PRELIMINARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED FLUID COUPLING FOR TRAINS

Abstract

In order to improve train availability and workplace safety standards, the rail industry is keen to modernise maintenance of trains through increased use of Robotic Autonomous Systems (RAS). Our research aims to address the mechanical challenges of automated fluid coupling in future applications of train-fluid servicing. Depending on the intricacy of the servicing RAS, a degree of misalignment will always exist between the robot end-effector and train fluid ports. Compliant end-effectors can generate flexing motions that facilitate misaligned insertions. Present work focuses on understanding the role of passive compliance within the end-effector of our demonstrator train-fluid servicing robot. Physical experiments were performed and using Design of Experiments we identify the effect of end-effector compliance parameters on misaligned insertions. Results show that maximum insertion force and work done increase exponentially with increasing misalignment. Certain arrangements of compliance parameters can significantly improve the coupling performance under misalignments. Nonetheless, forces observed are still too large and our research will continue to develop compliant end-effectors for better automated coupling that will reduce RAS force requirements.

Full text

Proceedings of TMCE 2020, 11-15 May, 2020, Dublin, Ireland, edited by I.Horváth and G.Keenaghan
 Organizing Committee of TMCE 2020, ISBN/EAN: 978-94-6384-131-3
205
PRELIMINARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED
FLUID COUPLING FOR TRAINS
Kourosh Eshraghi
Department of Mechanical and Aerospace Engineering
Brunel University London
Kourosh.Eshraghi@Brunel.ac.uk
Pingfei Jiang
Department of Mechanical and Aerospace Engineering
Brunel University London
Pingfei.Jiang@Brunel.ac.uk
Daniele Suraci
Department of Mechanical and Aerospace Engineering
Brunel University London
d.t.suraci@gmail.com>
Mark Atherton
Department of Mechanical and Aerospace Engineering
Brunel University London
Mark.Atherton@Brunel.ac.uk
ABSTRACT
In order to improve train availability and workplace
safety standards, the rail industry is keen to modernise
maintenance of trains through increased use of
Robotic Autonomous Systems (RAS). Our research
aims to address the mechanical challenges of
automated fluid coupling in future applications of
train-fluid servicing. Depending on the intricacy of the
servicing RAS, a degree of misalignment will always
exist between the robot end-effector and train fluid
ports. Compliant end-effectors can generate flexing
motions that facilitate misaligned insertions. Present
work focuses on understanding the role of passive
compliance within the end-effector of our
demonstrator train-fluid servicing robot. Physical
experiments were performed and using Design of
Experiments we identify the effect of end-effector
compliance parameters on misaligned insertions.
Results show that maximum insertion force and work
done increase exponentially with increasing
misalignment. Certain arrangements of compliance
parameters can significantly improve the coupling
performance under misalignments. Nonetheless,
forces observed are still too large and our research
will continue to develop compliant end-effectors for
better automated coupling that will reduce RAS force
requirements.
KEYWORDS
Compliant end-effector, passive compliance, design
of experiments, automated fluid coupling, train
maintenance
1. INTRODUCTION
1.1. Rail expansion and potential for
service automation
It is well documented that UK rail traffic will
considerably increase by 2047, requiring a
commensurate increase in national passenger fleet of
between 5,500 and 12,000 vehicles [1]. This increase
will produce a comparative challenge for maintenance
depots in terms of increasing their service capacity to
keep trains available and reliable. There will also be
an increasing demand for accurate service data and
asset condition monitoring, vital for modernising
maintenance, which will be challenging for manual-
based servicing. Not only capacity limits of current
manual labour will be stretched, but also the
desirability of subjecting humans to such tasks in the
21st century will continue to be scrutinised in terms of
working conditions and safety aspects.
It is generally accepted that autonomous servicing
could make a positive contribution to meeting the
above demands as supported by, for example, an

