Franklin Open 6 (2024) 100071

Contents lists available at ScienceDirect

Franklin Open
journal homepage: www.elsevier.com/locate/fraope

Multi-objective design optimization of an in-pipe inspection robot

Ahmed Zeid a, *, Amur Al-Yahmedi a, Riadh Zaier a, Issam Bahadur b
a
Department of Mechanical & Industrial Engineering, College of Engineering, Sultan Qaboos University, Muscat 123, Oman
b
Department of Mechanical Engineering, College of Engineering, Dhofar University, Salalah 211, Oman

A R T I C L E I N F O A B S T R A C T

Keywords: In addition to inspection, In-Pipe Inspection Robots (IPIRs) could be enabled to perform other tasks, such as
In-pipe inspection robots object retrieval and blockage clearing. In this study, a new IPIR is proposed with a design aimed at expanding its
Wall-pressing mechanism range of potential applications beyond inspection. The design methodology followed to facilitate the afore­
Design optimization
mentioned application expansion commences with the development of a parametric preliminary design of an
Multi-objective optimization
NSGA-II
IPIR, in which the sizing of the robot parts are determined based on an optimization problem formulated in this
work. The multi-objective optimization problem maximizes two contradictory objective functions, namely, its
payload capacity, and the range of pipe diameters in which it can fit (radial range). The problem’s Pareto front is
solved using the NSGA-II implemented by Matlab’s gamultiobj(.). The robot’s final design is selected based on one
of the Pareto-optimal solutions obtained from the Pareto front with a theoretical payload capacity of 11 kg and a
radial range of 33 mm. The robot was manufactured and tested to validate such theoretical results.

1. Introduction An IPIR’s payload capacity refers to the maximum payload it can

carry while traversing a pipeline before its performance is compromised.
In-Pipe Inspection Robots (IPIRs) have become a viable option for The importance of this parameter comes from the fact that it outlines the
pipeline inspection for years. The miniature size of these robots, enabled range of applications of which a given IPIR is capable. In tool retrieval
by the miniaturization of manufacturing processes and electronics, al­ tasks, where IPIRs can be sent down a pipeline to retrieve a foreign
lows them to seamlessly inspect narrow pipelines that have been re­ object of interest or rescue another inspection tool that has failed during
ported to be as small as 1 inch [1], rendering impossible any human service, the maximum payload the IPIR is capable of is a determining
visual inspection. Moreover, as these robots are actuated, some of them factor in its retrieval abilities. Similarly, the payload capacity is also a
have the ability to steer through different pipe bends as well, a challenge determining factor to the robot’s applicability to oil and gas downhole
faced by endoscopic inspection. [2], for instance.
Apart from inspection, and despite what their name might imply, The second design parameter of interest in the design of a versatile
these robots can also be equipped to conduct a number of in-pipe IPIR is the radial range. The IPIR’s radial range in this research work is
maintenance and repair tasks such as blockage clearing, wellbore dril­ defined as the range of pipe diameters into which an IPIR can be
ling, tool retrieval (fishing in oil and gas context), and data logging. Such deployed where no critical performance deterioration occurs. Such
tasks, however, are rather more challenging to design for, as they are performance deteriorations include robot slippage, stickiness, and
often more demanding than regular inspection tasks in terms of the instability. This parameter can hence outline the scalability and versa­
required robot robustness, payload capacity, and overall reliability. For tility of an IPIR to a number of pipe sizes and applications. In oil well
instance, wellbore drilling or tool retrieval and delivery are more applications, for instance, an IPIR must be adaptable to a number of
demanding in regard to the robot’s payload capacity, especially in a different wellbore diameters within a given site. Moreover, it must be
vertical pipeline configuration such as in a vertical oil well. In designing adaptable to the in-pipe diameter variations in a single wellbore weather
IPIRs for such tasks, several design parameters become of interest due to irregular surfaces, wall sedimentations, or depth [2]. Having an
including the robot’s payload capacity and radial range. These are pa­ IPIR capable of a wide radial range essentially enables it to complete
rameters whose effect becomes a decisive factor in a given robot’s ability tasks in various standard pipe sizes across different applications. An
to complete said tasks successfully. advantage which IPIRs aimed at a specific application such as [3–5] may

* Corresponding author.
E-mail address: Azeid7@gmail.com (A. Zeid).

https://doi.org/10.1016/j.fraope.2024.100071
Received 28 September 2023; Received in revised form 7 December 2023; Accepted 10 January 2024
Available online 12 January 2024
2773-1863/© 2024 The Authors. Published by Elsevier Inc. on behalf of The Franklin Institute. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
A. Zeid et al. Franklin Open 6 (2024) 100071

