Mechatronics 59 (2019) 168–177

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Analysis and optimal sizing of pneumatic drive systems for handling tasks
Adrian Raisch∗, Oliver Sawodny
Institute for System Dynamics (ISYS), University of Stuttgart, Waldburgstr. 17/19, Stuttgart 70563, Germany

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

Keywords: Energy savings and efficiency are declared aims of today’s politics. A large proportion of energy usage is related
Pneumatic systems to manufacturing with its drive systems technologies. Especially pneumatic drive systems have a large savings
Mechatronic systems potential – mostly often based on the fact that their compressed air consumption is a consequence of design, which
Component sizing
in practice is usually done by system designers based on their experience. In this paper, we give a detailed system
Energy efficiency
analysis, yielding promising optimization parameters. For the design optimization we then compare gradient-
Pneumatic actuators
System analysis and design based strategies with a genetic algorithm, showing that the much more efficient gradient-based schemes are
suitable for the application to pneumatic drives. Besides the usage of a standard interior-point optimization
scheme, a discrete-value optimization approach is derived. The optimization results are applied to a three-drive
setup, tested and validated with measurements – emphasizing the importance of optimization of pneumatic drives
and the effectiveness of the discrete-value optimization.

1. Introduction chosen properly for a specific motion task is derived in terms of an eigen-
frequency. Based on this eigenfrequency analysis, it is also possible to
Energy efficiency in automation and production industry has become easily adapt the system parameters in order to reduce the consumption
a field of deep interest over the past decades. Still today, the most com- of the drive system. An approach based on an iterative optimization
monly used drive technology in plant manufacturing is the usage of scheme is given in [9]. There, a genetic algorithm is applied to improve
compressed air drives – as they are cheap in acquisition, and robust the energetic behavior of pneumatic drives.
in operation with respect to external circumstances. The second focus of energy saving measures for pneumatic drives
However, compressed air systems are viewed as highly inefficient relies on an efficient operation. In [10], the energy efficiency of pneu-
and causing waste of energy [1]. This is mainly caused by the fact that matic servo drives is examined and energy optimal control strategies
for pneumatic drives, the energy consumption is an implication of the are investigated. A promising and systematically interesting method is
system design and not of its operation (as it is for electromechanical found in [11], where a pneumatic Wheatstone-bridge connection of the
drives) [2]: When operated by a standard switching control, pneumatic control valves is used to reduce the energy consumption by up to 85%
cylinders are always fully inflated and deflated independently from their depending on the application. In [12], a similar approach using four
load. Therefore, the design is of special concern in research of pneumatic fast switching control valves is taken into account – however with only
drives. partial position information at the ends of the cylinder stroke.
Most of the contemporary research regarding the energy efficiency of The authors of [13] also consider a control strategy with a supply
pneumatic drives is based on physical models of the components such as air shutoff for pneumatic cylinder drives resulting in remarkable en-
drives, valves [3,4], and tubes [5,6]. Energy saving measures for pneu- ergy savings. In [14] an energy efficient tracking strategy for pneumatic
matic drives found in literature can then be divided into a proper di- cylinders is presented, an issue also discussed in [15] where the conse-
mensioning and an efficient operation. quences for system design are taken into account.
Regarding the design process, it is the practical standard in indus- While the existing methods on energy savings of pneumatic drives
trial environments to select cylinders, valves and throttles based on ex- discussed before mainly lack of insight into the design process (such as
perience of the applicator. Addressing this issue, [7] considers the opti- genetic optimization algorithms) or are based on approximate formulas,
mization of a pneumatic drive system with a close link to implementa- this paper presents a detailed system analysis and gradient-based opti-
tion. In [8], an approximate formula for deciding whether a cylinder is mization strategies relying on those results. By this, we can reduce the

∗
Corresponding author.
E-mail addresses: [email protected] (A. Raisch), [email protected] (O. Sawodny).
URL: https://www.isys.uni-stuttgart.de (A. Raisch)

https://doi.org/10.1016/j.mechatronics.2019.04.003
Received 15 November 2018; Received in revised form 18 February 2019; Accepted 1 April 2019
0957-4158/© 2019 Elsevier Ltd. All rights reserved.
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

Fig. 2. Typical configuration for the usage of pneumatic drives. A double stroke
pneumatic cylinder is controlled by two 3/3-directional valves with supply pres-
sure 𝑝h .

the most relevant system parameters. In the following, a mathematical

Fig. 1. The experimental setup considered for validation in this contribution model for describing the corresponding dynamic behavior is presented.
consists of three pneumatically conducted transitions: (1): a horizontal motion
of 500 mm, (2): a vertical lifting by 200 mm, and (3): a horizontal displacement
of 100 mm. Thus, we combine three different typical motions. The cylinders are 2.1.1. Pneumatic cylinders
actuated by a valve terminal with a selected supply pressure. During the motion, Pneumatic drives convert energy from compressed air into mechan-
positions, chamber pressures, and compressed air consumption are captured. ical work. Therefore, thermodynamical aspects – which are explained
later – as well as a mechanical modeling is necessary. The mechanics of
pneumatic cylinders are described by Newton’s second law
number of function evaluations and system simulations for optimization, ( )
while aiming to minimize the total compressed air intake for a motion 𝑚𝑠̈ = 𝐴A 𝑝A − 𝐴B 𝑝B − (𝐴A −𝐴B )𝑝0 − 𝐹F − 𝐹T + 𝐹G − 𝐹H (1)
task.
with the piston position s, mass m, forces resulting from pressures in the
chambers pA,B and ambiance 𝑝0 multiplied by their respective area Ai ,
1.1. Contributions and the friction force

For sizing pneumatic cylinder drives, only heuristic rules based on 𝐹F = 𝑓c sign(𝑠̇ ) + 𝑓v 𝑠̇ (2)
approximate formulas or contributions on optimization with excessive as a combination of viscous and Coulomb friction. The friction force
numbers of function evaluations are available. So the key contribution (2) occurs at the contact surfaces of piston and housing, and thus we
of this paper is to demonstrate that gradient-based optimization schemes assume the friction parameters fc and fv to depend on the cylinder diam-
are suitable for optimizing the design of pneumatic cylinder drives. For eter with which the contact surfaces are scaling. In Fig. 3, an exemplary
this purpose, we give a validated model of pneumatic drives and analyze friction measurement for a cylinder with 𝑑 = 32 mm as well as identified
the system’s parameters. Based on the results of the analysis, we propose parameter sets for different sizes together with an affine approximation
the optimization of the relevant parameters. We then present and com-
pare three different optimization strategies for pneumatic drive systems 𝑓c = 𝑓c,0 + 𝑓c,d 𝑑 (3a)
and implement the results to the setup depicted in Fig. 1. The obtained
measurements confirm the optimization results and demonstrate the im- 𝑓v = 𝑓v,0 + 𝑓v,d 𝑑 (3b)
portance of a proper system design.
are illustrated. The gravity force