206
Kourosh Eshraghi, Pingfei Jiang, Daniele Suraci and Mark Atherton
economic and technical feasibility study of a robotic
autonomous system for train-fluid servicing [2].
However, it will be technically challenging for an
autonomous system to match manual-based servicing
in achieving effective mechanical connections
common to many train maintenance tasks, which are
performed under uncertain conditions. Meanwhile,
robotic inspection and evaluation of unfamiliar
conditions or damage is another challenge.
We have built a research demonstrator robot in order
to physically investigate the servicing of train-fluids
and to specifically focus on the end-effector design for
effective and reliable mechanical connection of fluid
couplings. This initial “CyberFluids” system (see
Figure 1) includes the servicing of fuel, Controlled
Emissions Toilets (CET) and wheel sand as pragmatic
examples, which has been well received at a
demonstration event to the rail industry in March 2019
[3]. Treating this fluid coupling application as akin to
a robot assembly task opens up the work to a large
body of research in which robot compliance is key to
successful robotic manipulations [4].
1.2. Robot compliance
In robotics, compliance (the inverse of stiffness) is
defined as the relationship between the motion and
forces generated by a robot manipulator and an
assembled part at their point of contact. Compliance
can commonly be in the form of a spring-like stiffness,
a generalised damper or a mechanical impedance.
Active compliance is achieved by controlling robot
actuators and servo-motions. Passive compliance is
defined as intrinsic structural deflections such as
flexibility of the robot base, limbs, joint transmissions
and compliant end-effector [5]. Active compliance
reduces reliance on physical elements and
corresponding complexity, also aims at ensuring
safety for humans. However, it also presents
disadvantages such as power losses, relatively slow
dynamic responses and reliance on sensors and control
systems. Therefore, passive compliance remains an
important feature of most robotic systems and a
logical consideration for any new application before
active compliance is developed on top.
Compliance of the end-effector is the primary concern
Figure 1 CyberFluids train fluid servicing robot

PRELIMNARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED FLUID COUPLING
207
in robotic assembly/disassembly where the total
positioning tolerance is greater than the assembling
tolerance for a ‘peg-in-hole’ insertion [6]. In such
cases, the aim is to avoid wedging or jamming of parts
through employing compliance at the end-effector so
that motion due to misalignment forces are
accommodated. In some peg-in-hole assembly
applications, a Remote Centre Compliance (RCC) [7]
device projects the compliance centre to a desirable
location below or at the point of contact between the
assembled parts. An RCC-equipped assembly gripper
will pivot the inserting part in a direction that
improves alignment as parts contact during the
insertion. A Variable Remote Centre Compliance
(VRCC) device adapts to various peg lengths by
adjusting its RCC projection point [8]. Hence, a single
device can proficiently insert various pegs. The idea
of RCC is well established and there are other
applications such as minimally intrusive surgery. A
recent study [9] investigated the design and analysis
of a passive RCC device using physical and nonlinear
finite element experiments. A projection accuracy of
+/- 0.015 mm was achieved for primary stage motions
of around +/- 4mm.
Active insertion systems, based on force control
algorithms, mimic an ideal RCC mechanism but more
recent developments include machine learning, a
mixture of sensors and vision systems [10] to solve the
peg-in-hole issue. Although the literature is rich, there
seems to be a lack of studies that investigate large
motion passive devices for the peg-in-hole insertion.
Most, if not all, designs incorporate RCC and the
inherent principle of instantaneous centre of motion to
project the compliance centre. This concept remains
valid where motions are small and hence, it is only
sufficient for precision assembly applications. For
robots outside of controlled environments, e.g. train
maintenance, the challenge is to accommodate larger
positional uncertainties that accumulate in
unstructured environments. In technical terms, a
compliant design is considered optimal if the insertion
forces for the misalignment range is minimal. This is
a motivation to investigate compliant end-effectors for
our CyberFluids train servicing robot.
In this paper, the effect of end-effector compliance on
coupling performance is investigated through physical
experiments and a resulting second-order regression
model. In section 2, we present the end-effector and
briefly analyse its performance using Design of
Experiments (DoE) where misalignments constitute
the uncontrollable factors. Misalignments are
representative of practical and inevitable conditions of
train position error or robot positional inaccuracy that
will be common in future applications of RAS to train
servicing. These initial experimental results are
presented in section 3 and will serve as a basis for
future comparison of end-effectors to be developed.
In section 4 the results are discussed, and some
recommendations are given to conclude the work in
section 5.
2. METHODOLOGY
2.1. CyberFluids’ end-effectors
CyberFluids is a Cartesian, track-based robot that runs
along a scaled-down mock-up train carriage that has 2
fluid ports for investigating autonomous servicing.
The track alongside the train carriage shown in Figure
1 is the robot X-axis. The robot has 7 Degrees of
Freedom (DoF) and provides 5-axis positioning for 3
end-effectors mounted on insertion arms (Z-axis).
Two of the Z-axis arms accommodate fluid couplers
and the third is for gripping the relevant dust caps.
The nominal size of each train port (and cap)
corresponds to the typical 2-inch fuel port and 3-inch
CET port. The cap gripper has an adjustable jaw to
accommodate both cap sizes. On the train side, the 2-
inch port is fixed to a manual 5-axis, non-back-
drivable positioning stage that can be used to
deterministically misalign the ports.
In current (manual) train-fluid servicing operations,
many different types of fluid couplers are used to
completely service the train. For each type of port
coupler different combinations of linear and rotary
motions are required to make the coupling. Therefore
there is a need to adapt and standardise train fluid ports
for automated fluid servicing. Dixon Ez-link cam and
groove couplers were selected for this purpose,
requiring only a linear insertion motion to reliably
make a secure and sealed connection. This linear
motion will also be less demanding of the robot and
end-effector while making the automated fluid
servicing faster. The CyberFluids end-effector was
designed to accommodate the Dixon coupler and
actuate its cam-locks. The 2-inch and 3-inch end-
effectors are identical in design with 2 pneumatic
actuators and a compliant interface with the robot arm.
We have selected the 2-inch version for this study. In
this section, the end-effector design is evaluated in
order to identify design parameters for the DoE.
The end-effector has passive compliance facilitated by
spring elements, as depicted in Figure 2. It has 3
shoulder bolts encapsulated by springs, the threaded
part of the bolt is fixed to the coupler flange while the

