridgestone offered two separate PAM models (i.e. RASC and Soft Arm robots); however, they
were unsuccessfully commercialized and became obsolete in 1990. Despite this failure, researchers and industry continued
their work on improving the PAM performance and proposed new designs of muscles. As an example, in 2002, Festo Corpo-
ration introduced the fluidic muscle MAS [8] which infuses the sleeve into the bladder. This design claimed to improve the
fatigue life of the muscle. Another new design called the Pleated Pneumatic Artificial Muscle (PPAM) [9] was also proposed to
eliminate muscle’s hysteresis and to improve muscle contraction distance and force.
Nowadays, the success of the PAM is in applications where other forms of actuations such as electric and hydraulic are not
optimum or feasible. The distinctive properties of the PAM also make it very appealing for the biomedical field. It has been
used for human motion rehabilitation devices [10,11]. This is attributed to the PAM compliant behaviour, high force to
weight ratio, high contraction ratio as well as human safety interaction behaviour.

2. Prior PAM models

A literature survey has shown that the PAM has been extensively studied and tested by researchers [12]; however, most
PAM analytical models analyze the muscle force F for static condition only. Most models developed to predict the PAM
dynamic behaviour are either based on lumped parameter approaches or are empirical in nature where the values of the
model parameters are experimentally determined.
The first PAM analytical model was developed by Gaylord [5] for the muscle force F in static condition and later by Schulte
et al. [13] as follows:

PpD290
F¼ ð3 cos2 h  1Þ ð1Þ
4

where P is the muscle pressure, and D90 is the diameter when the braid angle h is 90°. Many similar static models were later
developed by other such as Chou et Hannaford [14].
Tondu and Lopez [15] further modified the aforementioned static model by including a PAM dynamic force Fdyn for the
case of the PAM suspended vertically. This term takes into consideration the static muscle force and the friction forces. The
model’s governing equation is:

F dyn  mg ¼ m€x ð2Þ

where m is the mass of the hanging load, g is the gravitational constant and x is the muscle contraction distance. The param-
eters of this lumped model are selected and adjusted based on experimental results. Due to model simplification, Tondu and
Lopez have noted that the model given by (2) is less accurate when the muscle pressure is low (200 kPa).
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 95

Similar to (1), Petrovic first developed a static model based on the virtual work principle [16]. Assuming the PAM to be a
gas spring with a non-linear stiffness, and incorporating the system’s inertial and frictional forces, the static model is trans-
formed to a dynamic model as follows:

Mq _ þ f ðqÞp ¼ s
€ þ lðq; qÞp ð3Þ

where M is the mass of the system, q is the muscle contraction length, lðq; qÞp _ is the lumped model of the total system fric-
tion forces, f ðqÞp is the relationship for the static muscle force, and s is the dynamic muscle force. Petrovic presented no
results or analysis about the accuracy of his model.
Thongchai et al. [17] adopted a frequency modeling approach to predict the dynamic behaviour of a PAM set. Using the
Intelligent Machine Architecture (IMA) framework, they succeeded in developing three linear third order dynamic models,
each for a specific frequency range. However, the accuracy of such models could not be determined since the open loop
results of the developed models were not reported.
Kang et al. [18] also developed a model to analyze the frequency response of PAM. This model uses Gaylord’s model (1)
with a pressure dependent correction factor q.
qðpg Þ ¼ 1 þ cq1 expðcq2 pg Þ ð4Þ

where cq1 and cq2 are experimentally estimated constants. In this model the fibre to fibre friction was modeled as the com-
bination of coulomb and viscous friction forces.

F friction ¼ cv e_ þ cc sgnðe_ Þ ð5Þ

where e is the muscle strain, cv and cc are experimentally obtained coefficients for the viscous and coulomb friction, respec-
tively. Experimental testing of the model reported that predicted and experimental actuation forces have similar trends but
different magnitudes.
Reynolds et al. [19] developed a PAM dynamic model based on the Voigt Viscoelastic model. This is a lumped parameter
model of the following form:
€ þ By_ þ Ky ¼ F ce  Mg
My ð6Þ
where M is the load mass, B is the system damping coefficient, K is the spring stiffness constant, and Fce is the effective force
provided by the contractile element. The values of these parameters were identified experimentally for a step perturbation of
the load at constant pressure. Two lumped parameter models were derived to predict the PAM motion in the contraction and
in the relaxation phases. The developed models were experimentally validated for a triangular wave input of muscle pres-
sure between 55 kPa and 124 kPa. Reynolds et al. determined their model’s error accuracy to be 15% with respect to the aver-
age muscle contraction length.
Cao et al. [20] have subsequently recreated and extended the model created by Reynolds et al. They examined the
dynamic behaviour of PAM from 60 to 500 kPa. Utilizing a first order polynomial curve fit, they determined linear pressure
dependent equations for the force F, spring and damping coefficients K and B. From this analysis it was discovered that the
experimentally obtained coefficient plots (force, spring, and damping) displayed a drastic change in slope when the pressure
reached 200 kPa. To address this, two models were proposed. The first being a quadratic polynomial model:

FðPÞ þ B1 ðPÞx_ þ K 1 ðPÞx þ K 2 ðPÞx2 þ B2 ðPÞxx

_ ¼ L þ M€x ð7Þ
where L is the muscle force, x is the displacement and M is the mass. The second model was referred to as Piecewise model
which is the summation of two equations. One equation describes the system before the 200 kPa limit, and the second equa-
tion after 200 kPa. This second model displayed a high correlation with the experimental data. Both of these models were
then used on a second PAM of slightly different size. Both showed similar results to the first test and appear to bring the
model developed by Reynolds et al. [19] much closer to the measured values.
Based on the work done by Tondu and Lopez [15], a number of modified models have been created. One such model was
presented by Itto and Kogiso [21]. They developed a hybrid model with a focus on a mass flow model that converges to zero
at steady state. The governing mass flow equation is:
_
mðtÞ ¼/ ðtÞm_ i ðtÞ  ð1 / ðtÞÞm_ o ðtÞ ð8Þ