not particularly have [6]. of a static-force model, and geometrically in the form of a kinematic
Due to the large number of variables that affect each of the payload model. These two coupled mathematical models would form the basis
capacity and radial range of an IPIR as mentioned earlier, creating a for modelling the payload capacity and radial range objective functions
design that maximizes these two parameters is not straight-forward; the considered in the optimization problem. As for the problem’s decision
variables of interest would have to be tuned simultaneously into a variables, these are determined based on a screening process to the
unique combination that maximizes these two design parameters. As a variables featured in the developed model above where the variables
result, optimization is to be utilized in designing the intended IPIR in with the most flexibility in their values are selected. The optimization
order to identify a solution that maximizes the two objective functions problem is then constrained in light of the allowable mathematical re­
considered, i.e., the payload capacity and the radial range. This work, lationships between the decision variables selected. Consequently, the
hence, focuses on improving the range of applications achievable by a optimization problem is solved, and an ‘optimal’ design is applied to the
single IPIR. parametric preliminary design. The optimized design is lastly manu­
Design optimization as an approach for the design of IPIRs has been factured to allow for experimental validation of the developed model.
reported in the literature at different capacities. In [7], the authors As to be explained later in Section 4 of this manuscript, the two
exploited design optimization in sizing the wall-pressing mechanism of a objective functions considered are of contradictory nature in this
proposed IPIR. The optimization problem modelled was a particular design, i.e., an increase in the value of one of them results in a
Multi-Objective Optimization Problem (MOOP). The first objective decrease in the value of the other. In such scenarios, Multi-Objective
function in the problem was aimed at minimizing the mechanism’s size, Optimization (MOO) must be utilized. In solving a Multi-Objective
while the second was aimed at maximizing the force transmission effi­ Optimization Problem (MOOP), and although the single design point
ciency from the actuator to the mechanism. In their work, three different solution is possible, it is the Pareto front solution that is often sought.
mechanisms were considered, modeled, and optimized for the same The Pareto front, or frontier, is a graphical representation of the solution
quantities. These mechanism were a slot-follower mechanism, a space. Points, or solutions, along this frontier offer a certain trade-off
slider-crank mechanism with 4 bars, and another with 6 bars. The so­ between the objective functions considered such that each point repre­
lutions obtained for the three problems were then used as basis for sents the optimal value of one objective function after which the value of
mechanism-type selection as well as sizing. In [8], optimization was the other objective function(s) deteriorates. This type of optimality is
utilized to size the structure of a proposed IPIR. The robot featured a referred to as Pareto-optimality [14]. After the Pareto front of the MOOP
locking mechanism actuated by a stepper motor. The objective function formulated in this work is obtained, a single Pareto-optimal solution is
considered in their work was to minimize the required motor torque selected based on the trade-off it offers between the two objective
while maximizing the motor speed. In [9], the wall pressing mechanism functions considered and the relative importance between them.
of an in-pipe cleaning robot was sized in a way that minimizes the
pre-tension on the compression springs due to the bare weight of the 3. Mathematical modelling and optimization
robot, hence the now-alleviated springs can accommodate extra
compression for an additional payload. In [10], optimization was Fig. 1 shows a 3D CAD model of the proposed IPIR developed in a
employed in solving three different design problems. In their work, the previous work [15]. The robot is of a wall-pressing type, meaning that
authors modeled three separate Single-Objective Optimization Problems when pushed into a pipe with a smaller diameter relative to its size, its
(SOOPs), one for each of the considered design problems. The first links are forced to fold radially inward compressing the spring contained
optimization problem related to sizing the tensegrity mechanism within the suspension cylinder in order to match the pipe’s diameter as
featured in the robot to maximize the robot’s static stability. The second demonstrated in Fig. 1(b). This compression creates a normal force be­
optimization problem considered was a motor selection problem that tween the robot’s wheels and the pipe walls, resulting in sufficient
resulted in the selection which also determined the corresponding friction forces that support the weight of the robot.
nominal pipe diameter suitable for inspection by the robot. The third
and final optimization problem was aimed at maximizing the space 3.1. Objective function evaluation
occupied by the robot’s wall-pressing mechanism.
Furthermore, design optimization has been employed in the design As explained, a static-force and a kinematic mathematical model are
of IPIR at capacities different from sizing its components. In [11], the developed for the proposed design. Normally, the mathematical model
power consumption of a proposed Inchworm-type IPIR is minimized of an objective function is a linear/nonlinear combination of its decision
through determining the optimal frequency for the robot’s motion. In variables expressed in equation form. The way by which the objective
[12], optimization was used to control a proposed IPIR in a way that functions in hand are to be modelled, however, is not as straightforward.
maximizes its accuracy while minimizing its resulting power consump­ This is because neither of them can be represented by a closed-form
tion. In [13], the power consumption of a proposed screw-drive IPIR was mathematical expression. This type of objective functions are referred
minimized using two approaches. The first minimized the power con­ to as Black-Box objective functions. Based on investigation, which is to
sumption by obtaining an optimal path and input control conditions for be addressed in detail in Section 4 of this manuscript, it was found that
the robot between two given points, while the second minimized the the radial range of the proposed robot was rather a function of the
power consumption by introducing blades on the robot’s body, which payload. That is, for every payload value, a corresponding, and unique,
enhanced its dynamics in a fluid median and subsequently its power range of pipe diameters is allowable for the robot’s deployment. In this
consumption. work, such diameter range is referred to as the Specific Radial Range
To the best of our knowledge as evident by the above review, the two (SRR). Calculating an SRR that corresponds to a given payload, i.e.,
objective functions considered in this work were not explored together finding the minimum and maximum diameters that bound said range, is
before in the context of IPIR optimization, let alone in the context of not straightforward either and is not of a closed-form. This essentially
extending an IPIR’s range of potential applications. means that in order to find the minimum and maximum possible di­
ameters that correspond to a certain payload, one must perform a
2. Methodology simulation that starts at the minimum diameter physically achievable by
the robot, and incrementally increase the diameter value to the
In order to tackle the abovementioned problem statement, the maximum diameter physically achievable by the robot. At each step
following methodology is implemented. First, a new IPIR design is within this simulation, a number of conditions are checked, determining
proposed and 3D-modelled in Solidworks with preliminary dimen­ whether the given payload can be supported at the current diameter,
sioning. Next, the developed design is modeled structurally in the form from both structural and geometrical considerations. The SRR for a

