Automation in Construction 141 (2022) 104400

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

AR-based automatic pipeline planning coordination for on-site mechanical,

electrical and plumbing system conflict resolution
Liang-Ting Tsai a, Hung-Lin Chi b, *, Tzong-Hann Wu a, Shih-Chung Kang c
a
Department of Civil Engineering, National Taiwan University, Taiwan
b
Department of Building and Real Estate, The Hong Kong Polytechnic University, Hong Kong, China
c
Department of Civil & Environmental Engineering, University of Alberta, Canada

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

Keywords: Mechanical, Electrical and Plumbing (MEP) system conflicts are complicated problems given the involved
MEP system multidisciplinary nature and constraint-bearing coordination. Although Building Information Modeling (BIM)
Pipeline planning shows benefit to resolving the conflicts before construction, the on-site pipeline inspection and further adjust­
AR
ment are inevitably necessary due to practical issues caused by deviations with plans. Given this, the research
Conflict resolution
Path coordination
aims at developing an on-site pipeline inspection and automatic coordination approach by using Augmented
Reality (AR) and a grid-based path-planning algorithm. The approach allows the site practitioners to compare the
newly added pipe layout plan with the existing pipelines and re-plan the pipe layout where encountered con­
flicts, further obtaining a solution without clashes. The method considers MEP system constraints in practice and
searches conflict-free pipe layout plans in near real-time. Furthermore, the path coordination process also
considers MEP design criteria, including its function, construction, and maintenance. The planned pipe layout is
displayed through AR to link the information from the virtual model to the real world. The performance tests are
conducted in virtual scenarios with different grid cell sizes and priorities of pipes to evaluate the planning results.
The test results show that under room and aisle space scenarios, the implemented approach can obtain path
solutions with approximately 10 s, and most of them comply with the pre-defined MEP design requirements.
Through this study, automatic pipe planning with constraints is proved to be feasible to provide an effective on-
site conflict resolution approach to improve the MEP construction process and future maintenance.

1. Introduction Consequently, workers need to resolve the conflicts of MEP systems on-
site in a situation of lacking the global picture of design intention
Mechanical, electrical and plumbing (MEP) engineering is one of the behind.
most important topics that directly affect construction efficiency, Recently, digital construction concepts, like the adoption of Building
building safety and energy usage [1]. With great importance, the design Information Modeling (BIM), help further improve the performance of
of building MEP systems nowadays faces the challenge of conducting MEP system conflict resolution [3,4]. Different MEP systems can be in­
effective conflict resolution processes, given its complicated coordina­ tegrated into BIM collaborative design platform for automatic conflict
tion nature. It is because MEP coordination requires controlling different detection, while inspection and coordination on-site are still needed due
components under different systems (e.g., electricity, water, HVAC) to practical issues.
within limited space and given functional considerations. In the con­ Firstly, the ineffective collision detection result is a problem during
ventional approach, MEP engineers sequentially overlay and compare the automatic collision detection process using BIM. BIM tools are
2D drawings from multiple systems to detect and eliminate spatial and dedicated to resolving conflicts during the design and modeling phases
functional conflicts. Such a process is time-consuming, error-prone, before the construction. However, Wang et al. [5] have indicated that
expensive, and may not satisfy the construction demand [2] and it is 78% of collision detection results may be ineffective, which is because
even worse when considering collaborative installation scheduling [3]. models originally designed by different MEP engineers often

* Corresponding author.
E-mail addresses: [email protected] (L.-T. Tsai), [email protected] (H.-L. Chi), [email protected] (T.-H. Wu), [email protected]
(S.-C. Kang).

https://doi.org/10.1016/j.autcon.2022.104400
Received 3 August 2021; Received in revised form 28 May 2022; Accepted 1 June 2022
Available online 14 June 2022
0926-5805/© 2022 Elsevier B.V. All rights reserved.
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

encountered difficulties in the integration. It requires considerable time 2. Literature review