where / is a ratio 2 ½0; 1 that is a function of the valve command voltage, m_ i ðtÞ and m_ o ðtÞ are the mass flow in and out of the
PAM, respectively. Itto and Kogiso used the first law of thermodynamics and the principle of polytropic gases to develop a
number of mass flow equations that are applied individually in the form of ‘‘if then” statements depending on the pressure in
the PAM. However due to the complexity of the modeling, the volume of the muscle was approximated by a second order
polynomial based on experimentally obtained constants.
In a resulting paper Kogiso et al. [22] divided the model into transient and steady state components. In order to validate
the model, the effect of each parameter was evaluated through computer simulations.
Tothova and Pitel have presented a number of papers examining the non-linear behaviour of antagonistic PAM systems.
Their model is based on the fibre geometry and muscle pressure. Their basic geometric model assumes a cylinder with zero
wall thickness and a constant diameter. This model is shown in a paper by Borzikova et al. [23]. From this model, Tothova
96 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

and Pitel [24] developed what they call the advanced geometric muscle model. This new model assumes the diameter to be
variable with the length of the muscle. The ideal gas law with the assumption of constant temperature and the principle of
virtual work give the following force equation:

dV
F ¼ p
dh pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ! ð9Þ
pd21 8L2  6h
2
d1 4L2  h
2
d1 h
2
or F ¼ p þ þ  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
20 15pN 2 15N 15N 4L2  h
2

where p is the muscle pressure, L is half of the length of a single fibre, h is the muscle length, N is the number of windings, d1
is the diameter of the valve. Next the Bernoulli equation is used to create the following flow equation [24]
sffiffiffiffi
2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Q ¼ f v Cq Av Pin  Pout ð10Þ
q
where f v is a direction coefficient, C q is the flow coefficient at a pressure drop of 100 kPa, P in  Pout is the difference in pres-
sures on either side of the valve (the sign is corrected with f v ). This model was then validated with a computer simulation.
Following this, Pitel and Tothova conducted a physical experiment in order to further evaluate the model. Both the simple
and advanced geometric models were simulated along with a lumped parameter model modified from Hill’s muscle model
[25]. The experimental set up consisted of two pneumatic muscles in an antagonistic position connected to a weighted arm.
An imbalance in the pulling forces of the muscles would be seen as an angular deviation of the arm. It was concluded that the
advanced geometric model was the most accurate, with a mean absolute error of 1:12 which corresponds to a relative error
of 4.74%. This model assumes the muscle to be a cylinder and uses Bernoulli equation to calculate the flow through the valves
which is only valid for low velocities when the gas is assumed incompressible. This may lead to error when examining mus-
cles with higher pressure and flow rates.
Sorge [26] used a Newtonian approach to create a complex model of PAM. The model begins by introducing the axial and
longitudinal deformation of the bladder into the conventional Gaylord force model (1). The braid tension is refined by exam-
ining the texture effect of the overlapping fibres. A rotational friction moment in the sheath causes fibres to deviate from
their centerlines and add to the previously obtained equations. The main contribution of this article is a complex equation
describing the total force of the muscle. The dynamic aspect of this model is however somewhat limited. The force equilib-
rium equation (2) from Tondu and Lopez [15] is used. The flow through inlet or outlet orifices is described as a ratio of
upstream and downstream pressures. By using the polytropic law and deriving with respect to time the following filling
equation is obtained:
"  1=n #
np Gin  Gout p0 _
p_ ¼ V ð11Þ
V l0 p

where n is the polytropic exponent, G is the flow rate, V is the volume and the subscript X0 indicates a reference state. The
model is then compared to the experimental data collected by Chou and Hannaford [14] as well as Tondu and Lopez [15]. The
model was found to be in accordance with experimental results.
Tri Vo-Minh et al. [27] used the Maxwell-slip model to describe the hysteresis in PAM. The authors used a lumped param-
eter model with several parallel Maxwell-slip elements. Each element is determined by a stiffness k and a saturation force w
parameter. The hysteric output force then becomes the summation of the individual Maxwell-slip elements:
X
n
F hys out ¼ F i ðwi ; ki Þ ð12Þ
1

Conducting isometric muscle contraction testing, values of the stiffness and saturation force parameters were deter-
mined. With the summation of four Maxwell slip elements this model accurately recreated the total hysteresis of a specific
muscle after experimental testing. Although advanced control systems such as these can partially compensate for frictional
losses and hysteresis, the development of an accurate PAM model would greatly simplify the control scheme requirements.
For instance, a 2014 article by Woods et al. [30] examines the PAM actuators for the control of the edge flaps on helicopter
rotors. The paper describes an agonist antagonist PAM configuration for the bi-directional movement of a flap. The authors
use a PAM model describing the mass flow into an out of the muscles as well as losses along supply lines very similar to the
one presented in this paper. The main differences are the modeling of heat transfer, the force generation, and the bladders
mechanical contribution. First, the Woods model uses the basic Gaylord force model whereas this model uses the extended
version presented in [2]. Next, this model examines the heat transfer through the walls of the PAM, the Woods model does
not. Lastly, the mechanical contribution of the bladder is ignored in our model, not so in the Woods model. This particular
assumption regarding the mechanical properties of the bladder may be somewhat controversial since the bladder is often
included in PAM models.
A recent paper by Wang et al. [31] used the Mooney-Rivlin strain energy density theory to quantify the internal stress in
the bladder material. The authors examined the actuation force vs the contraction ratio in order to experimentally validate
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 97

the model. The experimental results proved to be in good agreement with the developed model. One drawback of the
Mooney-Rivlin approach is that it relies upon experimentally obtained constants. The authors themselves state that the
model may need to be adapted in order to accommodate a wider range of situations due to the highly pressure dependant
properties of the bladder materials. They also state that model inaccuracies may be due to the interaction between the blad-
der and the braids, recommending the need for further investigation.
A different bladder modeling approach was used by Ferraresi et al. [32] who presented a simplified model of the McKib-
ben PAM using the Young modulus. The prototype PAM used a braided sleeve with very few fibres and a thick inner bladder.
This allowed the authors to consider the entire muscle to be a uniform material. The main role of the fibres was to dictate the
relationship between the muscle radius and length.
Part of the difficulty of modeling the PAM comes from the uncertain role of the bladder elasticity. It is intuitively obvious
that the internal bladder of a PAM must be made of a flexible highly deformable material. However these materials have
non-linear mechanical properties. Their deformation can affect the overall force production of the muscle as well as affect
the friction and the hysteresis of the system. Models such as the Mooney-Rivlin theories can be used to describe the mechan-
ical properties of the bladder [31,33]. The models presented in this article are by no means a complete representation of all
current models, but serve to demonstrate the importance of the bladder material. This article however will not be modeling
the elastic properties of the bladder material. In the current paper the bladder is considered to be solely a seal. Its role is
uniquely to create a flexible air tight chamber. To this end a thin latex material was used for its low resistance to elongation
and its minimal wall thickness. For comparison, a thicker butyl rubber was also used.
The literature reviewed above, as well as many other in the open literature report that the dynamic models are mostly
empirical in nature or are approximations with parameters fitted using experimental data. There is no doubt that further
development of a dynamic model for PAM is needed for the advanced development of this actuator.