1.2. Outline of the paper 𝐹G = 𝑚𝑔 sin 𝛼 (4)

arises from the mounting orientation 𝛼, where we consider the most

The remainder of this paper is organized as follows: After a model common cases 𝛼 ∈ {0, 𝜋2 } in this publication.
derivation and validation in Section 2, we give a detailed system analysis The terminal force 𝐹T is included using a Hunt-Crossley contact force
in Section 3. Based on those results, we formulate a system sizing prob- model [16] to obtain a system model that retains its validity once the
lem and propose optimization strategies in Section 4. The algorithms motion is completed by
are then applied to the exemplary setup from Fig. 1, where the results
are discussed in Section 5, before closing the paper with some short ⎧𝛾|𝑠|𝛽 𝑠̇ − 𝛿|𝑠|𝛽 , 𝑠≤0
⎪
conclusions in Section 6. 𝐹 T = ⎨0 , 𝑠 ∈ (0, 𝑙) (5)
⎪−𝛾|𝑙 − 𝑠|𝛽 𝑠̇ − 𝛿|𝑙 − 𝑠|𝛽 , 𝑠≥𝑙
⎩
2. Model derivation and validation
with positive constants 𝛾, 𝛽, 𝛿. The holding force FH is an external force
which can be applied to an external object for manipulation. As de-
2.1. Modeling of pneumatic drives
scribed in [8], the relation of the cylinder areas can be approximated
by
Our goal is to optimize the energetic behavior of pneumatic drives. A
sketch of such a pneumatic cylinder drive is given in Fig. 2, together with 𝐴B = 𝜇𝐴A , 𝜇 = 0.88 (6)

169
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

Fig. 4. Identified flow function with critical pressure ratio 𝑏 = 0.20. The slightly
modified version matches the original flow function well and ensures differen-
tiability.

Fig. 3. Illustration of the cylinder friction: An exemplary measurement for the with the density of air 𝜌0 , sonic conductance C, critical pressure ratio
friction force together with its static approximation is depicted in the upper
b, upstream and downstream pressure 𝑝us , and 𝑝ds , as well as the valve
plot. The lower figure shows the identified parameters for different cylinder
position 𝑢 ∈ {−1, +1}.
sizes together with an affine approximation used in this contribution.
The corresponding flow function Ψ is defined by
⎧ 𝑝ds
by which (1) simplifies to ( ) ⎪ 1, 𝑝us
<𝑏
( ) 𝑝 ⎪√√ (𝑝 )
𝜋𝑑 2 Ψ ds , 𝑏 = ⎨√ ds −𝑏 . (13)
𝑚𝑠̈ = 𝑝A − 𝜇𝑝B − (1 − 𝜇)𝑝0 − 𝐹F − 𝐹T − 𝐹H − 𝑚𝑔 sin 𝛼, (7) 𝑝us √1 − 𝑝us 2, 𝑝ds
≥𝑏
4 ⎪ 1−𝑏 𝑝us
⎪
with d being the diameter of the cylinder. ⎩
As measure for the energy consumption J of a pneumatic drive, we
2.1.2. Pressure dynamics
consider the compressed air mass inflow
The air inside the two chambers j ∈ {A, B} passes through a ther-
𝑡f
modynamical process. As commonly described in literature [17], we as- 𝐽= 𝑚̇ + + 𝑚̇ + d𝜏 (14)
sume the air to be governed by a polytropic process, where the product ∫0 A B

of pressure and volume exponentiated by the polytropic index n is con- into chambers A and B during the whole double-stroke process from
stant for arbitrary times 𝑡1 and 𝑡2 : 𝑡 = 0 until 𝑡 = 𝑡f . The superscript + denotes that only positive portions
𝑝𝑗 (𝑡1 )𝑉𝑗𝑛 (𝑡1 ) = 𝑝𝑗 (𝑡2 )𝑉𝑗𝑛 (𝑡2 ) . (8) of the mass flows are being integrated.

Considering air as an ideal gas, it fulfills the relation 2.2. Adjustments to the model
𝑝𝑗 𝑉𝑗 = 𝑚𝑗 𝑅𝑇𝑗 , (9)
As we will next analyze the system described by (7), (10)–(14)
where mj is the mass of air inside the chamber, Tj is its temperature, and based on a sensitivity analysis and later also use those information
R is the specific gas constant. for gradient-based optimization, our model has to satisfy differentiabil-
The total differential of the polytropic process Eq. (8), combined with ity requirements. Hence, the flow function, which is non-differentiable
the ideal gas Eq. (9), and the corresponding relations for the chamber around 𝑝ds = 𝑏𝑝us and 𝑝ds = 𝑝us , is smoothed. We do so by introducing a
volumes results in the pressure dynamics quadratic approximation for 𝑏 ≤ 𝑝ds ∕𝑝us < 𝑏 + 𝜖1 , and 𝑏 − 𝜖2 ≤ 𝑝ds ∕𝑝us <
( )
𝑛 𝜋𝑑 2 1 with small 𝜖 1,2 . The result of this modification is illustrated in Fig. 4.
𝑝̇ A = 2 𝑅𝑇 𝑚̇ A − 𝑠̇ 𝑝A (10) Regarding the model consistency it should be pointed out, that the mod-
𝜋𝑑 4
𝑠 + 𝑉D,A
4 ( ) ified flow function vanishes for 𝑝ds = 𝑝us .
𝑛 𝜋𝑑 2
𝑝̇ B = 𝑅𝑇 𝑚̇ B + 𝜇 𝑠̇ 𝑝B . (11)
𝜋𝑑 2 4 2.3. Overall cylinder model
𝜇 4
(1 − 𝑠) + 𝑉 D,B

Here, 𝑚̇ A and 𝑚̇ B denote the mass flows of air into the two chambers, For a more compact notation, we introduce the state x and parameter
VD,A , VD,B are dead volumes inside the cylinder and the corresponding vectors 𝜉, Φ
transmission lines, while the temperature 𝑇A = 𝑇B = 𝑇 is assumed to be [ ]
constant. 𝑥 = 𝑠 𝑠̇ 𝑝A 𝑝B T , (15)
[ ]T
𝜉 = 𝑝h 𝑑 𝐶A 𝐶A 𝐶B 𝐶B ,
E R E R (16)
2.1.3. Pneumatic valves
The previously given pressure dynamics described by (10) and [ ]
Φ= 𝑚 𝛼 T. (17)
(11) are driven by their air inflows and outflows 𝑚̇ A and 𝑚̇ B delivered
through a valve. The standard in modeling the mass flow through pneu- Here, 𝜉 describes design variables (supply pressure, cylinder diameter,
matic valves in literature is the description as a nozzle [18]. and sonic conductances for extension and retraction of the two chambers
An elliptical approximation of the mass flow through the orifice re- A and B), while in Φ parameters representing the given motion task
sults in the commonly used {C,b}-description as introduced in [19]: (load mass and orientation) are collected. Then, the system dynamics of
( ) a single pneumatic cylinder described by (7), (10)–(13) can be written
𝑝 as
𝑚̇ = 𝜌0 𝑝us 𝐶Ψ ds , 𝑏 𝑢 (12)
𝑝us
𝑥̇ = 𝑓 (𝑥, 𝜉, Φ), 𝑥(0) = 𝑥0 (𝜉). (18)

170
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

This sensitivity ODE (20) is a linear time-variant (𝑛x × 𝑛𝜉 )-matrix

ODE which can be solved simultaneously with a simulation run of the
underlying system ODE.