2
A. Zeid et al. Franklin Open 6 (2024) 100071

Fig. 1. Working Princible of a Wall-Presing IPIR: (a) Relaxed State, (b) Compressed State.

given payload hence extends from the smallest diameter that does not
violate the conditions and ends before the first diameter that violates
them. In order to determine the SRRs that correspond to a set of payloads
within a given range, the described simulation is repeated for every
payload increment within the said range.
Since radial ranges are only specific to their respective payloads, a
single SRR is not representative the overall radial range of the robot over
the entire payload range. Hence, another parameter that bypasses this
issue is introduced, the Common Radial Range (CRR). This range is
defined as the intersection of all Specific Radial Ranges over a given
payload range. Within the robot’s CRR, all payloads within the given
payload range can be supported by the robot’s wall-pressing
mechanism.
Technically, the process by which the two objective functions in this
work are evaluated is to be implemented programmatically in Matlab in
the form of a nested For-loop. The outer loop increments through a
preliminary payload range that starts from the free mass of the robot,
taken as the mass of the preliminary design obtained from the Solidworks
CAD model, and ends with an arbitrarily high payload value to ensure
that the simulation does not exit prematurely before reaching the actual
payload capacity. Since the optimized robot is quite likely to have a mass
different from that of the preliminary robot’s, the optimization problem
is solved a number of times where in each iteration the mass of the robot Fig. 2. The Distance h.
is taken as the mass of the robot from the previous iteration. The iter­
ative process is terminated when the assumed robot mass is close enough payload value within the payload search range after which the condi­
to the robot mass obtained through the optimization process. Moreover, tions must be violated to achieve the desired configuration.
the maximum payload value analyzed in the simulation is to be adjusted Lastly, and upon exiting from the nested loop, the robot’s radial
during the same iterative process discussed such that it is taken as a range, measured by its Common Radial Range, is calculated as the
value only slightly higher than the maximum payload capacity observed intersection of all the Specific Radial Ranges obtained within the search
in the obtained Pareto front from each iteration. Setting the payload range.
range through this iterative process enables a more accurate solution
exploration with better resolution and dispersion over the range of 3.2. Static-force modeling
payloads analyzed in the simulation process.
The inner loop, on the other hand, increments through a range of The following mathematical analysis is performed on the robot under
values of the distance h, which represents the axial distance covered by static-equilibrium conditions inside a vertical pipe when it is passively
the slide blocks along the lead screw measured from the lead screw’s capable of supporting its own weight through the springs and starts to
sleeve as shown in Fig. 2. When h is increased i.e., the slide blocks are move upwards. This scenario in particular is selected for the analysis
slid towards one another along the lead screw, the robot’s diameter because, compared to moving downward a vertical pipe or traversing
increases as can be depicted from Fig. 1 and Fig. 2. The distance h is sideways in a horizontal pipe, the investigated scenario is the one in
hence controlled to adapt the proposed robot to different pipe diameters. which the desired motion opposes gravity the most and is hence most
Looping through increments of this parameter effectively loops determinate to the analysis.
through the range of diameters geometrically achievable by the robot. Starting with the static-force analysis, and under static equilibrium,
The range of h hence starts from zero, to a value hmax where the upper the minimum normal force, N, between the pipe walls and each of the six
slide block becomes on the verge of touching the gear carrier. For every wheels required to support the robot of total mass M is calculated as
payload within the defined range, an SRR is obtained as per the criteria such:
discussed earlier. The robot’s payload capacity is hence taken as the