3. Experimental setup

To analyze the PAM dynamic behaviour and to validate the developed dynamic model, an experimental setup was
designed and built in laboratory as shown in Fig. 2. This setup was designed based on the required type of muscle contraction
test, namely, concentric contraction. Such a contraction occurs when the generated muscle force exceeds the pulling forces
and the muscle contracts. In this test, the muscle is suspended vertically while being fixed at one end and attached to a mass
on its other end. The muscle contraction distance was monitored using a Linear Variable Displacement Transducers (LVDT)
sensor. For operating the PAM, two proportionally controlled electro-mechanical valves are used to inflate and deflate the
muscle. This also controls the muscle contraction distance. The muscle and supply pressure are measured by a pressure
transducer and the supply pressure is controlled using a pressure regulator and a capacitance tank. Finally, all operating
parameters are measured and recorded using a LABVIEW data acquisition system.

Fig. 2. PAM concentric contraction testing setup.

98 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

4. Proposed PAM dynamic model

The PAM dynamic model is an analytical relationship that permits the study of the muscle motion as a function of time.
Several methods exist to derive such a model, in this paper, a Newtonian based approach is proposed. This method uses the
muscle force model that was derived in [2] which assumes that the muscle is composed of three shape elements; a frustum
cone that models each muscle end and a cylinder to model the muscle middle section. These muscle segments are illustrated
in Fig. 3.
With reference to Fig. 3, D is the muscle diameter, d is the muscle end diameter, L is the overall muscle length, Lm is the
muscle middle section length, LL is the horizontal length of the cone and Lz is the cone generator length. Based on this mus-
cle’s structure, the muscle volume V is derived as:
"   2 #
2
2 D d dD D2
V¼ pLL þ þ þ p ðL  2LL Þ ð13Þ
3 2 2 4 4

This volume take into account both end cones as well as the central cylinder. This formulation is used in the derivation of
the force model described in [2]. The force of the muscle is not validated in this work. For the calculation of muscle volume
and bladder area, the muscle is modeled as a simple cylinder. Moreover, the force Fm produced by the muscle is derived as:
F m ¼ NT f cos he cos b ð14Þ
where N is the number of fibres in the muscle’s sleeve, he is the end muscle braid angle and Tf is the tension within the mus-
cle’s fibres which is given by:
! ! ! !
T f ¼ T h1 þ 2T h2  T l ð15Þ
where Tl is fibre tension resulting from the longitudinal stresses, Th1 and Th2 are the fibre tension resulting from the hoop
stresses in the cylindrical and two cone frustum sections, respectively. Based on the experimental setup shown in Fig. 2,
the muscle dynamic model is presented in Laplace domain as:
X
F x ¼ ms2 x ð16Þ

where Fx are the forces acting along the muscle contraction axis, m is the mass of the hanging load, and x is the muscle con-
traction distance. Substituting force components yields:
NT f cos he cos b  mg  f fd  f fb ¼ m  a ð17Þ

where the first term in this equation represents the net force Fm produced by the muscle, the second term is the load of the
mass, the third term is the dynamic friction ffd resulting from the fibre to fibre contact, and the fourth term is the friction
between the fibres and the bladder. Solving for the muscle contraction distance yields to:
ZZ  
1
x¼ ðF m  mg  f fd  f fb Þ dt ð18Þ
m

4.1. Fluid dynamic modeling - valve flow

The PAM system is operated by controlling gas flows and pressures that inflate and deflate the muscle through the valve’s
orifice. The flow through the valve will be considered an isentropic process. Assuming a uniform flow of gas through the
cross section of the valve’s orifice A12, the incoming mass flow rate to the muscle m _ 12 can be expressed as a function of
the density q2 and the speed u2 of the gas at the cross section of the orifice as follow:
_ 12 ¼ q2 A12 u2
m ð19Þ
For an incompressible fluid, the Bernoulli formula can be used to express the flow velocity as a function of pressure. How-
ever, air is a compressible fluid. Therefore, the first law of thermodynamics is used.

Fig. 3. The Muscle’s structure.

 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 99

   
u2 u2
m h1 þ 1 þ gz1 þ Q ¼ m h2 þ 2 þ gz2 þ W ð20Þ
2 2
where h is the enthalpy, m is the mass, g is the gravitational acceleration, z is the height, Q is the heat and W is the work.
Assuming the supply air reservoir is large such the velocity u1 is negligible, no heat exchange, no work and the potential
energy of the air is negligible, (20) yields to:
u22
h1  h2 ¼ ð21Þ
2
The change in enthalpy can be expressed as a function of temperature T and specific heat capacity cp as:
h1  h2 ¼ cp ðT 1  T 2 Þ ð22Þ
or
Rc
cp ¼ ð23Þ
c1
where c is the ratio of specific heat and R is the gas constant.
Comparing (21) with (22):
u22
cp ðT 1  T 2 Þ ¼ ð24Þ
2
re-arranging and divide by cp T 1
u22 T2
1 ¼ ð25Þ
2cp T 1 T 1
replace cp by Eq. (22)
ðc  1Þu22 T 2
1 ¼ ð26Þ
2RcT 1 T1
or
 1
ðc  1Þu22 T1
1 ¼ ð27Þ
2RcT 1 T2
From literature [28], the ratio of the stagnation pressure over the partial pressure is:
 c1
c  c1
c
P0 T0 ðc  1Þ 2
¼ ¼ 1þ M ð28Þ
P T 2
where M is the Mach number
u
M¼ ð29Þ
c
where c is the speed of sound in the medium
pffiffiffiffiffiffiffiffiffi
c¼ cRT ð30Þ
It is important to note that in (28), T is the local temperature and not the stagnation temperature. For compressible fluids,
Andersen [29] states that the total pressure can be defined as the pressure at local isentropic stagnation state. Assuming that
the muscle gas undergoes an isentropic process through the restriction caused by the valve, the muscle supply pressure P1
(total pressure) can be expressed as a function of the muscle’s local pressure P2 as follows:
 c1c
P1 ðc  1Þu22
¼ 1 ð31Þ
P2 2RcT 1
Though this formulation is similar to the more common form found in literature, shown by (33), there is one important
difference. In the common derivation, (24) is rearranged to isolate the stagnation state. Eq. (32) illustrates this difference.
u22
cp T 1 ¼ cp T 2 þ ð32Þ
2
Leading to
0 " #2 1c1
c