208
Kourosh Eshraghi, Pingfei Jiang, Daniele Suraci and Mark Atherton
bolts are free to pivot and slide within corresponding
clearance holes located in the arm flange. In this
arrangement when all bolts slide simultaneously the
motion is linear in the Z-direction. The clearance
holes effectively act as spherical joints that have
corresponding angular motions, which are amplified
along the bolt length. The bolts are equally spaced
around a pitch circle (radius, PR) that coincides with
the centre of the coupler. Spring pre-compression
ensures that the end-effector returns to its original
position after a misaligned insertion. Maximum linear
sliding in Z is determined by the compressed length of
the springs. The coupler and its flange rotate and slide
relative to the arm flange. The maximum rotation
occurs when the bolts have two-point contact in the
hole. Equation (1) is based on simple geometry and
can be solved nonlinearly to estimate this maximum
angle.
𝑏𝑠𝑖𝑛(𝜗)+𝑙𝑐𝑜𝑠(𝜗)= 𝑡 (1)
Where b is hole diameter, l is hole length, t is bolt diameter
and θ is the maximum angle of rotation.
The coupler has 3 DoF (θYZ, θXZ and Z). As shown in
Figure 2, the X and Y linear motions of the coupler are
coupled to the rotations in yaw (θXZ) and pitch (θYZ).
Hole clearance can encourage small, non-elastic
motions in the remaining directions (X Y and θXY)
however, keeping the clearance to a minimum, these
motions can be neglected as relatively small. The
maximum range of linear motion is not symmetrical
either side of the X-axis i.e. in pitch motions. As
shown in Figure 2a and b, this is because bolts and
springs either side of the X-axis have different
distances to the coupler centre. This means that when
the plate pivots the amplification effect will be
different depending on the direction of motion. For
similar reasons, compliance in the pitch-axis is also
not symmetrical.
2.2. Controllable factors: compliance
We are interested in understanding the effects of
spring stiffness, K, pitch radius of the bolt holes, PR
and the orientation of the set of holes, O. As discussed
in the previous section, clearance will remain a
constant and bolt length is not considered (or distance
between the arm and coupler flanges) in order to
reduce the number of variables and experiments.
Table 1 lists the experimental parameters and Figure 3
shows the modified experimental end-effector that can
accommodate up to 5 adjustable levels for hole
orientation, O, and pitch radius, PR. In order to easily
adjust the spring stiffness, K, it was decided to use
Belleville spring washers (see Figure 4). A number of
these spring washers can be stacked in parallel or
series to achieve a large range of stiffness,
Figure 2 Arrangement of CyberFluids end-effectors

PRELIMNARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED FLUID COUPLING
209
deformation or load characteristics [11]. The selection
of Bellville springs is not only constrained by the
required range of stiffness or deformation, but also by
bolt diameter. If the clearance between the washer and
bolt is too large, even when the spring is compressed,
washers can slide in the radial direction. This is
undesirable as washers will not make contact at
consistent points and this will cause an indeterminate
change in spring parameters that will induce
experimental error. It is also crucial to prevent the
washers from jamming in the screw thread. Thus bolt
length, L, is selected to accommodate the longest
spring washer stack. For spring arrangements with a
lower free length, standard spacer washers are
included in the stack to fill the remainder of the bolt
length. All springs arrangements are pre-compressed
to 15% of the total stack deformation (lower bound
spring operating range recommended in DIN2093).
As shown in Figure 5, the selected spring has a
nonlinear force-displacement relationship thus:

 ≠ 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝐾 (2)
In order to derive a single parameter that serves as the
stiffness constant (design parameter, K), the force-
displacement relationship shown in Figure 5 is
linearised. Since the relationship between the number
of washers and deformation/load is linear, the
regression fit is independent of the number of washers
stacked. With R2=0.898 and P-Value= 0.00404, the
regression model has a good fit. Hence linearisation
is a simple and reliable method of comparing the
stiffness of various washer stack arrangements.
Equation 3 is used to identify the number of washers
(in series) that will deliver the required linear stiffness.
Table 1 Experimental parameters
Figure 3 Experimental end-effector
Figure 4 Parameters of the selected Belleville
spring washer [11]