3.1. Influence on energy consumption

To calculate and analyze the compressed air consumption for a gen-

eral polytropic coefficient n, we start from the pressure dynamics
( )
𝑛
𝑝̇ = 𝑅𝑇 𝑚̇ − 𝑉̇ 𝑝 . (21)
𝑉
Without loss of generality, we first consider the extension process of a
pneumatic cylinder, where chamber A is the inflated driving chamber.
Using 𝑉 = 𝐴𝑠 + 𝑉0 , we reformulate (21) as

(𝐴𝑠 + 𝑉0 )𝑝̇ + 𝑛𝐴𝑝𝑠̇ = 𝑛𝑅𝑇 𝑚̇ . (22)

Integrating (22) from 𝑡 = 0 to 𝑡 = 𝑡f where 𝑡f is chosen sufficiently large,

such that 𝑝(𝑡f ) = 𝑝h and 𝑠(𝑡f ) = 𝑙 yields
𝑡f 𝑡f
(𝐴𝑠 + 𝑉0 )𝑝̇ + 𝑛𝐴𝑝𝑠̇ d𝜏 = 𝑛𝑅𝑇 𝑚̇ d𝜏. (23)
∫0 ∫0

Fig. 5. Model validation for a double stroke cylinder. Depicted are position and Employing
pressure behavior of measurement and model. 𝑡f [ ] 𝑡f
𝑡
(𝐴𝑠 + 𝑉0 )𝑝̇ d𝜏 = (𝐴𝑠 + 𝑉0 )𝑝 0f − 𝐴 𝑠̇ 𝑝d𝜏, (24)
∫0 ∫0
2.4. Identification and validation we obtain for the compressed air flow
[ ]
Next, the previously derived model equations are validated us- 𝑡
𝑡f (𝐴𝑠 + 𝑉0 )𝑝 0f (𝑛 − 1)𝐴 𝑡f
ing measurement data. To do so, the parameters for the friction 𝑚̇ d𝜏 = + 𝑠̇ 𝑝d𝜏 . (25)
model, sonic conductances and critical pressure ratio are identified. ∫0 𝑛𝑅𝑇 𝑛𝑅𝑇 ∫0
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
In Fig. 5, the comparison of simulation and measurement is de- (∗)
picted for a 500 mm double stroke performed with a Festo DSNU-
25-500 pneumatic cylinder and a load mass on the extension of Assuming 𝑠̇ 𝑝 ≥ 0, ∀𝑡 ∈ [0, 𝑡f ] for a purely expanding process, we know
2 kg. This corresponds to the linear transition 1 from the setup pre- that (∗ ) ≥ 0. By this, we can bound the consumption from above and
sented in Fig. 1. The parameters of the simulation are identified to below:
𝑓c = 11.13 N, 𝑓v = 12.53 𝑁𝑠/m, 𝐶AE = 5.05 𝑚𝑚3 ∕𝑠𝑃 𝑎, 𝐶AR = 5.55 𝑚𝑚3 ∕𝑠𝑃 𝑎, [ ] [ ]
𝑡 𝑡
(𝐴𝑠 + 𝑉0 )𝑝 0f 𝑡f (𝐴𝑠 + 𝑉0 )𝑝 0f
𝐶BE = 5.15 ∕mm3 sPa, 𝐶BR = 6.40 𝑚𝑚3 ∕sPa, and 𝑏 = 0.27. Obviously, the
≤ 𝑚̇ d𝜏 ≤ . (26)
model equations describe the system behavior in a satisfying manner 𝑛𝑅𝑇 ∫0 𝑅𝑇
and hence are suitable for the subsequent analysis and optimization.
For a general polytropic coefficient n ∈ [1, 𝜅], (25) can be utilized to
calculate the derivatives of the mass inflow w.r.t. the parameters 𝜉 via
3. System analysis the system sensitivities and
{ } {[ ]𝑡 }
Before optimizing the system parameters we first take a closer look 𝜕
𝑡f
𝜕 (𝐴𝑠 + 𝑉0 )𝑝 0f
𝑚̇ d𝜏 =
on those and their effects on the system behavior and boundary con- 𝜕𝜉𝑟 ∫0 𝜕𝜉𝑟 𝑛𝑅𝑇
ditions such as travel time, impact velocity and holding force. In the { } 𝑡f
𝜕 (𝑛 − 1)𝐴
following section, system parameters arising from the drive design, par- + 𝑠̇ 𝑝d𝜏
𝜕𝜉𝑟 𝑛𝑅𝑇 ∫0
ticularly supply pressure 𝑝h , diameter d and sonic conductances 𝐶𝑖𝑗 are ( )
analyzed. Where not analytically possible, we will use the sensitivity (𝑛 − 1)𝐴 𝑡f 𝜕 𝑠̇ 𝜕𝑝
+ 𝑝 + 𝑠̇ d𝜏. (27)
equations [20]. The sensitivity 𝑛𝑅𝑇 ∫0 𝜕𝜉𝑟 𝜕𝜉𝑟

𝜕𝑥 ̃
𝑆(𝑡)∶= (𝑡, 𝜉, Φ) ∈ ℝ𝑛x ×𝑛𝜉 (19)
𝜕𝜉 3.1.1. Isothermal process
describes the dependence of the solution x(t) to an ODE 𝑥̇ = In the common case of an isothermal process with 𝑛 = 1, the esti-
𝑓 (𝑥, 𝜉, Φ), 𝑥(𝑡0 , 𝜉) = 𝑥0 (𝜉) with respect to a (small) variation in the pa- mate (26) becomes the exact solution for the mass inflow. Applying the
rameters 𝜉. Differentiating the ODE 𝑥̇ = 𝑓 (𝑥, 𝜉, Φ) together with its initial previous derivations also for the retraction process then yields
conditions towards 𝜉 yields ( )
𝜌 𝑙𝜋𝑑 2
𝐽 = 0 (𝑉D,A + 𝑉D,B )(𝑝h − 𝑝0 ) + (1 + 𝜇)𝑝h (28)
𝜕𝑥(𝑡0 , 𝜉) 𝜕𝑥 (𝜉) 𝑝0 4
= 0 (20a)
𝜕𝜉 𝜕𝜉
From (28), we can directly conclude the influence of supply pressure
( ) and cylinder diameter on the energy consumption
𝜕 d𝑥(𝑡, 𝜉, Φ) 𝜕𝑓 (𝑥, 𝜉, Φ) 𝜕𝑥(𝑡, 𝜉, Φ) 𝜕𝑓 (𝑥, 𝜉, Φ)
= + (20b) ( )
𝜕𝜉 d𝑡 𝜕𝑥 𝜕𝜉 𝜕𝜉 𝜕𝐽 𝜌 𝑙𝑑 2 𝜋
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏟⏞⏞⏞⏟ = 0 𝑉D,A + 𝑉D,B + (1 + 𝜇) (29a)
=𝑆(𝑡) 𝜕𝑝h 𝑝0 4
=𝑆̇ (𝑡)