P1 @ ðc  1Þ u2
¼ 1þ pffiffiffiffiffiffiffiffiffiffiffi A ð33Þ
P2 2 cRT 2
100 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

Whereas both (31) and (33) are equivalent, (33) includes the local temperature. In this case since the local temperature is
the temperature inside the muscle, it is much more convenient to use the formulation found in (31). In our case the stagna-
tion state is in the supply tank, therefore the stagnation temperature is simply the ambient room temperature.
Solving (30) for the velocity term u2 yields to:
( "  c1 #)1=2
2cRT 1 P2 c
u2 ¼ 1 ð34Þ
c1 P1

Given the gas isentropic process assumption, the density and temperature relationships can be expressed as a function of
the pressures as follow:
 1c
q2 P2
¼ ð35Þ
q1 P1

 c1
c
T2 P2
¼ ð36Þ
T1 P1
Assuming an ideal gas, the density relationship (35) yields to:
 1
P1 P2 c
q2 ¼ ð37Þ
RT 1 P1
_ 12 yields to:
Substituting (34) and (37) into (19) and solving for the flow relationship m
( " 2c  cþ1 #)1=2
A P 2c P2 P2 c
_ 12 ¼ p12ffiffiffiffiffi1
m  ð38Þ
T1 ðc  1ÞR P1 P1

Eq. (38) expresses the muscle in-flow as a function of the ratio of the muscle pressure to the supply pressure. The max-
_ Cr . This flow is achieved when the gas is at a critical velocity u2c,
imum flow through the orifice is given by the critical flow m
given by:
pffiffiffiffiffiffiffiffiffiffiffi
u2c ¼ cRT 2 ð39Þ
The ratio of the flow W12 to Wcr is expressed by N12 as:
2 2  cþ1 31=2
P2 c P2 c
_ 12 6 P1  P1
m 7
N12 ¼ ¼ 6 7 ð40Þ
_ cr 4 c1  ccþ1
m 1
5
2
2 cþ1

Using (40), the flow relationship (38) can be cast as:

KP1 A12 N12
_ 12 ¼
m pffiffiffiffiffi ð41Þ
T1
where K is a factor that depends on R and c and is given by:
"  ccþ1 #1=2
2 c 1
K¼ ð42Þ
R cþ1

The flow relationship given by (41) will be used to predict the gas flow going through the orifice to and from the muscle.
As the gas flows from the supply source to the muscle through the orifice, the muscle pressure P2, the muscle volume V2, as
well as the muscle gas temperature T2 vary accordingly.

4.2. Muscle pressure model

Once the gas has passed through the valve, the variation of its properties can no longer be considered isentropic. Ignoring
the kinetic and potential energies, an energy conservation equation can be written as follows

Q_ þ W
_ þ ðh1 m _ 23 Þ ¼ U_ 2
_ 12  h2 m ð43Þ

where Q_ is the heat transfer rate through the muscle wall, W _ is the work done on the muscle with respect to time, h1 and h2
_ 12 and m
are the enthalpies of the gas as it enters and exits the muscle, respectively, m _ 23 are the mass flow rates of air into and
out of the muscle, respectively and U_ 2 is the change in internal energy of the system with respect to time.
The work, enthalpy, heat transfer and internal energy terms can all be described separately as follows. The heat transfer:
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 101

Q_ ¼ hc Aq ðT atm  T 2 Þ ð44Þ

k
hc ¼ ð45Þ
tb
where hc and Aq are the heat transfer coefficient and the heat transfer surface area, respectively. The heat transfer coefficient
itself is defined as the thermal conductivity of the bladder material k divided by the bladder thickness tb. The heat transfer
area is defined as the area of the bladder that is not being covered by the fibre mesh.
 q
Aq ¼ Atotal  Afibre ¼ r2pL  w Ly N  z ð46Þ
2
A single fibre is in contact with the bladder along its entire length minus the intersection with other crossing fibres. The
number of contact surfaces between fibres can be expressed as
 2
N N
q ¼ 2n  ð47Þ
2 2
where N is the number of fibres in the mesh and n is the number of full revolutions done by each fibre. Assuming the fibres
lie flat against the bladder, then the fibre bladder contact area can be calculated as the total area of the fibres facing the blad-
der, minus the area occupied by the crossing of the other fibres. Since the fibres cross each other in an over-under pattern,
half of the total number of fibre crossings will be subtracted. The fibre bladder contact area can be written as follows.
 q
Afibre ¼ w Ly N  z ð48Þ
2

Atotal ¼ 2p  L  r ð49Þ
where w is the width of the fibres, Ly is the length of the fibres, L is the length of the muscle and z is the portion of the fibre
length that defines the frictional area. The area calculations consider the muscle to be a cylinder with radius r. The cylindrical
assumption does not represent the true shape of PAM, however the inaccuracy induced by this fact is considered to be
minimal.
The work of the system is described as:
_ ¼ P2 V_
W ð50Þ

where V_ is the variation of the muscle volume with respect to time. The enthalpy and the internal energy terms:
h1 ¼ cp T 1 ; h2 ¼ cp T 2 ð51Þ

U 2 ¼ cv m2 T 2 ð52Þ
The enthalpy of the inflow and outflow depend on the gas temperature and specific heat. The internal energy term
depends on the mass of gas inside the muscle. The mass inside the muscle m2 can be determined based on the conservation
of mass plus the initial mass that was inside the muscle (if any) prior to inflation.
Z
m2 ¼ minitial þ _ 12  m
ðm _ 23 Þdt ð53Þ