3
A. Zeid et al. Franklin Open 6 (2024) 100071

Mg
N= (1)
6μs

where μs is the coefficient of static friction between the rubber wheels

and the acrylic pipe, and g is the gravitational acceleration constant.
In this analysis, μs is taken as 0.3; a typical value for such application
[16]. Due to the influence of μs on the proposed robot’s payload ca­
pacity, this assumption is to be validated experimentally in this work,
nevertheless.
Moving to the links of the wall-pressing mechanism, Fig. 3 shows the
FBD of one of them since they are all of identical geometry and loadings.
∑
By equating the summation of moments about point O to zero, i.e., MO
= 0, the following relationship is derived:
Kxd
N cot(θ) = (cos(∅) + sin(∅)cot(θ)) − μs N − Ww − WL (2)
L

where θ is the angle the link makes with the vertical, ∅ is the angle the
suspension cylinder makes with the vertical, K is the spring stiffness in, x Fig. 4. Kinematic Diagram of the Five-bar Linkage.
is the spring’s compression, d is the distance shown in the figure, Ww is
the weight of the wheel, L is the length of the link, and WL is the weight
of the link. Sl = So + Lf − x (5)
It is worth mentioning that the case where the link is under dynamic
∑
equilibrium, i.e., MO ∕
= 0 is to be considered in a future study that where So is the length of the outer suspension cylinder, and Lf is the free
explores the dynamic behavior of the robot at diameter changes along length of the spring.
the pipeline. In order to allow for enough spring compression, the outer suspen­
sion cylinder must be of enough length to allow for the spring to undergo
full compression before the inner suspension rod reaches the bottom of
3.3. Kinematic modeling the outer suspension cylinder. For that, So is to be sized as per the
following relationship:
Moving on to the kinematic analysis, Fig. 4 shows the Five-bar
linkage created by the wall-pressing mechanism and the rest of the So = 0.5Lf + 20 (6)
robot.
where 0.5Lf represents an initial approximation of the spring’s
In light of the figure, the two following relationships are obtained:
maximum compression, and the term "20″ accounts for an extra axial
Sl cos(∅) + dcos(θ) = H (3) clearance in millimeters. This assumption is deliberately lenient in its
application in most cases, and it aims to avoid imposing an unnecessary
Sl sin(∅) − dsin(θ) = Hf − Nf (4) constraint on the model. For the selected design point, the value 0.5Lf is
to be substituted with the exact maximum compression of the chosen
where Sl is the length occupied by the suspension cylinder, H is an in­ spring.
termediate variable, and Hf and Nf are the centre hinge’s and the slide
The height H from Eq. (3) can also be further expanded as such:
block’s radial offsets, respectively.
Furthermore, Sl can be expanded as: H=
LR tH tN
− − − h (7)
2 4 2

where LR is the length of the lead screw, tH and tN are centre hinge’s and
the slide block’s thicknesses, respectively, and h is the slide block’s
displacement introduced previously.
Lastly, the robot’s outer diameter can be calculated as such:
OD = Hf + Lsin(θ) + r (8)

where r is the wheel radius.

To sum up, for a given candidate solution, and in order to calculate
its corresponding objective functions’ values, the simulation-based
process demonstrated in Fig. 5 is followed. For every M value ranging
from 1.5 to 20 kg, determined by the iterative process introduced earlier
in the manuscript, h is varied from 0 to hmax , where hmax is calculated as
such:
LR tH
hmax = − − 38 (9)
2 2

where “38″ is a fixed distance in mm that accounts for the height of the
gear carrier plus a small clearance to ensure a collision-free motion
between the top slide block and the gear carrier.
With the mass M known, the required normal force, N is calculated
from Eq. (1). Next, Eq. (2) through Eq. (5) can then be solved simulta­
neously to obtain θ, ∅, Sl , and x with So and H already calculated from
Fig. 3. FBD of the Link.

4
A. Zeid et al. Franklin Open 6 (2024) 100071

Fig. 5. Pseudocode for Evalvuating the Objective Functions.

Eq. (6) and Eq. (7), respectively. Finally, the robot’s outer diameter OD a given payload only outlines the range of pipe diameters that are rec­
can be calculated using Eq. (8). The simulation progresses with the ommended at this payload value in light of the traction forces required.
conditions being checked for after every iteration, as explained, until This basically means that a given payload may exist outside its corre­
they are violated. In each of these iterations, the SRR corresponding to sponding SRR but only at a traction force higher than the one required to
each M value is obtained. The M value at which the developed Matlab support its weight. Such a case leads to excessive power consumption, an
code exits denotes the maximum total robot mass i.e., the mass of the operational condition that is not preferred for the proposed battery-
robot plus the mass of the payload that the robot structure can support powered robot. In this work, this potential range of pipe diameters is
(Mmax ). referred to as the maximum radial range.
The robot’s payload capacity can then be calculated as such:
Pmax = Mmax − m (10) 3.4. Decision variables and constraints

where Pmax is the payload capacity, Mmax is the maximum total robot The flexible variables in the above analysis are the link length, L, the
mass, and m is the base robot mass. spring’s free length, Lf , the distance d, the spring stiffness, K, and the
The Common Radial Range, CRR can then be calculated as: lead screw length, LR , which were hence taken as the decision variables
CRR = max(ODmin ) − min(ODmax ) (11) of the optimization problem in this work. These variables are.
Lastly, the decision variables are only constrained by the relation­
where ODmin and ODmax are vectors containing the minimum and ship:
maximum specific diameter values corresponding to the values of h d ≤ 0.4L (12)
iterated over, respectively.
As for the conditions checked for each payload-diameter pair, they This ensures that the position of the suspension cylinder does not
include the conditions listed in Table 1. surpass the centre of the link in Fig. 3 for two main reasons. First, it
At this point, it should be made clear that the SRR that corresponds to enhances the compactness of the robot by enabling it to occupy smaller
pipe diameters and, secondly, it helps in maintaining the lengths of both
the suspension cylinders and springs at small manageable values, which
Table 1 reduces their chances of link buckling. These visual observations were
The Checked Conditions.
made while developing the 3D CAD model of the preliminary design.
Condition Clarification