210
Kourosh Eshraghi, Pingfei Jiang, Daniele Suraci and Mark Atherton
𝑁 = 𝐿𝑖𝑛𝑒𝑎𝑟 𝑆𝑡𝑖𝑓𝑓𝑛𝑒𝑠𝑠 𝑜𝑓 1 𝑤𝑎𝑠ℎ𝑒𝑟
𝑅𝑒𝑞𝑢𝑖𝑟𝑒𝑑 𝐿𝑖𝑛𝑒𝑎𝑟 𝑆𝑡𝑖𝑓𝑓𝑛𝑒𝑠𝑠  (3)
The number of washers must obviously be a discrete
value and thus, due to rounding up/down, there are
errors in achieving the exact required linear stiffness.
If a large number of washers are stacked in series this
error becomes very small. Using the selected spring,
for a range of 3.2 to 6 N/mm (91 to 49 spring washers)
maximum error in the stiffness is 0.064%
2.3. Noise factors: misalignments
For an XYZ (3 DoF) Cartesian RAS conceptualised in
[2], robot X and Y-axis misalignments lead to poor
coupling. Axial misalignments in Z (insertion
direction) only contribute to the coupling seal hence
neglected in present work. For these experiments train
fluid ports are aligned parallel to the robot Z-axis and
the CyberFluids robots will only use XYZ motion to
make the fit. Initial experiments determined that ports
do not couple when X and Y axes are misaligned more
than 5 mm. Hence, the experimental range of
misalignments is selected accordingly.
Misalignments are measured using Vernier callipers
from a fixed datum on the fluid port positioning stage.
2.4. Performance measurement: readings
A perfectly aligned fluid coupling is expected to
couple with minimal force; a high insertion force will
indicate high friction and/or physical clash that will
occur due to misalignment between the coupler and
fluid port. Therefore, considering the complete
insertion cycle, we can use the energy quantity ‘work
done’ by the motor, as a scalar measure of coupling
performance. Insertion force is also monitored, as this
is important for sizing actuators, robot structure and
preventing damage to the robot or train parts.
Robot servo-drives can monitor motor current and
position, which is used to determine insertion force
and linear position of the end-effector arm. The
relationship between motor current and torque is
linear and defined by a motor torque constant
specified by the manufacturer. In order to obtain work
done, torque must be converted to force and integrated
over the linear distance travelled.
Motor Torque, T= k∗I (4)
Gearbox,T= T∗5 (5)
Insertion Force,F= T/r (6)
Work , W = ∫𝐹𝑑 (7)
Where 𝑘 is the motor torque constant, Irms is the
root mean square of the ‘torque generating’ current
and r is the radius of the pulley drive in the insertion
arm.
The servo-drive is capable of recording 200 samples
for motor position and current. The robot insertion
speed was set to a nominal value of 25mm/s, and a
sampling rate of 20ms was used in order to capture the
entire event of coupling with sufficient precision.
2.5. Design of Experiments (DoE)
A DoE [12] approach was employed in order to
maximise the chances of meaningful experimental
results. DoE methods introduced in many textbooks
place emphasis on factors with two levels, often called
2k-p design, where k is the number of factors and p is
the degree of fractionation. DoE methods such as full
factorial and fractional factorial design are widely
applied in 2-level factor DoE. Despite the advantages
of having only 2-level factors such as reduced
experiment size and simple analysis of main effects
and interactions [13], it is inadequate for predicting
precise and non-linear behaviour of the system output
due to factor changes. Response surface methodology
(RSM) can be seen as a branch of DoE with the
purpose of fitting regression models and optimising
processes and systems [14]. Central Composite
Design (CCD) was adopted in this study due to its
wide application [14]. When allocating design points
three types of CCD can be applied, namely
circumscribed (CCCD), inscribed (ICCD) and faced
(FCCD). CCCD tends to create new extreme limits for
factors, indicated by the four axial points outside
design space. ICCD can be seen as a scaled-down
version of CCCD with axial points created inside the
y = 5.9619x + 50.709
y = 3.2102x + 50.709
0
40
80
120
160
200
0 20 40
Force (N)
Spring Deformation (mm)
49
Washers
Figure 5 Force-displacement relationship of the selected
Belleville washers

PRELIMNARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED FLUID COUPLING
211
design space. For CCCD larger factor limits are
physically impossible to achieve with the current
design. For ICCD no combination of factors at
extreme levels are investigated, e.g. largest
misalignments in X and Y at the same time. Therefore
FCCD was adopted in this study which investigates
the influence of robot end-effector design factors on
the resultant energy consumption to perform an
effective coupling. An FCCD for five factors of K,
PR, O, XM and YM requires 36 experiments with each
factor at 3 levels, including the extremes and
midpoints of the experimental range (see Table 1).
Typically, system responses are represented by
regression functions with appropriate empirical
models that allow prediction with known factors.
These regression functions can exist in various
formats such as first-order or second-order polynomial
models, describing linear and quadratic behaviour of
system responses respectively. Equation 8 is a general
expression of a second-order regression model.
R
=
β

+

β

x





+

β

x






+


β









x

x

+
ε
R represents system response, n stands for the number
of factors, xi stands for each independent factor, β are
the coefficients for each independent term and ε is the
error term.
For a complex device like the robot end-effector
design studied here, a second-order regression model
is chosen as non-linear behaviour of its response upon
factor changes is expected. Each factor in an FCCD
is configured to three levels: low, medium and high.
These levels are generally codified as -1, 0 and 1
respectively. Actual settings for each factor were then
interpolated referring to their actual limits. Table 2
shows the experimental plan in actual values using
FCCD with a revised order to minimise human effort
in changing end-effector configurations.
Furthermore, centre runs were performed at different
points during the experiments in order to effectively
capture more of the experimental errors.
Table 2 Faced Central Composite Design of Experiments plan with the obtained results
(
8
)