In order to combine the equations for work, heat transfer, and internal energy into (43), the following manipulations must
first be done. Assuming an ideal gas, the state of the muscle can be identified by using the ideal gas equation. The muscle gas
mass m2 is given as follows:
P2 V 2
m2 ¼ ð54Þ
RT 2
Using the chain rule, (54) is differentiated with respect to time. The perfect gas law is then used to simplify as:
_ 2 P_2 V_ 2 T_ 2
m
¼ þ  ð55Þ
m2 P 2 V 2 T 2
With the differentiation complete, (55) is rearranged.
m2 T 2 _ m2 T 2 _
P2 þ V 2 ¼ m2 T_ 2 þ m
_ 2T 2 ð56Þ
P2 V2
Next, from (52)
R
U_ 2 ¼ _ 2 T 2 þ m2 T_ 2 Þ
ðm
c1
102 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

Recalling that
Rc cp
cp ¼ and c ¼
c1 cv
Placing (56) inside the parenthesises
 
R m2 T 2 _ m2 T 2 _
U_ 2 ¼ P2 þ V2 ð57Þ
c  1 P2 V2
Distributing R inside the parentheses and replacing terms using the perfect gas law (54)
1
U_ 2 ¼ ðV 2 P_2 þ P 2 V_ 2 Þ ð58Þ
c1
Next using (43) and replacing the heat transfer, work, and energy terms with (44), (50), and (57) then solving for the
pressure
1
hc Aq ðT atm  T 2 Þ  P2 V_ 2 þ ðh1 m
_ 12  h2 m
_ 23 Þ ¼ ðV 2 P_2 þ P2 V_ 2 Þ
c1

½ðc  1Þhc Aq ðT atm  T 2 Þ  ðc  1ÞP2 V_ 2  P2 V_ 2 þ cRðh1 m _ 23 Þ ¼ V 2 P_2

_ 12  h2 m

Finally dividing by the volume and taking the integral we obtain an equation for the pressure.
Z
V_ c  1 cR
P2 ¼ cP2 þ _ 12  T 2 m
hc Aq ðT atm  T 2 Þ þ ðT 1 m _ 23 Þ ð59Þ
V V V

4.3. Temperature modeling

The volume of muscle gas was determined in (13) based on the assumed muscle structure in Fig. 3. The muscle pressure is
identified as nonlinear with respect to the muscle contraction distance. Assuming a polytropic process, the muscle gas tem-
perature is related to the muscle pressure as follows:
t ng  1 p
¼ ð60Þ
T ng P
where the value of ng varies between 1 for an isothermal process to 1.4 for an adiabatic process.

4.4. Valve modeling

For muscle operation, as shown in Fig. 2, two proportionally controlled electro-mechanical valves are used to control the
input flow to the muscle and output flow from the muscle. The EPV valve series from Instrument Laboratory valve consists
only of one moving part that is the armature. As electrical current is applied to the valve’s core, a magnetic force is generated
that pulls on the armature. The armature will only travel if the magnetic force is greater than the force of the flat spring on
the armature. For system model simulation, this valve must be analytically identified. Proper valve modeling involves iden-
tifying the effective area of the valve, the derivative of the effective area of the valve with respect to the armature position,
the required actuation force by the core and the derivative of this actuation force with respect to the armature position.
These analytical relationships can only be achieved by dismantling the valve and completing extensive experimental tests.
This was deemed unfeasible and thus for this research, the valve is modeled as an orifice whose effective area is A12 and
hence the flow is a function of the controller input u (electrical current).
The flow of the EPV valve series can be linearly modeled as function of the input current u since their mechanical design
offers limited hysteresis in the flow response and thus allows its hysteresis to be neglected. Using testing results from the
Instrument Laboratory for the valve EC-P-05-4025 at a supply pressure of 344.7 kPa, a linear relationship is derived:
_ 12 ¼ 0:07289  u  0:01056 m3 =min
m ð61Þ
However, (61) cannot be used directly in the PAM modeling since in this study, different supply pressures were applied to
the system. Using (41) and (61), the valve opening area is derived as a function of the input current as:

A12 ¼ 1:80  106  u  2:61  107 m2 ð62Þ

Eqs. (61) and (62) are derived for a supply pressure of 344.7 kPa, however, this is greater than the average muscle oper-
ating pressure and thus it may alter the valve properties. Therefore, the valve was calibrated in the laboratory for 15 distinct
values of supply pressures ranging from 30 kPa to 280 kPa. Based on the experimental results and (41), a valve effective area
has been calculated for different supply pressure tests. However, so far, all flow relationships have been developed when the
valve’s back pressure is atmospheric. This is never the case for the input flow m_ 12 to the muscle, since its back pressure is the
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 103

actual muscle pressure. Thus, the valve back pressure effect on the valve is not considered in these results and relationships.
To characterize further the valve behaviour, including its back pressure, a physical setup was designed to analyze the beha-
viour of a fully open valve when inflating a specific tank volume to a specific pressure and then to deflated back to atmo-
spheric pressure. In this test, the supply pressure and the tank pressure were measured using a pressure transducer
while being recorded as a function of time using a data acquisition system. This test was repeated for eight tank volumes
of 2.1 L, 1.85, 1.6 L, 1.35 L, 1.1 L, 0.85 L, 0.6 L and 0.35 L as well as for two supply pressures of 207 kPa and 138 kPa. Using
the experimental results, a parametric model for the valve is developed. This has resulted in an effective area of the valve
of 0.002580 cm2 when the input current of the valve is 0.35 Amp. This is approximately a 32% reduction in the effective area
of the valve when compared to (62). Finally, to validate the valve characterization in this study, a model was developed to
simulate the pressure vs. time behaviour for a constant tank volume inflation and deflation as experimentally achieved.
Assuming an ideal gas and a polytropic process, the tank pressure Pt relationship was derived with a constant volume as
follows:
mt _
_t
Pt ¼ m ð63Þ
nPt
where mt is the mass of the tank gas and smt is the change in mass of the tank gas with respect to time which is also equiv-
alent to the net flow rate to the tank Wt. This flow rate relationship is given by:
KP1 A12 N12 KP2 A23 N23
m _ 12  m
_t¼m _ 23 ¼ pffiffiffiffiffi  pffiffiffiffiffi ð64Þ
T1 T2
where the index 1 denotes the supply state, the index 2 denotes the tank state and the index 3 denotes the atmospheric state.
Substituting (64) in (63) and rearranging yields:
Z   
KP 1 A12 N12 KP 2 A23 N23 nPt
pt ¼ pffiffiffiffiffi  pffiffiffiffiffi  dt ð65Þ
T1 T2 mt
Using the ideal gas equation, the above equation yields to:
Z   
KP 1 A12 N12 KP 2 A23 N23 nRT 2
pt ¼ pffiffiffiffiffi  pffiffiffiffiffi  dt ð66Þ
T1 T2 V2
where mt is the mass of the tank gas, Tt is the temperature of the tank gas and Vt is the volume of tank gas. The temperature Tt
is determined using the polytropic process relationship given by (60).
Using MATLAB/Simulink, the proposed model for the valve and a constant volume given by (66) has been successfully
validated by predicting the experimental results of the tank pressure vs. time for eight distinct volumes and two supply pres-
sures. For a supply pressure of 208 kPa, the model results and the experimental results for a tank volume of 2.1 L (largest
volume) and 0.35 L (smallest volume) are shown in Figs. 4 and 5, respectively. Thus, this model which includes the valve
characterization will be used later on for developing the PAM dynamic model.