171
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

Fig. 7. Sensitivity of the impact velocity w.r.t. the cylinder diameter for a larger
parameter set of horizontal motion tasks. Depending on the task configuration,
a change in d will lead to an increase or decrease of the impact velocity.
Fig. 6. Illustration of the piston velocity until the impact time and the corre-
𝜕 𝑠̇
sponding sensitivities 𝜕𝜉 (𝑡, 𝜉, Φ): Depicted are the influence of supply pressure
𝑟 What we can conclude from (30a)-(31b) is that the holding force
𝑝h , cylinder diameter d, and sonic conductances for inflating chamber A and
grows approximately quadratic with the cylinder’s diameter d and ap-
deflating chamber B. For a better comparability of the different sensitivities in
proximately linear with respect to the supply pressure 𝑝h .
different units, the results are scaled by the nominal values 𝜉̄𝑟 of the sensitiv-
ities and for the given illustration, an additional multiplication by 0.1 is done
3.2.2. Impact Velocity
representing a parameter variation of [10]%.
As the piston velocity 𝑠̇ (𝑡) is the result of a nonlinear dynamic process,
the impact velocity

𝜌 𝑙𝑑𝜋 𝑣∗ = 𝑠̇ (𝑇 ), (32)
𝜕𝐽
= 0 (1 + 𝜇)𝜌0 𝑝h (29b)
𝜕𝑑 2𝑝0 where T corresponds to the travel time, cannot be analyzed analytically.
However, employing the sensitivity ODEs (20), the influence of the sys-
and that the sonic conductances do not influence the compressed air
tem parameters on the velocity can be considered using
consumption. Considering the relative change in consumption, we also
see that a decrease in the supply pressure 𝑝h will lead to a decrease in 𝜕𝑣 𝜕 𝑠̇ ||
= . (33)
the cost J by approximately the same percentage, while the influence of 𝜕𝜉 𝜕𝜉 ||𝜉̄
the diameter d is quadratic but decreases with growing dead volumes Fig. 6 shows the time behavior of the piston velocity, and the corre-
VD,j . sponding normalized influence of the system parameters. Regarding the
piston velocity and its value at the impact time instance, intuitive re-
lations for supply pressure and sonic conductances of exhausting and
3.2. Influence on boundary conditions
pressurizing hold: increasing 𝑝h , 𝐶AP , or 𝐶BE will lead to higher velocities
and vice versa. This can also be seen in Fig. 6, where the sensitivities of
The typical boundary conditions for a pneumatic motion task in prac-
those three parameters are positive during the whole stroke. However,
tice are holding forces, impact velocities, and travel times. Those three
the cylinder diameter d can affect the impact velocity in two ways as il-
are analyzed in the following.
lustrated in Fig. 7. Depending on the task configuration, the velocity can
in- or decrease, basically arising from the growing volume that needs to
3.2.1. Holding Force be inflated. If the inflow of pressurized air is the limiting factor in the
Similarly to the compressed air consumption, the influence on the dynamical process, an increasing diameter will lead to larger transition
holding force can be analyzed analytically. The holding force is the re- times. Fig. 7 displays the influences in case of a horizontal motion task
sulting force at stroke end and is given by (𝛼 = 0). Note that the results for 𝛼 ≠ 0 are qualitatively similar, since the
weight force only acts as a constant additional force during the stroke.
𝐹H,E = 𝐴A (𝑝h − 𝑝0 ) − 𝑚𝑔 sin 𝛼 (30a)
3.2.3. Travel Time
𝐹H,R = 𝐴A (𝜇𝑝h − (1 − 𝜇)𝑝0 ) + 𝑚𝑔 sin 𝛼 (30b)
The travel time is the time difference between the start of the ex-
for expansion (𝐹H,E ) and retraction (𝐹H,R ). The holding force represents tension or retraction process and the time where the piston reaches the
the available force that can be applied by the cylinder to an external corresponding end stop:
object at stroke end. If there is no external object applying a force, the
𝑇𝑘 = min 𝑡 − 𝑡start
𝑘 (34)
holding force is absorbed by the terminal force (𝐹H = 𝐹T ). In the same 𝑡 > 𝑡start
𝑘
way as the compressed air consumption, the holding forces are not influ- {
𝑠(𝑡) ≤ 0, 𝑘 = E
enced by the sonic conductance but by the supply pressure and diameter s. t.
𝑙 − 𝑠(𝑡) ≤ 0, 𝑘 = R
of the cylinder. Differentiating (30a) and (30b) with respect to 𝑝h and d
yields In (34), the index k corresponds to the extension and retraction pro-
cess, 𝑡start is the time where the process is started (i.e. the valve position
𝜕𝐹H,E 𝜋𝑑 2 𝜕𝐹H,E 𝜋𝑑 𝑘
= , = (𝑝 − 𝑝0 ), (31a) switches), and 𝑠end 𝑘
is either 0 or the cylinder length l – depending on k.
𝜕𝑝h 4 𝜕𝑑 2 h
The travel time can either be obtained from (34) directly, or an addi-
𝜕𝐹H,R 𝜋𝑑 2 𝜕𝐹H,R 𝜋𝑑 tional dynamic state for Tk with
= 𝜇, = (𝜇𝑝h − (1 − 𝜇)𝑝0 ). (31b)
𝜕𝑝h 4 𝜕𝑑 2 ( )
𝑇̇ 𝑘 = 𝜎 (𝑠end
𝑘 − 𝑠(𝑡))𝑢(𝑡) , 𝑇𝑘 (0) = 0 (35)

172
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

Fig. 8. Influence of the exhaust-sided sonic conductance on the travel time.

Depicted is the relative change resulting from a variation of 10%.

can be appended. Since we are interested in the parameter dependencies

for Tk , we employ the latter one and approximate the unit step function
𝜎( · ) by a differentiable arctan-function. As the behavior of the piston
velocity directly results in the travel time, the influence of the design
parameters are qualitatively similar. Increasing the sonic conductances,
or the supply pressure lowers the travel time, while the influence of
the cylinder diameter d is twofold. Fig. 8 illustrates the influence of the
exhaust-sided sonic conductance: Increasing the parameter leads to a
decreased travel time. However, for configurations dominated by the
load mass, the influence is small.