xmax ≥ x ≥ 0.5 Spring compression accuracy bellow 0.5 mm is not guaranteed in

3.5. Solving the optimization problem
practice and compression beyond spring’s maximum compression
limit is not possible
π
>θ>0 Link angle is limited by this range to avoid mechanism singularity Since in Pareto optimization multiple solutions are desired, the
2 points NSGA-II [17] is suited to such application as it inherently deals with a
π
>∅>0 Suspension angle is limited by this range to avoid mechanism population of candidate solutions rather than a single candidate in each
2 singularity points
iteration. Moreover, the Genetic Algorithm (GA) (on which the NSGA-II

5
A. Zeid et al. Franklin Open 6 (2024) 100071

is based) is fundamentally a global optimization technique that is robust the objective function’ values within these two points.
against the locality trap to a certain extent as it explores the entirety of In this study, a design point that achieves a balance between the two
the solution space even if a good enough feasible solution has already objective functions is favored, with a slight inclination towards
been found, which renders it reliable in converging to a global optima. enhancing the payload capacity. This bias stems from the recognition
Lastly, the implementation of NSGA-II is readily available through that payload capacity often holds greater significance in defining the
Matlab and the Matlab Global Optimization Toolbox. It is also worth range of applications associated with an IPIR. For example, an IPIR with
mentioning that simple gradient-based optimization techniques have increased payload capacity and a constrained radial range can effec­
their limitations when dealing with Black-Box Optimization as such tively perform demanding tasks like tool retrieval, despite the limited
techniques require the objective functions to be continuously differen­ range of pipe diameters. Conversely, an IPIR with a reduced payload
tiable, a requirement that is either unguaranteed or computationally capacity and a wide radial range may struggle to support the weight of a
expensive in Black-Box objective functions [18]. retrieved tool across various pipe diameters. Consequently, during the
Based on the above, the NSGA-II is used in solving the MOOP selection of the final design point, it was determined that candidate
formulated in this work, namely, Matlab’s gamultiobj(.) function. Since design points must exhibit a payload capacity to CRR ratio of no less
gamultiobj(.) only supports function minimization, the negative values of than 2:1 and no greater than 3:1. Beyond this range, the CRR function
the two considered objective functions are minimized in this work, diminishes in relative importance. To appropriately represent these ra­
effectively maximizing their values. tios, both objective functions were normalized. Three candidate design
points highlighted in red were identified within the predefined bound­
ary lines, as illustrated in Fig. 7.
3.6. Solution selection
To assess the real-world performance of these three design points, the
finalization of the actual spring designs to be implemented in each of
Upon solving the optimization problem formulated above, the Pareto
them is a prerequisite. Subsequently, their realized or corrected per­
front shown in Fig. 6 obtained. The computational time required to
formances, specifically in terms of payload and radial range, will be
generate the front was around 2 h with these solver settings: ‘Population
compared. The spring designs for each design point must align with the
Size’: 300, ‘Maximum Stall Generations’: 30, ‘Function Tolerance’: 10− 4,
reported Lf and K values or closely approximates them. Table 2 shows
‘Constraint Tolerance’: 10− 4.These values were set based on trial and
the corrected objective function values for the points after implementing
error, as they were found to reduce the computational time required
the off-the shelf spring designs whose Lf and K values best approximate
without compromising the solver’s convergence to the global solutions.
The computer uses an Intel i7–8550 U CPU at 1.80 GHz and 16 GB RAM. those required for each of the design points. Table 2 presents the cor­
From the obtained Pareto front, it can be seen that the values of the rected objective function values for the points after incorporating off-
payload capacity function extend from 0 to 17.4 kg, whereas the values the-shelf spring designs.
of the CRR function extend from 3.5 to 151.4 mm. The obtained Pareto- Given that the performances of the three candidate points are still
optimal points were found to generally follow the shape of the power indistinguishable in terms of superiority, an aggregate function is
function overlaid on the figure (R2 = 0.981), which represents an introduced to facilitate a more definitive comparison by combining the
approximate of the problem’s Pareto front. Points ’A’ and ’B’ represent two objective functions. The aggregate function is formulated as follows:
the optimal design points that would have been obtained had the √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
G = PC2norm + CRR2norm (13)
problem been set-up as a SOOP with either CRR or payload capacity
objective functions considered, respectively. For instance, in applica­
tions where the proposed parametric design is required to carry where G is the aggregate function, PCnorm is the normalized payload
maximum payload with no pipe size restrictions, design point ‘A’ is capacity, and CRRnorm is the normalized CRR. Note that the normalized
preferred. On the contrary, in lightweight applications such as simple objective functions are used here such that the aggregate function is not
inspection tasks, the payload is merely the camera and a wide range of affected by scale difference in the two functions. The objective functions
pipe diameters is favored, making point ‘B’ an ‘optimal’ design point. are normalized by dividing them by their respective Frobenius norm.
Hence, any design point along the obtained Pareto-optimal points ob­ As indicated in Table 2, design point (1) was identified with the
tained, apart from the two points addressed, offer a unique trade-off in highest aggregate function values, leading to its selection for

Fig. 6. The Obtained Pareto Front. Fig. 7. The Normalized Pareto Front.