4.5. Dynamic friction modeling

In this work the frictional force is assumed to be the result of fibre-fibre contact as well as fibre bladder contact. Further-
more, the undulations due to the cross fibre weave are ignored, and cross fibres are assumed to simply lie flat on top of each
other. The expression for the fibre-fibre friction force ffr is derived as:

Fig. 4. Valve and constant volume model validation results for a tank volume of 2.1 L and a supply pressure of 207 kPa.
104 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

Fig. 5. Valve and constant volume model validation results for a tank volume of 0.35 L and a supply pressure of 207 kPa.

8
>
>
P 2 Aff ld v <0
< P F þ PA l
> P
F x 6 0; v ¼ 0
x ff d
f fr ¼ P ð67Þ
>
>
> P 2 Aff ls F x > 0; v ¼ 0
:
P2 Aff ld v >0
where Fx is the static friction force, ls and ld are the static and dynamic coefficients of friction, respectively. Such a form,
however, is highly nonlinear and includes a discontinuity at the transition from the static friction regime to the dynamic
one. Such discontinuities are difficult to deal with in numerical simulation. Consequently, for the dynamic case, (67) is
replaced by:
f frd ¼ PAff ldL ð68Þ

where
ldL ¼ ld TanhðK f v Þ ð69Þ
where ffrd is the dynamic frictional force, ldL is the new dynamic friction coefficient, kf is a constant that defines how close
(68) approximates the Coulomb friction and v is the PAM contraction velocity. For the results presented in this paper, kf = 1.
In addition to the fibre-fibre friction, friction between the innermost fibres and the bladder is considered. During pressur-
ization there is a slight movement of the fibres relative to the bladder. As mentioned previously, the number of contact sur-
faces between fibres can be expressed as
 2
N N
q ¼ 2n  ð70Þ
2 2
where N is the number of fibres in the mesh and n is the number of full revolutions done by each fibre. The fibre bladder
friction force can be written as follows.
 q
F fb ¼ P2 lfb w Ly N  z ð71Þ
2
where lfb is the fibre bladder friction coefficient, w is the width of the fibres, Ly is the length of the fibres, and z is the portion
of the fibre length that defines the frictional area. Thus contact area between two crossed fibres is
F ff ¼ w  z ð72Þ

5. PAM model simulation

Given the complexity of the model, only a block diagram approach was proven to be feasible to simulate the proposed
PAM dynamic model. Fig. 6 is a simplified presentation of the MATLAB/Simulink model used to show the inter-
relationships between the muscle analytical relationships. This model is comprehensive and the simulation takes into
account the exact values of all the parameters corresponding to experiments with the exception of three, namely, the exact
nature of the polytropic process of inflating and deflating the muscle, the exact model for the frictional losses, and the
dynamics of a pneumatic valve and the air flow through it.
Fig. 6 shows two main blocs (#8 and 13), the pressurized fluid is first fed to the fluid dynamic model (bloc #8) which
yields to a muscle pressure that drives the muscle motion dynamic model (bloc # 13). The valves orifice effective areas (bloc
# 1) were determined from the experimentally validated model in earlier section. The gas flow (bloc #3) uses (41) to deter-
mine the mass flow through the inflow valve and the outflow valve. The mass flow is then passed to pressure model (bloc #6)
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 105

Fig. 6. PAM dynamic model block diagram.

using (59) which itself is a function of heat transfer (bloc #4) and muscle volume (bloc #7) determined by (73) and (74),
respectively. It is assumed that the heat transfer happens through the bladder only and not through the fibres.
The pressure output is fed back to mass flow (bloc #3) and passed forward to the muscle motion dynamic model (bloc
#13). The temperature inside the muscle (bloc #5) is determined using (60) which relies on pressure variation and polytropic
exponent. The polytropic nature of the temperature equation makes it suitable for multiple possible testing scenarios. The
temperature is directed into the flow gas bloc and serves in the mass flow calculation. The temperature is directed into the
flow model (bloc #3) and serves in the mass flow calculation. The muscle pressure is the main driving variable for the muscle
force model (bloc #10) using (75). Using (76), the sum of all forces, namely, load (block # 9), muscle force (bloc # 10), and
friction (bloc #11) is calculated. The resulting value is the net contraction force generated by the muscle. Dividing by the
mass of the system and integrating twice with respect to time will give the contraction distance of the muscle as given
by (18). The contraction distance is then compared to the original muscle length in order to determine the instantaneous
length of the muscle. The contraction velocity is returned into the muscle friction bloc.

6. Dynamic model validation

Comparisons of the experimental and theoretical simulation results for two muscle configurations are shown in the fol-
lowing figures. Both muscles have an original sleeve diameter of 2.0 cm and an end muscle diameter of 1.3 cm. The muscle
configuration has two fibre revolutions (n = 2) and an original sleeve length of 24 cm. Both muscles use the same outer mesh
material (Flexo PET by TechFlex) and differ only in the internal bladder material. One of the muscles uses a butyl rubber blad-
der with a thickness of 0.8 mm while the other bladder is a latex balloon with a thickness of 0.254 mm.
The proposed model introduces a friction between the bladder and the fibres as well as heat transfer through the walls of
the muscle. Figs. 7and 8 show the muscle pressure validation with respect to time for the experimental test and the model

Fig. 7. Muscle pressure validation results for the butyl rubber muscle prototype for three different polytropic process coefficients.
106 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

Fig. 8. Muscle pressure validation results for the latex balloon muscle prototype for three different polytropic process coefficients.