3.3. Interim conclusions

Fig. 9. Influence of the design parameters on cost function and boundary con-
ditions for the extension. Depicted are the simulation results of 420 differently
Summing up the results from the system analysis, we can conclude: parametrized motion tasks. The four subplots illustrate the influence of the de-
sign parameters 𝜉 = {𝑝h , 𝑑, 𝐶AP , 𝐶BE } on the cost functional, as well as the bound-
• Crucial for the energy consumption of pneumatic drives are supply
ary parameters travel time, impact velocity, and holding force for the extension
pressure ph and diameter d. Regarding the fulfillment of boundary
process. The bars represent the range of the values and the color mapping of
conditions of motion tasks, also the sonic conductances for in- and the bars indicate the number of obtained simulation results in a specific area:
deflating have a significant impact. the brighter the color is, the more simulations delivered a result in this specific
• The sensitivities and the analytic derivations (27)-(31b) yield the di- area.
rections of descent of cost and constraints later used for optimization
in Section 4.2.
• For the optimization of real-world applications, usually components The key goal of this contribution is to find a (parameter) configura-
from a catalog have to be chosen. This corresponds to a discrete set tion for a set of connected pneumatic drives, such that a given motion
of parameters such as the cylinder diameter d. From the analysis, task is fulfilled and the energy – or compressed air – consumption is min-
we see that rounding the diameter has a noticeable impact on the imized. For the consideration of multiple cylinders, we expand (18) by
behavior but can be compensated by adjusting other parameters. introducing the drive index 𝑖 = 1, … , 𝑛c . By this, we can formulate this
design problem as an optimization problem
In Fig. 9, the influence of the design parameters on the compressed 𝑛c
∑
air consumption and boundary conditions on the extension are illus- minimize 𝐽𝑖 (36a)
𝜉𝑖
trated graphically. Depicted are the relative changes corresponding to 𝑖=1
a parameter increase of 10 % over a variety of motion tasks. By this, subject to 𝑥̇ 𝑖 = 𝑓 (𝑥𝑖 , 𝜉𝑖 , Φ𝑖 ), (36b)
Fig. 9 gives a quick overview on the outcome of parameter variations,
𝑥𝑖 (0) = 𝑥0,𝑖 (𝜉𝑖 ), (36c)
closing our system analysis.
𝜉𝑖 [1] = 𝜉𝑙 [1], ∀𝑖, 𝑙 ∈ {1, … 𝑛c }, (36d)
4. System sizing problem and solution strategies
𝑔(𝜉𝑖 ) ≤ 0, (36e)
As mentioned in the introduction, pneumatic drives are commonly
⎡𝐹̄H,𝑘,𝑖 − 𝐹H,𝑘,𝑖 (𝜉𝑖 )⎤
selected by engineers based on their experience. Additionally, large ℎ𝑖 (𝑥𝑖 , 𝜉𝑖 ) = ⎢ 𝑇𝑘,𝑖 (𝑥𝑖 ) − 𝑇̄𝑘,𝑖 ⎥ ≤ 0. (36f)
safety buffers are taken into account when designing an industrial ma- ⎢ ∗ ⎥
⎣ 𝑣𝑘,𝑖 (𝑥𝑖 ) − 𝑣̄ ∗𝑘,𝑖 ⎦
chine. This often results in oversized drives consuming lots of energy.
What is usually quite well known to the system designer, are the task Here, the cost function (36a) is given by the sum over (14), (25),
specific parameters for the motion task – in our case the load mass, a or (28), respectively, for all cylinders i. The underlying dynamics of the
predefined transition time, and demands on initial and terminal condi- pneumatic cylinders are incorporated by (36b), and the boundary con-
tions. ditions of the motion tasks as described before, are represented by (36f):

173
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

The holding forces have to be larger than a demanded force 𝐹̄H,𝑘,𝑖 , the gradient-based optimization scheme, directly working with discrete sets
travel times must be less or equal the demanded travel times 𝑇̄𝑘,𝑖 , and of parameters.
the impact velocities have to be below 𝑣̄ ∗𝑘,𝑖 . The index k corresponds First, the non-discretized parameters – in our example case the sup-
to the process of {E}xtension and {R}etraction. Additionally, (36d) de- ply pressure 𝑝h and the sonic conductances 𝐶𝑗,𝑖 𝑘 – are discretized. We