6
A. Zeid et al. Franklin Open 6 (2024) 100071

Table 2 maximum outer diameter the robot can geometrically achieve before
Corrected and Aggregate Objective Functions’ Values for the Candidate Design violating the geometric conditions considered. As a result of the trends
Points. of these two arrays, the SRR increases nonlinearly along the same
Design Corrected Payload Corrected CRR Aggregate Objective payload range since it is none but the difference between the two arrays
Point Capacity (kg) (mm) Function as can be seen from Fig. 7(b).
(1) 10.990 33.066 0.713 Moreover, and from Fig. 7(a), it can be observed that the proposed
(2) 11.465 30.379 0.699 robot’s CRR of 33 mm is actually equivalent to the SRR at 0 kg payload,
(3) 11.736 27.802 0.681 as it represents the narrowest radial range achievable at all payloads,
while its maximum radial range of 91.77 mm is equivalent to that at the
11 kg payload. As a result, the CRR can be defined by the smallest
implementation. It’s important to note that using the aggregate function
diameter the robot can achieve, which occurs at a 0 kg payload, and the
as the single objective function for the problem would yield different
robot’s SRR at the same payload. This gives an explanation to the con­
outcomes. This is due to the fact that the value of the summation in Eq.
tradictory nature between the payload and the CRR objective functions
(13) is not unique.
that characterizes this analysis. This was found to be induced by two
decision variables; Lf and d, that each carries this contradictory behavior
4. Simulation results and discussion
when it comes to improving the objective functions. It can be shown that
a shorter spring, and hence suspension cylinder, results in a more
For the selected design, the mathematical analysis represented by
compact robot that is capable of achieving smaller diameters at 0 kg
Fig. 5 was implemented in Matlab. Fig. 8(a) shows the variation in the
payload. Such capability increases the robots CRR as explained earlier
arrays ODmin and ODmax from Eq. (11) in mm, while Fig. 7(b) shows the
from Fig. 7(a). On the other hand, a shorter spring allows for a smaller
Specific Radial Ranges in mm corresponding to these arrays, both as
compression range and hence a lower payload capacity. Similarly, the
functions of the payload in kg for the selected design point. As can be
shorter length d is, the more compact the robot is, but the lower the
seen from Fig. 7(a) below, ODmin decreases nonlinearly from 275 to 235
compression range of which the suspension cylinder is capable. The
mm when the payload increases from 0 to 11 kg. In its essence, ODmin
effect these two variables have on the objective functions becomes
corresponds to the minimum spring compression that creates enough
clearer in light of Table 3 where the objective functions’ values of three
traction to support a given payload. Higher spring compressions at the
arbitrary design point are compared.
same payloads, although possible, would translate to higher power
The table compares point A with points B and C which differ from
consumption and dissipation rates, which are not desired for the pro­
point A only in their Lf and d value, respectively. As can be seen from the
posed robot since it is partially aimed at power conservation and effi­
table, and by comparing point B to A, a decrease in the spring length
ciency. For that reason, and at lower payloads, the required spring
corresponds to a decrease in the payload capacity and an increase in the
compressions are minimal, which causes the robot to have larger ODmin
CRR as explained above. Furthermore, and by comparing point C to A, a
values, and vice versa for larger payloads. Furthermore, ODmax increases
reduction in the length d has the same effect on the two objective
nonlinearly from 308 to 326 mm before flattening out when payload
functions.
increases from 0 to 11 kg, which is a direct result of the cap the condi­
tions listed in Table 1 impose. Larger outer diameters correspond to
5. Model validation
flatter suspension angles, which work as a spring-force multiplier and
reduces the required spring compression to support the same payload as
In this step, the manufactured robot (Fig. 9(a)) is put through a test
compared to smaller outer diameters as explained earlier. These com­
that assess the fidelity of the mathematical model developed for the
pressions, and at smaller payloads, drop beyond 0.5 mm, violating
design. The validation step consists of a single experiment in which the
Condition 1, and limiting the ODmax value the robot can achieve. At
proposed robot is placed in a vertical pipe under normal operating
higher payloads on the other hand, the condition is less likely to be
conditions while a set of known weights are attached to its end and as
violated since the required spring compressions are large enough. The
illustrated in Fig. 9 (b, c). The weights are added incrementally until the
flat portion towards the end of the ODmax curve corresponds to the
robot is found to lose contact with the pipe walls and starts slipping. The
total mass at which slippage occurs denotes the experimental payload
capacity of the robot, which is to be compared to the one obtained
theoretically from the model, validating it.
As discussed earlier, the current mathematical model assumes a μs
value of 0.3. It is necessary, nevertheless, to measure the actual μs value
for the current experimental apparatus. To do so, the robot was placed
inside the pipe with no payload and the lead screw was manually rotated
in the direction that increased the robot’s outer diameter until the robot
was observed to support its own weight. The robot was then retrieved
from the pipe, and the current h distance was measured as 1.917 mm.
Thereafter, and through the Matlab code, it was found that at a 270 mm
pipe and no payload, an h value of 1.917 mm is required only when μs is
equal to 0.114, denoting the experimental coefficient of static friction