Fig. 9. Muscle contraction distance validation results for the butyl rubber prototype for three different polytropic process coefficients.

with different polytropic process coefficients ng. These graphs illustrate that the simulation presents a pressure profile clo-
sely resembling the experimental tests.
The contraction distance of the prototypes can be seen in Figs. 9–11. The experimental curve of the butyl rubber proto-
type shows a gradual contraction to steady state value. This behaviour is not seen in the model simulation. Rather, the model
curves have a faster output response than the experimental results in the muscle contraction phase. Iterative variation of the
dynamic friction coefficient did little to correct the slope of the contraction distance vs time. The inability of the model to
accurately replicate the transient response of the butyl rubber prototype illustrates a deficiency in the model. In the current
model, the elastic strain of the bladder is not considered. The latex balloon prototype was created to showcase the behaviour
of the PAM when the elastic elongation of the bladder is insignificant. The latex balloon material was chosen due to its low
stiffness and high compliance. In this paper it is used to demonstrate the scenario in which the bladder strain does not con-
tribute to the muscle behaviour. The contraction distance plot of the latex balloon muscle can be seen in Fig. 10. The exper-
imental results of the latex balloon muscle show a contraction curve that matches the model simulation much more closely
that the butyl rubber muscle shown in Fig. 9. The rapid contraction of the muscle is in accordance with the simulation and
confirms the gradual transient response of the butyl rubber muscle can be explained by the mechanical properties of the
bladder.
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 107

Fig. 10. Muscle contraction distance validation results for the latex balloon prototype for three different polytropic process coefficients.

Fig. 11. Muscle contraction distance validation results for the butyl rubber prototype for three different polytropic process coefficients with fibre-bladder
friction coefficient Ufb = 0.04.

The contraction plot for the butyl rubber muscle also shows a large offset in the steady state length. The friction between
the bladder and the fibres proposed in this model can justify this offset. By increasing the friction coefficient between the
bladder and the fibres to Ufb = 0.04 (from 0.015), new results shown by Fig. 11 are obtained.
As expected from Fig. 11, the friction coefficient between the bladder and the fibres affects the steady state contraction
distance. Furthermore, the coefficient is higher for the butyl rubber than for the latex balloon. The undulation of the fibre
weave is not considered in this model. The complex interaction between the fibres and the bladder may be the cause of these
frictional characteristics. Further study on the fibre bladder interaction is required.
Finally it is worth noting the effect that the polytropic index has on the muscle model simulations. The previous figures
show that the polytropic index ng has a marked effect on the transient and steady state responses of the model. For ng = 1.4
and 1.2, the model curves are comparable to each other, however, for ng = 1, the model value diverges from the other values.
This lends evidence to support the heat transfer component of the proposed model since the isothermal case seems the least
accurate.
During experimental testing, a thermocouple was inserted between the fibres and the bladder of the muscle. Fig. 12 com-
pares the internal gas temperature predicted by the model against the experimentally obtained temperature.
Fig. 12 shows that the polytropic constant ng directly affects the temperature predicted by the simulation. When poly-
tropic process is assumed to be isentropic (ng = 1.4), the model predicts a temperature rise of approximately 4 °C. The exper-
imental results show a temperature rise of just over 1 °C followed by a drop in temperature to bellow the initial value. Using
visual inspection, this plot seems to indicate that a polytropic constant of around ng = 1.2 would be the most accurate
108 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

Fig. 12. Muscle temperature results for the latex balloon prototype for three different polytropic process coefficients with fibre-bladder friction coefficient
Ufb = 0.04.

Table 1
Values of parameters used in the simulation of the latex bladder PAM.

Parameter Symbol Value

Thermal conductivity k 0.14
Diameter of the end connector Dend 0.013 m
Initial diameter of muscle Dref 0.02 m
Initial muscle length Lo 0.24 m
Single fibre length Ly 0.25 m
Cone length Lz 0.008 m
Hanging mass m 2.27 kg
Number of fibre rotations n 2
Number of fibres N 216
Bladder thickness tb 0.254 mm
Fiber thickness tf 0.254 mm
Fiber Bladder Friction coefficient lfb 0.015
Fiber-fibre friction coefficient ld 0.8
Ratio of specific heat c 1.4

assumption. However one must be careful when making the previous statement and must also realise the limitations of the
results. The model proposed in this paper only considers heat generated by the gas, yet there are a number of factors that
contribute to the temperature variation of the PAM. For instance the strain in the bladder and the various frictions would
both generate heat. It is therefore unlikely that the recorded temperature is an accurate measurement of the internal gas
temperature, as the model predicts. In order to truly record the temperature variations of the muscle gas, an alternative mea-
surement method is required.
The proposed model introduces the heat transfer through the walls of the muscle as well as a friction term between the
fibres and the bladder. The model simulation demonstrates that the rapid contraction of the muscle can be explained by the
absence of the bladder strain in the model. Furthermore, the friction between the bladder and the fibres affects the steady
state length of the muscle and is different for the two bladder materials. The results also indicate that the isothermal case
(ng = 1) does not match the experimental curve. The best results are obtained when the polytropic index is higher (ng = 1.2,
ng = 1.4).

7. Conclusions

A Newtonian based fully analytical dynamic model of the PAM has been developed in this work. The model incorporates
an analysis of the pneumatic system and a fibre to fibre friction model. The model results were compared to those from
experiments conducted on prototype PAM’s. The results presented show that the muscle pressure values of the proposed
 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110 109

Table 2
Values of parameters used in the simulation of the butyl rubber bladder PAM.

Parameter Symbol Value

Thermal conductivity k 0.09
Diameter of the end connector Dend 0.013 m
Initial diameter of muscle Dref 0.02 m
Initial muscle length Lo 0.23 m
Single fibre length Ly 0.25 m
Cone length Lz 0.008 m
Hanging mass m 2.27 kg
Number of fibre rotations n 2
Number of fibres N 216
Bladder thickness tb 0.8 mm
Fiber thickness tf 0.254 mm
Fiber Bladder Friction coefficient lfb 0.015
Fiber-fibre friction coefficient ld 0.8
Ratio of specific heat c 1.4

model compare reasonably with the experimental ones, and that the muscle contraction results compares very well with the
experimental ones. The main advantage of this analytical model over other empirical or experimentally based ones reported
in the literature is that it does not require a characterization of a particular muscle configuration.