212
Kourosh Eshraghi, Pingfei Jiang, Daniele Suraci and Mark Atherton
3. RESULTS
Table 2 shows the results that were post-processed
according to the formulations of Section 2.4. The
trapezium rule of integration was employed to
numerically calculate the work done. Two
independent regression models were developed based
on work done and maximum insertion force (Max-
Force). The model fit statistics are listed in Table 3.
Both models are significant (P-value < 0.05) and
express very good prediction capability with R2 values
close to 1. Table 4 shows the regression coefficients
and P-values that represent the significance of each
term in the regression model. The main and quadratic
effects of misalignments are very sizeable. Spring
stiffness, K, and pitch radius, PR, are effective design
parameters while hole set orientation, O, is not. The
quadratic effect of design factors are insignificant
however, most interaction terms, especially those
involving K, are significant.
Figure 6 is a typical experimental reading in which it
is possible to see the force-displacement relationship
of the robot arm. Typically, 45 N is required to drive
the arm at the set speed of 25 mm/s. Based on the Max-
Force prediction model, Figure 7 shows that when
both misalignments are large, maximum force could
reach up to a value of 660 N.
Figure 8 shows the mean effect of the significant
factors in terms of the observed work done. By
averaging, the observed responses when each factor
was at its higher or lower limits, the typical effect of a
design factor is determined [15]. Softer springs and
lower values of PR results in lower work done.
Obviously, coupling with large misalignments should
produce a higher value of work done. However, mean
response is significantly lower in the –Y and +X
directions. This is understandable for the Y-direction
as pitch compliances are not symmetrical. It is
interesting to have observed this in the X-direction as
this could indicate systems’ (end-effector, robot and
fluid port) preferred directions of compliance. This
phenomenon and the lack of symmetry are also
observable in Figure 7 as the red and blue lines do not
overlap, and the minimum point of the curve is slightly
away from the zero-misalignment point.
Figure 9 is a model-based surface plot showing the
effects and interactions of PR with K. When both
parameters are at their lower level, significantly less
work done is required to make the coupling. Plots of
Figure 10 are based on the Max-Force prediction
model. Again, the quadratic effect and lack of
symmetry is very clear. Design factors K and PR
reduce Max-Force significantly. When using the
softest springs, the average reduction is 43 N, when
pitch radius is minimum this value is 73N, when both
parameters are set to their lower values its 115 N with
a maximum reduction of 170N at -5mm misalignment
in the X-axis. Note that in cases of Y-misalignment,
Max-Force reaches values below zero. These
regression model predictions are inevitably not an
exact representation of physical reality. This is due to
the fit of the regression model and in Figure 11 the
residuals (difference in real and predicted response)
for each observation highlight that some of the
experiments have large residuals.
Table 3 Fit statistics of the regression models
Table 4 Coefficients of the regression models

PRELIMNARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED FLUID COUPLING
213
Figure 6 Experimental Force-Position curve for
misaligned insertion
Before
the clash,
F=45N
Arm Position (m)
Force (N)
Figure 7 Prediction of Max Force at nominal design
condition and varying misalignments
Max-Force (N)
X-axis Misalignment, XM
Effect of interaction
Figure 9 Predicted response surface plot: showing the interaction of design factors
Figure 8 Mean effect of experiment factors based on physical data

214
Kourosh Eshraghi, Pingfei Jiang, Daniele Suraci and Mark Atherton
Figure 11 Plots of error in real and estimated responses (plot of residuals)
Figure 10 Prediction of maximum insertion force with varying misalignment conditions
: showing the effect of design
factors on Maximum insertion Force
Max-Force (N)
X-axis Misalignment, XM Y-axis Misalignment, YM
Observation Number
Observation Number