mands that all cylinders are driven by a common supply pressure 𝑝h do this in steps of 20 mbar for the supply pressure and 0.05 /mm3 s · Pa
connecting the systems with 𝜉[1] denoting the first element of 𝜉, and for the conductances. At this point note that with standard mechani-
(36e) describes additional practical boundaries on the parameters 𝜉 in- cal pressure regulators, and losses in tubes and connections, already a
troduced subsequently. Problem (36) now is a parameter optimization matching within this precision for the supply pressure as well as for
problem with 6𝑛c optimization variables and 4𝑛c underlying differential the sonic conductances is hard to achieve in practical realization. By
equations, which have to be solved in each iteration. this, we obtain the discrete-value set  of parameters 𝜉. At this point
note that the cardinality of  grows exponentially with the number
4.1. Design parameters and their practical boundaries of parameters and hence already for large discretization steps, it gets
practically impossible, to solve the optimization problem (36) by full
The design parameters (16) for the system sizing problem (36) have enumeration.
to fulfill some typically given boundaries included in (36e): In order to reduce the computational effort, we are interested in em-
ploying a gradient-based scheme to the discretized parameter set. Con-
• The supply pressure ph basically can be set freely using a pressure
sidering the gradients of cost and boundary conditions, we already know
regulator for the system. However, for achieving a smooth transition
the gradients for cost (29a), (29b) and holding forces (31a), (31b) from
behavior of the cylinder, it should not be chosen below 3 bar.
the analysis section. For the boundary conditions on travel time and im-
• The cylinder diameter d typically is from a discrete value set as
pact velocity, the relations between the system’s sensitivity ODEs from
manufacturers produce fixed sizes of cylinders. Common diameters,
the analysis and the gradients of the optimization problem can be uti-
which we consider here are d ∈ {8, 12, 16, 20, 25, 32, 40} mm.
lized in order to reduce the number of function evaluations by supplying
• The sonic conductances are limited from above by the interconnec-
gradients to the optimization algorithm.
tion of conductances of valve and tubes. It can further be decreased
Regarding the effect of a parameter variation Δ𝜉 around a configu-
by throttles and is typically below 20/mm3 s · Pa.
ration 𝜉̄ on the boundary conditions (36f), we obtain
4.2. Optimization strategies ℎ(𝑥(𝜉̄ + Δ𝜉), 𝜉̄ + Δ𝜉)
| |
= ℎ(𝑥(𝜉) ̄ + 𝜕ℎ | 𝜕𝑥 Δ𝜉 + 𝜕ℎ | Δ𝜉 + (Δ𝜉 2 )
̄ , 𝜉) (37)
For solving (36), three strategies are used in this contribution: two 𝜕𝑥 ||𝜉̄ 𝜕𝜉 𝜕𝜉 ||𝜉̄
state-of-the art optimization schemes (genetic algorithm and an interior- ( )
| |
= ℎ(𝑥(𝜉) ̄ + 𝜕ℎ | 𝑆 + 𝜕ℎ | Δ𝜉 + (Δ𝜉 2 ).
̄ , 𝜉) (38)
point algorithm), as well as a gradient-based optimization scheme di- |
𝜕𝑥 |𝜉̄ |
𝜕𝜉 |𝜉̄
rectly relying on discrete-value sets of the parameters. To achieve a ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
better comparability, the results of the interior-point algorithm for the =∶ΛTℎ
cylinder diameters are rounded and then compared to the genetic algo-
rithm and discrete-value optimization strategy. where ΛTℎ is the gradient of the respective boundary conditions.
Now, we can employ an iterative optimization strategy using the
4.2.1. Genetic algorithm previously derived relations for calculating and approximating the out-
In order to obtain a globally comparable solution to the optimiza- comes of parameter variations. Starting with an initial parameter con-
tion problem (36), a genetic algorithm is used for optimization [21]. figuration 𝜉̄0 , in each iteration step 𝜈, we
In literature, applications of genetic algorithms to the optimization of 1. Perform a simulation of the model and its sensitivities with a current
pneumatic drive systems can be found – e.g. in [9] yielding suitable re- parameter configuration 𝜉̄𝜈 . By this, we obtain the state trajectories
sults. However, it is well known and later confirmed here that genetic for xi,𝜈 , together with the corresponding boundary conditions hi,𝜈
algorithms require excessive function evaluations and thus large com- and their gradients ΛTℎ,𝜈 around the current configuration 𝜉̄𝜈 .
putation times. Therefore, we consider gradient-based strategies for the 2. Build a local catalog of parameter variations. This is done by com-
optimization of pneumatic drive systems in more detail in the following. bining parameter sets around the current configuration 𝜉̄𝜈 in a matrix
P𝜈 . For this, we replace single or multiple entries of 𝜉̄𝜈 with different
4.2.2. Interior-Point optimization values of the discretized parameter set. In order to suffice for the lo-
The very simplest way to realize a gradient-based optimiza- cality of the sensitivities, we allow steps in supply pressure of up to
tion scheme for the considered problem is the utilization of Mat- 300 mbar, of up to 0.75 /mm3 s · Pa in conductances, and one cylinder
lab/fmincon with an interior-point optimization strategy [22]. In size. So, each column of P𝜈 corresponds to a parameter variation Δ𝜉
order to reduce the necessary number of simulations, we supply the gra- around the current parameter configuration 𝜉̄𝜈 .
dients for cost J and holding forces analytically by (29a), (29b), (31a), 3. Then, we estimate the outcomes of all parameter variations by using
and (31b). The gradients for travel time and impact velocity are then the analytic expressions and (38) with a cheap matrix-vector multi-
calculated by finite differences. plication for
( ) [ ]
𝜕ℎ 𝜕ℎ ||
4.2.3. Discrete optimization scheme ΛTℎ,𝜈 𝑃𝜈 = 𝑆+ | 𝜉̄𝜈 +Δ1 𝜉, … , 𝜉̄𝜈 +Δ𝑁 𝜉 . (39)
𝜕𝑥 𝜕𝜉 || ̄
The interior-point optimization will already lower the number of 𝜉𝜈
iterations and function evaluations compared to gradient-free meth- 4. For the next iteration, we choose the best predicted feasible con-
ods like the considered genetic algorithm dramatically. However, as figuration with respect to the cost. If there is no predicted feasible
mentioned before in Section 4.1, at least the cylinder diameter is a configuration around the current setup 𝜉̄𝜈 , we choose the configura-
value from a discrete set. When also considering drives without the tion with the least predicted constraint violation,
use of extra throttles, usually the sonic conductances as well are com-
ing from a discrete set. So in practice, the results of the two previ- until the configuration stays the same in the next iteration.
ously sketched optimization schemes have to be rounded in order to Remark (Connection between sensitivity equation and adjoint equa-
obtain a realizable result. To overcome this issue, we next propose a tion). Depending on the number of design parameters n𝜉 , the

174
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

computational effort to calculate the sensitivity might increase signifi- Regarding the number of performed simulations and the obtained re-
cantly. In literature, a connection between the sensitivity and the adjoint sults, we can conclude that gradient based optimization for pneumatic
equation can be found [23], reducing the computational effort. The uti- drive systems is a valid approach, despite of the system nonlinearities.
lization of the adjoint system basically leads to the problem of solving In case of the interior-point and genetic algorithm results it should be
𝑛x ODEs backwards in time per asked sensitivity at a defined time in- pointed out that the simulation of the sensitivity ODEs was not neces-
stance. So, relying on the number of state variables, design parameters, sary. This results in less computational effort per simulation compared
and boundary conditions, the evaluation of the adjoint system can lower to the results for the discrete scheme, where the system ODE (18) and
the computation effort compared to solving the 𝑛x × 𝑛𝜉 sensitivity ODEs. the sensitivity ODEs (20) need to be simulated simultaneously. Still,
The solution of the adjoint system does not provide the time solution of when working directly with discretized variables and available cylin-
the sensitivities, however allows a simple calculation of the sensitivities der sizes, the computational effort can be lowered further compared to
at time 𝑡f . Especially to maintain a better interpretability of the results, standard optimization strategies. This can be an advantage when imple-
we adhere to the use of the sensitivity in this paper. menting the optimization scheme for a user application from a manu-
facturer point of view. Here, calculation times need to be reduced as far
5. Optimization results as possible. Besides the parameter results, Table 1 also gives the com-
putation times for our example obtained by calculations in MATLAB on
5.1. Application example a Intel Xeon Gold CPU with 2.1 GHz. Expectably, the genetic algorithm
exceeds the computation time of gradient based optimization schemes
The application example for this contribution was already intro- (293.52 min for the genetic algorithm, 10.61 min for the interior-point
duced in Fig. 1 and consists of three different transitions. In Fig. 10, a de- algorithm, and 34.3 s for the proposed discrete scheme). The slight mis-
tailed view on the process is given: First, a load mass of 2 kg is displaced match between measured and simulated compressed air consumption of
horizontally by 500 mm [Fig. 10(a) → Fig. 10(b)]. Then, it is lifted around 2.5% originates from the thermal measurement principle of the
by 200 mm [Fig. 10(b) → Fig. 10(c)], and finally ejected by 100 mm used Festo-SFAB mass flow sensors.
[Fig. 10(c) → Fig. 10(d)]. All three transitions are performed as double In Fig. 11, measurements for the initial and optimal parameter
strokes, meaning that the cylinders retract to their original position af- configurations from the presented discrete optimization scheme are
terward. Optimizing the supply pressure 𝑝h , the cylinder diameters di , given. Depicted are the normalized positions, the chamber pressures
as well as the sonic conductances for inflating and venting the cylinder of the driving chambers, as well as the compressed air consumption
chambers leads to 6𝑛c = 18 optimization variables for the considered for both cases. By reducing the cylinder size and the supply pressure,
sizing problem. For the experiments conducted here, a Festo Motion the overall compressed air consumption was lowered by 2.44 standard
Terminal Valve system is used, being able to set defined conductances liters or 55.5%. It is notable that the initial position trajectory of
for the different branches. The chamber pressures, piston positions, and the first cylinder (black, dashed) behaves slower than the one of the
mass flows are measured during our experiments. For the initial config- optimized setup (black, solid), while the lifting task (blue lines, around
uration of cylinder diameters and supply pressure, a conservative static 5 s) is accelerated. This is a hint that the original 32 mm diameter
choice was made – ensuring the fulfillment of the motion task. of cylinder 1 was over-sized, resulting in a slower behavior, as also
more volume needs to be filled with compressed air. The predicted
5.2. Results and measurements results from simulations and optimization again emphasize the well-
suited model, while the measured amount of savings underlines the
The previously introduced optimization schemes are now applied to importance of already small changes of the system setup for pneumatic
the exemplary setup from Fig. 10 as previously explained. In Table 1, the drives.
optimization-related results – except the sonic conductances – are de- Fig. 12 illustrates the necessary number of iterations for the scheme
picted. For clarity of presentation, those can be found in Table A.1 in the presented in Section 4.2.3. Over 800 optimizations were performed with
Appendix. As the interior-point optimization delivers cylinder sizes not initial conditions distributed inside and at the boundaries of the allowed
matching available diameters, we round those results (conservatively) parameter values (compare Section 4.1). While the number of iterations
up and obtain the resulting maximum constraint violation in the tran- stays below 40 even for initial values far from the optimal configuration,
sition time. It is noticeable that all three strategies – genetic algorithm, the obvious trend of increased iterations for an increased initial distance
interior-point and the discrete scheme – deliver comparable results. from the optimization result can be seen.