Appendix A

See Tables 1 and 2.

References

[1] M. Doumit, J. Murillo, A. Lawrynczyk, N. Baddour, Design and evaluation of pneumatic artificial muscle for powered transfemoral prostheses, J. Med.
Biol. Eng. 34 (5) (2014) 439–447.
[2] M. Doumit, A. Fahim, M. Munro, Analytical modeling and experimental validation of the braided pneumatic muscle, IEEE Trans. Robot. 25 (6) (2009)
1282–1291.
[3] C.R. Johnson, R. Pierce, Expansible cover, U.S. Patent 2 238 058, April 14, 1941.
[4] D.H. Hugh, Tensioning device for producing a linear pull, U.S. Patent 2483088, September 27, 1949.
[5] R.H. Gaylord, Fluid actuated motor system and stroking device, U.S. Patent 2844126, July 22, 1958.
[6] V.L. Nickel, J. Perry, A.L. Garrett, Development of useful function in the severely paralyzed hand, J. Bone Joint Surg. Am. 45 (1963) 933–952.
[7] K. Inoue, Rubbertuators and applications for robots, in: Presented at the 4th International Symposium on Robotics Research, 1988.
[8] D. Bergemann, B. Lorenz, A. Thallemer, Actuating means, U.S. Patent 6349746, February 26, 2002.
[9] F. Daerden, Conception and Realization of Pleated Pneumatic Artificial Muscles and their Use as Compliant Actuation Elements Ph.D. dissertation, Vrije
Universiteit Brussel, Brussel, Belgium, 1999.
[10] D.P. Ferris, J.M. Czerniecki, B. Hannaford, An ankle-foot orthosis powered by artificial pneumatic muscles, J. Appl. Biomech. 21 (2) (2005) 189–197.
[11] G. Kim, S. Kang, H. Cho, J. Ryu, M. Mun, K. Kim, Modeling and simulation of powered hip orthosis by pneumatic actuators, Int. J. Control Autom. Syst. 8
(1) (2010) 59–66.
[12] B. Tondu, Modelling of the McKibben artificial muscle: a review, J. Intell. Mater. Syst. Struct. 23 (3) (2012).
[13] H.F. Schulte, Jr., D.F. Adamski, J.R. Pearson, Characteristics of the braided fluid actuator, Engineering, College of – Technical Reports, <https://deepblue.
lib.umich.edu/handle/2027.42/7479.,1961> (accessed 06.10.16).
[14] C.P. Chou, B. Hannaford, Measurement and modeling of McKibben pneumatic artificial muscles, IEEE Trans. Robot. Autom. 12 (1) (1996) 90–102.
[15] B. Tondu, P. Lopez, Modeling and control of McKibben artificial muscle robot actuators, IEEE Control Syst. 20 (2) (2000) 15–38.
[16] P.B. Petrović, Modeling and control of an artificial pneumatic muscle, Part one: model building, in: Presented at the 10th Conf. on Mechanical
Vibrations, Timisoara, 2002.
[17] S. Thongchai, M. Goldfarb, N. Sarkar, K. Kawamura, A frequency modeling method of rubbertuators for control application in an IMA framework, in:
Presented at American Control Conference, 2001.
[18] B.S. Kang, C.S. Kothera, B.K.S. Woods, N.M. Wereley, Dynamic modeling of Mckibben pneumatic artificial muscles for antagonistic actuation, in:
Presented at IEEE International Conference on Robotics and Automation, 2009.
[19] D.B. Reynolds, D.W. Repperger, C.A. Phillips, G. Bandry, Modeling the dynamic characteristics of pneumatic muscle, Ann. Biomed. Eng. 31 (3) (2003)
310–317.
[20] J. Cao, S.Q. Xie, M. Zhang, R. Das, A new dynamic modelling algorithm for pneumatic muscle actuators, Intell. Robot. Appl. (2014) 432–440.
[21] T. Itto, K. Kogiso, Hybrid modeling of McKibben pneumatic artificial muscle systems, in: Presented at IEEE International Conference on Industrial
Technology, 2011.
[22] K. Kogiso, K. Sawano, T. Itto, K. Sugimoto, Identification procedure for McKibben pneumatic artificial muscle systems, in: Presented at IEEE/RSJ
International Conference on Intelligent Robots and Systems, 2012.
[23] J. Boržiková, J. Pitel, M. Tóthová, B. Šulc, Dynamic simulation model of PAM based antagonistic actuator, in: Presented in Carpathian Control
Conference, 2011.
[24] M. Tothova, J. Pitel, Simulation of actuator dynamics based on geometric model of pneumatic artificial muscle, in: Presented at IEEE 11th International
Symposium on Intelligent Systems and Informatics, 2013.
[25] J. Pitel, M. Tothova, Dynamics of pneumatic muscle actuator: measurement and modeling, in: Presented at Control Conference International
Carpathian, 2014.
[26] F. Sorge, Dynamical behaviour of pneumatic artificial muscles, Meccanica 50 (5) (2014) 1371–1386.
[27] T. Vo-Minh, T. Tjahjowidodo, H. Ramon, H.V. Brussel, A new approach to modeling hysteresis in a pneumatic artificial muscle using the maxwell-slip
model, IEEEASME Trans. Mechatron. 16 (1) (2011) 177–186.
[28] P. Balachandran, Fundamentals of Compressible Fluid Dynamics, PHI Learning Pvt. Ltd, 2006.
110 M.D. Doumit, S. Pardoel / Mechanical Systems and Signal Processing 91 (2017) 93–110

[29] B.W. Andersen, The Analysis and Design of Pneumatic Systems, Wiley, 1967.
[30] D.K. Woods, C.S. Kothera, G. Wang, N.M. Wereley, Dynamics of a pneumatic artificial muscle actuation system driving a trailing edge flap, Smart Mater.
Struct. 23 (9) (2014).
[31] X. Wang et al, Non-linear quasi-static model of pneumatic artificial muscle actuators, J. Intell. Mat. Syst. Struct. 26 (5) (2015).
[32] X. Ferraresi et al, Flexible Pneumatic Actuators: A Comparison between The McKibben and the Straight Fibres Muscles, J. Robot. Mech. 13 (1) (1999).
[33] G.K. Klute, B. Hannaford, Accounting for elastic energy storage in McKibben artificial muscle actuators, ASME JDSMC 122 (2) (2000).