Table 3
Objective Functions’ Values for Three Design Points.
Point Decision Variable Vector* Payload Capacity (kg) CRR (mm)

A [120, 33, 48, 9.9, 200] 10.571 34.053

B [120, 30, 48, 9.9, 200] 8.01 43.175
C [120, 33, 40, 9.9, 200] 9.52 51.429
*
Fig. 8. Simulation Results: (a) ODmin & ODmax and (b) Specific Radial Range Vs. X = [L, Lf , d, K, LR ].
Payload, both for The Selected Design Point.

7
A. Zeid et al. Franklin Open 6 (2024) 100071

Fig. 9. (a) The Prototyped Proposed Robot, (b) Schematic Diagram of the Validation Experiment Setup, (c) Top View of the Experiment.

between the rubber wheels and the plastic pipe used. between the rubber wheels and the springs is likely to have an effect on
Ideally, a similar-sized pipe with a μs value of 0.3 would be used in the experimental results to a certain extent. The implementation of
the intended test. However, this approach was not implemented in the metal wheels should eliminate this effect, nevertheless.
absence of methods, other than experimentation, to accurately deter­ The overall agreement between the theoretical and experimental
mine the coefficient of static friction between the wheels’ material and results obtained validates the mathematical model developed and favors
the pipe’s walls. Instead, the robot’s payload capacity obtained theo­ the 11 kg payload capacity and CRR of 33 mm claimed at the assumed μs
retically is adjusted for the value μs = 0.114 and the developed model’s value of 0.3.
fidelity is assessed against the corrected value using the available pipe.
The proposed robot’s corrected payload capacity at μs = 0.114 was 6. Conclusion and future work
calculated to be 2.449 kg.
First, and in order for the robot to be able to support the above­ In conclusion, a new battery-powered IPIR was designed, optimized,
mentioned payload, the required value of h was calculated from the and validated. The proposed design was modeled structurally and
model, and was found to be 8.248 mm. After the lead screw position was geometrically before the developed model was formulated as a MOOP
adjusted accordingly, the robot was inserted into the test pipe and the that sizes the proposed robot in such a way as to maximize its payload
pre-described payload capacity test was performed. The robot was capacity and radial range. The optimization problem was solved using
observed to slip at a payload of 2.493 kg, evidenced by the very slow slip the NSGA-II implemented by Matlab’s gamultiobj(.). The proposed robot
rate noticed, denoting the robot’s experimental payload capacity at μs = was observed to have a theoretical payload capacity of 11 kg and a radial
0.114. However, and due to the difficulty associated with identifying range of 33 mm as per the mathematical model developed. These
the slip onset, the reported results are susceptible to an inaccuracy of theoretical results were validated experimentally and were determined
±50 g. This tolerance limit exposes the experimentally reported result to to be solely 3.84 % less than those obtained experimentally. Such
a maximum percentage difference of 3.84 % from the theoretically ob­ payload capacity and radial range values would enable the proposed
tained one. Such a variation can be attributed to two main factors. robot to perform a number of in-pipe tasks that extend beyond simple
Firstly, the geometric tolerances in the suspension cylinders since they visual inspection in a wide range of pipe sizes, increasing its range of
were not precision-machined and rather depended on the skill of the potential applications.
manufacturer. Such small variations are likely to induce small variations In this study, the performance of the manufactured prototype was
in the springs’ compressions and ultimately the robot’s payload capac­ evaluated solely at a single pipe diameter. Future enhancements beyond
ity. Secondly, small-scale deformations in the rubber wheels occurred at the scope of the current work involve conducting experiments across a
high spring forces at higher h values as highlighted by Fig. 10. This variety of pipe diameters and materials to comprehensively assess the
deformation is merely a redirection of a small portion of the compression fidelity of the developed model. Additionally, it’s imperative to
the springs were expected to undergo. The variation in the elasticity acknowledge that the robot was exclusively tested in a dry pipe envi­
ronment. Therefore, to provide a more comprehensive evaluation,
testing under wet conditions and in active pipes (with flow) becomes
necessary. These proposed improvements will contribute to a more
robust understanding of the prototype’s capabilities and its applicability
in diverse real-world scenarios.

CRediT authorship contribution statement

Ahmed Zeid: Conceptualization, Methodology, Software, Writing –

original draft. Amur Al-Yahmedi: Conceptualization, Writing – review
& editing, Supervision. Riadh Zaier: Conceptualization, Writing – re­
view & editing, Supervision. Issam Bahadur: Writing – review &
editing.

Fig. 10. The Observed Wheel Deformation.