Fig. 10. Detailed view of the considered process: The load mass is first transitioned by 500 mm [(a) → (b)], then lifted by 200 mm [(b) → (c)], and finally displaced
by 100 mm [(c) → (d)].

175
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

Table 1
Comparison of the optimization results.

Algorithm 𝑝h (bar ) d1 (mm) d2 (mm) d3 (mm) J (s.l.) # sim. comp. time constr. violation

Genetic Algorithm 3.06 25 20 12 1.87 21850 293.52 min 0%

Interior-Point 3.00 22.35 17.44 12.21 1.50 780 10.61 min 0%
Interior-Point (rounded) 3.00 25 20 16 1.88 9.24%
Discrete Scheme 3.3 25 16 12 1.91 15 0.57 min 0%
Measured initial setup 4.8 32 20 12 4.40
Measured optimal setup 3.3 25 16 12 1.96

Fig. 11. Measurement results for the optimization of

the considered setup: depicted are the normalized posi-
tions of cylinder 1 (500 mm, black), cylinder 2 (200 mm,
vertical, blue), and cylinder 3 (100 mm, yellow), both
before (dashed) and after optimization (solid lines). Sec-
ondly, the corresponding chamber pressures of the driv-
ing chambers are depicted. Third, the compressed air
consumption in standard liters is shown for the ini-
tial (gray) and resulting (dark blue) configuration. The
compressed air consumption is reduced from 4.40 s.l. to
1.96 s.l. by adaption of cylinder diameters, supply pres-
sure and sonic conductances. (For interpretation of the
references to colour in this figure legend, the reader is
referred to the web version of this article.)

Acknowledgment

This work is part of EnAP-project (http://www.enap-projekt.de),

supported by the German Federal Ministry for Economic Affairs and
Energy under Grant 03ET1385B: User oriented use of energy-efficient
drive technology in production. (original title, in German: Anwenderor-
ientierter Einsatz energieeffizienter Antriebstechnik in der Produktion).

Conflict of interest

None.

Fig. 12. Number of necessary iterations of the presented scheme from

Appendix A. Results for the sonic conductances
Section 4.2.3 for different initial values together with their mean value.

In order to shorten the results in Table 1, the sonic conductances of

6. Conclusions optimization and implementation are given here, in Table A.1.

Based on a validated model for pneumatic cylinder drives, a detailed

system analysis was performed in this paper. The analysis highlighted Table A.1
the relevance of the parameters such as cylinder diameter, supply pres- Sonic conductances from optimization and measurements. All sonic
sure, and sonic conductances on the dynamic behavior. Based on mod- conductances are given in 10 mm3 /s · Pa.
eling and analysis, optimization strategies for pneumatic drive systems
Gen. Algo. Int.-Point Discrete meas. init. meas. opt.
were presented and applied to a three-drive system. As the results from
gradient-based optimization strategies are in no way inferior compared 𝐶A,1
E
0.8436 0.6409 1.0000 0.9999 1.0000
to the results of a genetic algorithm, we can conclude that gradient 𝐶A,1
R
0.6473 0.6003 0.6550 0.6549 0.6550
𝐶B,1
E
0.7861 0.6017 0.6000 0.5995 0.6000
based strategies are suitable for the sizing pneumatic drives. The com-
𝐶B,1
R
0.6343 0.6771 0.6000 0.6016 0.6000
putational effort can be reduced further by directly optimizing with a 𝐶A,2
E
0.6354 0.8174 1.0000 0.4664 1.0000
discrete set of parameters, as cylinder sizes are typically available from 𝐶A,2
R
0.8155 0.9253 1.0000 0.3996 1.0000
a discrete set. For this purpose, an optimization heuristic was presented. 𝐶B,2
E
0.6699 0.8489 0.6400 0.4356 0.6400
The optimization and measurement results underline the validity of the 𝐶B,2
R
0.7266 0.8211 0.6000 0.4500 0.6000
chosen approaches as well as the importance of a proper system sizing. 𝐶A,3
E
0.6992 0.8205 0.6000 0.5250 0.6000
𝐶A,3
R
0.6705 0.7755 1.0000 0.4440 1.0000
Future work will include the application towards energy saving mea-
𝐶B,3
E
0.8096 0.7875 0.6000 0.5100 0.6000
sures and to the production plants of the industrial partners of the fund-
𝐶B,3
R
0.8003 0.8913 0.6000 0.4640 0.6000
ing EnAP research project.

176
A. Raisch and O. Sawodny Mechatronics 59 (2019) 168–177

Table A.2
Overview of all symbols and variables.