PRELIMNARY STUDY ON END-EFFECTOR COMPLIANCE IN AUTOMATED FLUID COUPLING
215
4. DISCUSSION
Compliances of the end-effector are not symmetrical
in pitch motions. When the coupler pivots downwards
it only deforms one spring, if it pivots upwards it will
press on two springs. Thus, pivoting downwards due
to a corresponding misalignment requires less work or
force. This has encouraged a lack of symmetry in the
results which is also augmented by experimental error,
robot inherent compliance, backlash and other
hysteresis that create preferred directions of motion.
Robot compliances could overshadow the end-
effector compliance and explain the relatively lower
mean effect of design parameters. For future
experiments, where only end-effector compliance is of
interest, fixing and rigidising all but the insertion axis
of the robot will be beneficial in isolating the effects
of the end-effector compliance. Nevertheless testing
the end-effector ‘in-situ’ is important for representing
the real application.
It can also be observed that the second-order
regression model has limited ability to capture the
actual non-linearity. Although the regression statistics
suggest a very good fit, when the work done or forces
are low the predicted response can go negative in cases
of Y-misalignments. Other than the highly nonlinear
profile this could have been exited by the lack of
symmetry in the end-effector compliance. Increasing
the number of experimental levels from 3 to 5 could
improve the model and resolve this issue.
Overall results show that generally more compliant
springs reduce the work done. We also observed an
interaction between pitch radius and spring stiffness.
When both are at their lowest values, a significant
increase in performance is observed. This is intuitive
in that moments exerted by the spring are reduced at
lower pitch reduces. When there are no
misalignments, a force of about 45 N is required to
drive the robot arm at the nominal speed. Force
observed beyond this value is due to contact friction
and physical clash. When the port is misaligned 5mm
in X and Y, the maximum insertion force is around
660N. This value is very large when compared to the
typical force exerted manually by a human. Train
fluid ports are not designed for such loads, which also
increase requirements on the robot such that larger
motors and structures become necessary.
As understood from a typical peg-in-hole problem
when the misaligned peg travels across the chamfer
and into the hole, angular motions of the peg could
result in 2-point contacts that encourage jamming.
The current end-effector only moves in pitch, yaw and
the insertion directions and thus coupling is prone to
jamming as misalignments are linear. The
CyberFluids robot can generate large enough forces
that exceed the jamming force. As this occurs, robot
arm bounces forward and the springs release energy
hence the sudden drop of force in Figure 6.
Simple modifications to the CyberFluids end-effector
can increase its performance and help to prevent
reduce insertion forces. By incorporating clearance
holes in the coupler flange, and removing the locking
nuts from the spring side, the coupler will attain
pivoting capabilities relative to the bolt. In 2D, this
arrangement becomes analogous to a parallelogram
linkage. The double pivoting action stacks to allow
linear motion in X or Y. It also desirable to have
symmetrical compliance in the X-Y plane, thus 4
equally spaced springs should be incorporated. 2 along
the Y-axis and 2 along the X-axis. Compliance in the
insertion direction is not necessary but it is inherent to
this end-effector design concept. Nonetheless, this
feature could be useful as a safety feature for robots
that cannot limit force/torque of actuators.
Flexure mechanisms [16] have become very popular
over the last two decades. There are many inherent
advantages in solving the same design problem using
monolithic, distributed compliance mechanisms.
Good examples are Constant Force Mechanisms
(CFM) [17] that regulate surface contact forces and
generate compliance at the end-effector. Therefore, in
developing new compliant end-effectors, flexure-
based mechanism incorporating passive compliance
capable of handling a larger range of misalignments
will be developed for a peg-in-hole scenario. Such an
end-effector can solve another limitation of our work;
where we assumed ports would be horizontally
located on the train. Depending on how the rail
industry goes forward with modifying the fluid ports,
simple robots with compliant end-effectors could
deliver a better economic solution than very
sophisticated robots with many sensors and DoF.
5. CONCLUSION
Automated servicing of trains is being seriously
considered by the rail industry with the aim of
releasing humans from unsuitable tasks and improving
health and safety in maintenance depots. The benefit
of having passive compliance in a robot end-effector
will help to improve the robustness of fluid coupling
whilst reducing the reliance on accurate robot end-
effector positioning systems. In this work, we have
investigated the role of end-effector compliance in