8
A. Zeid et al. Franklin Open 6 (2024) 100071

Declaration of competing interest the ASME Design Engineering Technical Conference, 2014, https://doi.org/
10.1115/DETC2014-34057.
[8] D. Qu, H. Song, J. Sun, Force Analyzing and Parameter Optimization of the
The authors declare that they have no known competing financial Telescopic In-Pipe Robot, Appl. Mech. Mater. (2015), https://doi.org/10.4028/
interests or personal relationships that could have appeared to influence www.scientific.net/amm.738-739.941.
the work reported in this paper. [9] Z. Yang, Y. Fan, Y. Yan, Development of an Adaptive Wall-pressing Mechanism for
an In-pipe Cleaning Robot, in: Journal of Physics: Conference Series, 2022, https://
doi.org/10.1088/1742-6596/2365/1/012023.
Acknowledgment [10] S. Venkateswaran, D. Chablat, P. Hamon, An optimal design of a flexible piping
inspection Robot, J. Mech. Robot. (2021), https://doi.org/10.1115/1.4049948.
[11] Q. Liu, Y. Chen, T. Ren, Y. Wei, Optimized inchworm motion planning for a novel
The authors would like to acknowledge the partial support of Sultan in-pipe robot, Proc. Inst. Mech. Eng. C J. Mech. Eng. Sci. (2014), https://doi.org/
Qaboos University, Petroleum Development Oman, and Ejaad Platform 10.1177/0954406213502409.
in this work. Their contributions are gratefully appreciated. [12] H. Tourajizadeh, M. Rezaei, A.H. Sedigh, Optimal Control of Screw In-pipe
Inspection Robot with Controllable Pitch Rate, J. Intellig. Rob. Syst. Theo. Appl.
(2018), https://doi.org/10.1007/s10846-017-0658-7.
References [13] H. Tourajizadeh, V. Boomeri, M. Rezaei, A. Sedigh, Dynamic Optimization of a
Steerable Screw In-pipe Inspection Robot Using HJB and Turbine Installation,
[1] K. Suzumori, T. Miyagawa, M. Kimura, Y. Hasegawa, Micro inspection robot for 1- Robotica (2020), https://doi.org/10.1017/S0263574719001784.
in pipes, IEEE ASME Trans. Mechatron. (1999), https://doi.org/10.1109/ [14] P. Ngatchou, A. Zarei, M.A. El-Sharkawi, Pareto multi objective optimization, in:
3516.789686. Proceedings of the 13th International Conference on Intelligent Systems
[2] R. Tao, Y. Chen, L. Qingyou, A helical drive in-pipe robot based on compound Application to Power Systems, 2005, https://doi.org/10.1109/
planetary gearing, Advanced Robotics (2014), https://doi.org/10.1080/ ISAP.2005.1599245.
01691864.2014.913499. [15] A. Zeid, A. Al-Yahmedi, R. Zaier, I. Bahadur, Design and control of a diameter-
[3] H.M. Kim, Y.S. Choi, Y.G. Lee, H.R. Choi, Novel mechanism for in-pipe robot based adaptable in-pipe inspection robot, in: IEEE International Multidisciplinary
on a multiaxial differential gear mechanism, IEEE ASME Trans. Mechatron. (2017), Conference on Engineering Technology, 2023, https://doi.org/10.1109/
https://doi.org/10.1109/TMECH.2016.2621978. IMCET59736.2023.10368241.
[4] J.H. Kim, G. Sharma, S.S. Iyengar, FAMPER: A fully autonomous mobile robot for [16] Q. Liu, J. Zhao, H. Zhu, G. Wang, J.D. McLennan, Review, classification and
pipeline exploration, in: Proceedings of the IEEE International Conference on structural analysis of downhole robots: Core technology and prospects for
Industrial Technology, 2010, https://doi.org/10.1109/ICIT.2010.5472748. application, Rob. Auton. Syst. (2019), https://doi.org/10.1016/j.
[5] S.G. Roh, W.K. Do, J.S. Lee, H. Moon, R.C. Hyouk, Modularized in-pipe robot robot.2019.02.008.
capable of selective navigation inside of pipelines, in: IEEE/RSJ International [17] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist multiobjective
Conference on Intelligent Robots and Systems, 2008, https://doi.org/10.1109/ genetic algorithm: NSGA-II, IEEE Trans. Evol. Comput. (2002), https://doi.org/
IROS.2008.4650968. 10.1109/4235.996017.
[6] P. Li, S. Ma, B. Li, Y. Wang, Multifunctional mobile units with a same platform for [18] I. Bajaj, A. Arora, M.M.F. Hasan, Black-Box Optimization: Methods and
in-pipe inspection robots, in: IEEE/RSJ International Conference on Intelligent Applications, Springer Optimization and Its Applications, 2021, https://doi.org/
Robots and Systems, 2008, https://doi.org/10.1109/IROS.2008.4650583. 10.1007/978-3-030-66515-9_2.
[7] R. Henry, D. Chablat, M. Porez, F. Boyer, D. Kanaan, Multi-objective design
optimization of the leg mechanism for a piping inspection robot, in: Proceedings of