Symbol Description Symbol Description

p0 ambient pressure 𝜈 index for the iteration step of the optimization

𝛼 angle of orientation m mass
𝜇 approximation constant for the piston areas 𝜉̄ nominal or specific value of design parameters
𝑚̇ compressed air mass flow 𝑠̈ piston acceleration
𝛾 constant for contact force model of FT A piston area
𝛽 constant for contact force model of FT s piston position
𝛿 constant for contact force model of FT 𝑠̇ piston velocity
J cost function n polytropic index
fc Coulomb friction coefficient 𝜅 polytropic index for isentropic process
b critical pressure ratio p pressure
d cylinder diameter f right hand side of the system ODE
l cylinder length S sensitivity
VD dead volume C sonic conductance
𝑣̄ ∗ demanded (maximal) impact velocity R Specific Gas Constant
𝑇̄ demanded (maximal) travel time 𝜌0 Standard density of air
𝐹̄H demanded (minimal) holding force x state vector
𝜉 design parameters A subscript for chamber A
pds downstream pressure B subscript for chamber B
send end position of cylinder (send ∈ {0, l}) E superscript for Extension
tf end time of simulation R superscript for Retraction
Ψ flow function ph supply pressure
FF friction force Φ task parameters
g gravity T Temperature
FG gravity force FT terminal force at end stop of the cylinder
FH holding force T travel time
v∗ impact velocity pus upstream pressure
i index for cylinder u valve position
j index for cylinder chambers A, B Δ𝜉 variation of design parameters
r index for design parameters fv viscous friction coefficient
k index for Extension and Retraction V volume

References [17] Beater P. Pneumatic Drives - System Design, Modelling and Control. Springer Berlin
Heidelberg; 2007.
[1] Radgen P, Blaustein E, editors. Compressed air systems in the european union - [18] ISO. Pneumatic fluid power - determination of flow-rate characteristics of compo-
energy emissions, saving potential and policy actions. Fraunhofer ISI; 2001. nents using compressible fluids. ISO 6538 2013.
[2] EnEffAH - energy efficiency in production in the drive and handling technology field. [19] Sanville FE. A new method of specifying the flow capacity of pneumatic fluid
Project report; 2012. http://www.eneffah.de/EnEffAH_Broschuere_engl.pdf. power valves. BHRA 2nd International Fluid Power Symposium; 1971. D37–
[3] Ohligschläger O. Pneumatic cylinder drives - thermodynamic fundamentals and dig- D49.
ital simulation. RWTH Aachen; 1990. (in german, original title: Pneumatische Zylin- [20] Dickinson RP, Gelinas RJ. Sensitivity analysis of ordinary differential
derantriebe - thermodynamische Grundlagen und digitale Simulation). equation systems a direct method. J Comput Phys 1976;21(2):123–43.
[4] Glück T, Kemmetmüller W, Kugi A. Trajectory optimization for soft landing of fast- doi:10.1016/0021-9991(76)90007-3.
switching electromagnetic valves. In: Proceedings of the 18th IFAC World Congress. [21] Deep K, Singh KP, Kansal M, Mohan C. A real coded genetic algorithm for solving in-
Milano, Italia; 2011. p. 11532–7. doi:10.3182/20110828-6-IT-1002.01822. teger and mixed integer optimization problems. Appl Math Comput 2009;212:505–
[5] Krichel SV, Sawodny O. Non-linear friction modelling and simulation of long 18. doi:10.1016/j.amc.2009.02.044.
pneumatic transmission lines. Math Comput Model Dyn Syst 2014;20(1):23–44. [22] Nocedal J, Wright S. Numerical Optimization. Springer; 1999.
doi:10.1080/13873954.2013.811268. [23] Cao Y, Li S, Petzold L, Serban R. Adjoint sensitivity analysis for differential-algebraic
[6] Turkseven M, Ueda J. An asymptotically stable pressure observer based on load equations: the adjoint DAE system and its numerical solution. SIAM J Sci Comput
and displacement sensing for pneumatic actuators with long transmission lines. 2003;24(3):1076–89.
IEEE/ASME Trans Mechatron 2017;22(2):681–92.
[7] Hepke J, Weber J. Energy saving measures on pneumatic drive systems. In: The 13th
Adrian Raisch received the B.Sc. and M.Sc. degrees in engi-
Scandinavian International Conference on Fluid Power, SICFP2013; 2013. p. 475–
neering cybernetics from the University of Stuttgart, Stuttgart,
83. doi:10.3384/ecp1392a47.
Germany, in 2013 and 2015, respectively. He was with the
[8] Doll M, Neumann R, Sawodny O. Dimensioning of pneumatic cylinders for motion
Cymer Center for Control Systems and Dynamics, UC San
tasks. Int J Fluid Power 2015;16(1):11–24. doi:10.1080/14399776.2015.1012437.
Diego, USA in 2015 under the supervision of M. Krstić. Since
[9] Harris P, Nolan S, O’Donnell GE. Energy optimisation of pneumatic actuator systems
2015 he has been a Research Assistant at the Institute for Sys-
in manufacturing. J Clean Prod 2014;72:35–45. doi:10.1016/j.jclepro.2014.03.011.
tem Dynamics, University of Stuttgart, Germany. His current
[10] Ke J, Wang J, Jia N, Yang L, Wu QH. Energy efficiency analysis and optimal control
research interests include modeling and control of pneumatic
of servo pneumatic cylinders. In: IEEE Conference on Control Applications; 2005.
and electrical drive systems, with a focus on energy efficiency.
p. 541–6. doi:10.1109/CCA.2005.1507182.
[11] Doll M, Neumann R, Sawodny O. Energy efficient use of compressed air in pneumatic
drive systems for motion tasks. In: International Conference on Fluid Power and
Mechatronics; 2011. p. 340–5. doi:10.1109/FPM.2011.6045785.
[12] Pfeffer A, Glück T, Kugi A. Soft landing and disturbance rejection for pneumatic
drives with partial position information. In: Proceedings of the 7th IFAC Symposium Oliver Sawodny received the Dipl.-Ing. degree in electri-
on Mechatronic Systems & 15th Mechatronics Forum International Conference, vol. cal engineering from the University of Karlsruhe, Karlsruhe,
49. Loughborough, UK; 2016. p. 559–66. doi:10.1016/j.ifacol.2016.10.661. Germany, in 1991, and the Ph.D. degree from the Ulm Uni-
[13] Raisch A, Hülsmann S, Sawodny O. Saving energy by predictive supply air shutoff versity, Ulm, Germany, in 1996. In 2002, he became a Full
for pneumatic drives. In: European Control Conference (ECC); 2018. p. 965–70. Professor with the Technical University of Ilmenau, Ilmenau,
[14] Wang J, Gordon T. Energy-efficient tracking control of pneumatic cylinders. In: 2011 Germany. Since 2005, he has been the Director of the Insti-
50th IEEE Conference on Decision and Control and European Control Conference; tute for System Dynamics, University of Stuttgart, Stuttgart,
2011. p. 7956–61. doi:10.1109/CDC.2011.6160684. Germany. His current research interests include methods of
[15] Hildebrandt A, Neumann R, Sawodny O. Optimal system design of SISO- differential geometry, trajectory generation, and applications
servopneumatic positioning drives. IEEE Trans Control Syst Technol 2010;18(1):35– to mechatronic systems.
44. doi:10.1109/TCST.2008.2009879.
[16] Hunt K, Crossley E. Coefficient of restitution interpreted as damping in vibroimpact.
J Appl Mech 1975.

177