216
Kourosh Eshraghi, Pingfei Jiang, Daniele Suraci and Mark Atherton
enhancing the mechanical connection of fluid ports
under positional uncertainties. Results show that
misalignments have an exponential effect on the work
done and the maximum force of insertion. When the
fluid port is misaligned, having softer springs at a
lower pitch radius can reduce the maximum insertion
force by up to 160N. Likewise, there is a reduction of
work done by the insertion motor indicating an overall
better coupling. Yet still the forces involved are too
large and need to be reduced. As discussed, small
modifications to the current end-effector design can
result in better coupling performance. This is a next
step in our research on end-effector compliance
design, which focuses on relaxing the insertion force
relationship to misalignments.
Through this preliminary study, it is apparent that a
Design of Experiments with more levels will increase
the accuracy of the prediction model. In addition, the
inherent robot compliances and hysteresis can
enhance or disturb the coupling process. Hence,
future experiments will use a highly rigid rig for
isolating end-effector compliance more effectively.
ACKNOWLEDGEMENTS
We especially thank and acknowledge the Rail Safety
and Standards Board (RSSB) for funding and
supporting this research (contract RSSB 2675).
REFERENCES
[1] Eversholt Rail Group; Angel Trains; Porterbrook;
Rail Delivery Group. Long term passenger rolling
stock strategy for the rail industry. 6th ed., Rail
Delivery Group, 2018.
[2] Atherton, M.; Hill, S.; Harrison, D.; Ajovalasit, M.
Economic and technical feasibility of a robotic
autonomous system for train fluid servicing. J. Rail
Rapid Transit, 2019.
[3] Rail Safety and Standards Board, Fluid servicing
robot – developed by Brunel University with TBG
Solutions, 2019
https://www.youtube.com/watch?v=cMn5AIZbx0U
&t=4s accessed 5 November 2019.
[4] Ang, M.H. and Andeen, G.B., 1995. Specifying and
achieving passive compliance based on manipulator
structure. IEEE transactions on robotics and
automation, 11(4), pp.504-515.
[5] Wang, W., Loh, R.N. and Gu, E.Y., 1998. Passive
compliance versus active compliance in robot-based
automated assembly systems. Industrial Robot: An
International Journal, 25(1), pp.48-57.
[6] Zhang, Y., Lu, H., Pham, D.T., Wang, Y., Qu, M.,
Lim, J. and Su, S., 2019. Peg–hole disassembly using
active compliance. Royal Society open science, 6(8),
p.190476.
[7] Whitney, D.E., 1979. What is the remote centre
compliance (RCC) and what can it do?. In
Proceedings of the 9th International Symposium on
Industrial Robots, 1979.
[8] Lee, S. (2005). Development of a New Variable
Remote Center Compliance (VRCC) With Modified
Elastomer Shear Pad (ESP) for Robot Assembly.
IEEE Transactions on Automation Science and
Engineering, 2(2), pp.193-197.
[9] Gandhi, P., Bobade, R. and Chen, C. (2018). On novel
compliant mechanisms for remote center motion.
Advances in Mechanical Engineering, 10(4),
p.168781401876192.
[10] Xu J, Hou Z, Liu Z, et al. Compare contact model-
based control and contact model-free learning: A
survey of robotic peg-in-hole assembly strategies.
Preprint (2019), arXiv:190405240.
[11] Shigley, J. (2006). Standard handbook of machine
design. 2nd ed. Norwich, NY: Knovel.
[12] Montgomery D. C., Design and analysis of
Experiments, 5th ed. USA: John Wiley & Sons, Inc.,
2001.
[13] Ankenman, B. E. “Design of Experiments with Two-
and Four-Level Factors,” J. Qual. Technol., vol. 31,
no. 4, pp. 363–375, 1999.
[14] Myers R. H., “Response surface methodology -
Current status and future directions,” J. Qual.
Technol., vol. 31, no. 1, pp. 30–44, 1999.
[15] Atherton, M. A. and R. A. Bates, “Searching for
improvement,” in Design and Information in Biology,
vol. 27, WIT Press, 2006, pp. 345–379.
[16] Howell, L., Magleby, S. and Olsen, B. (2013).
Handbook of compliant mechanisms. Chichester
(United Kingdom): John Wiley & Sons.
[17] Chen, Y. and Lan, C. (2012). An Adjustable Constant-
Force Mechanism for Adaptive End-Effector
Operations. Journal of Mechanical Design, 134(3).