to filter the effective results out manually when a massive amount of
ineffective clashes is detected. Besides, the manual modeling process is On-site MEP system conflict refers to the occurrence of overlapping,
time-consuming and does not compatible with the current time and cost collision, interference, or functional impediment that happened among
restriction [2]. In consequence, these ineffective collision detection re­ or across different MEP system layouts, causing troubles on the system
sults can only be solved in the construction field, and it would increase installation, operation, and maintenance at sites. They are essential tasks
the difficulty of the construction and maintenance process because of the required to be eliminated ahead at design, planning, or even early site
lack of an accurate layout plan to follow. work stages through further coordination, to save the expense and time
Secondly, deviations often occur when the installation process takes to avoid the late rework. Previous research works were conducted to
place after workers ideally layout the pipe routes from the pre-designed achieve the efficiency and effectiveness of conflict resolution. Together
2D drawings or 3D models. Because of the construction errors, the MEP with enabling technologies and methods, including AR and pipe path
systems may no longer fit properly during the installation process when planning methods, have been discussed individually in the following
there are deviations between structures or among the MEP layouts [5]. It sub-sections:
may lead to the delay because of re-design, layout, and report the MEP
model or drawing conflicts. As a consequence, the MEP original design 2.1. MEP system conflict resolution
has to be adjusted as well.
Lastly, the constructability of the MEP systems concerned by prac­ To overcome the present MEP conflict resolution challenges, previ­
titioners at sites is hard to be addressed at the design phase. It is the same ous researchers addressed them through three different aspects: design,
case as maintainability. Previous research pointed out that during the monitoring, and refactoring. Firstly, how to accurately design the MEP
pipe arrangement, visualization can help handle the detection and system is a core problem to avoid conflicts and increase workplace
response to collisions between pipes and obstacles [6]. However, con­ safety, quality, and efficiency. Korman et al. [1] captured the knowledge
struction, operations, and maintenance issues may not be easily of MEP design criteria: function, construction, and maintenance, from a
discovered in the 2D drawing or 3D models because of the uncertainties research project. They implemented this knowledge into a computer tool
implied comparing with plans. External MEP maintenance tools are to assist the MEP design process. Many researchers have also analyzed
needed to improve the inspection task and pipeline accessibility right the influences of CAD-based planning algorithms on MEP design [9–11].
before the actual construction [7,8]. Tserng et al. [9] provide a spatial planning algorithm that packages large
As the current MEP design is facing the three challenges mentioned and complex MEP systems into several smaller fabricated components to
above, practically the detailed MEP arrangement process still depends improve construction quality and productivity, reduce construction
on the decision of an on-site engineer or technician. Moreover, the cycle time, and minimize cost. Currently, researchers further took
systems may not satisfy the MEP functional requirements, affecting the advantage of BIM and heuristic optimization approaches, such as Par­
construction process afterward and increasing the difficulty of mainte­ ticle Swarm Optimization (PSO) [40], 3D A* [12] and Simulated
nance. These have been reported as essential issues, and enabling Annealing (SA) [3], to further develop automatic planning of pipe
technologies for further improvements of MEP coordination is installation layout, sequence as well as schedule [3]. Although these
demanded. research results provide references in practice, Lee and Kim [11] suggest
Given the challenges and potential enabling technologies, this that other factors may still affect coordination productivity, such as skill
research aims to develop an inspection and coordination approach using levels of the coordination participants, design complexity, the number of
automatic path-planning algorithms and AR visualization to resolve the change orders, and project delivery methods [11]. Furthermore, most of
current MEP conflict resolution issues. In advance of an MEP system the automatic pipe coordination results at the design stage could be
installation, the proposed approach assists the on-site engineers or further affected by the deviation in practice and demand a further re-
general contractors to compare the AR-visualized planned pipe models routing approach at sites when necessary.
with the as-built pipelines and further inspecting conflicts through Furthermore, researchers introduced monitoring approaches to cope
mobile devices at sites. If conflicts are detected, on-site coordination of with the identification of potential as-built deviations between design
the planned pipe paths, through automatic re-routing, can be carried out and actual work. The representative monitoring strategy is by collecting
for obtaining a set of path solutions without conflict. The path planning accurate field information through laser-scanning technologies. Such
algorithm adopts considerations of the function, construction, and technologies with automated methods to detect and recognize point
maintenance of the MEP system. Two research problems have been clouds, further used for comparing the as-built and as-planned status of
derived from this aim: the MEP system [13–17]. Bosché et al. [13] tested a Scan-vs-BIM
method. They identified it is challenging when tracking MEP systems
1. Are the on-site pipe planning results able to comply with the pre- constructed at sites, due to adjustments made on-site that lead to actual
defined design requirements of the current MEP coordination? component layouts varying with designed ones. Bosché et al. [13] thus
2. Does the processing time of the pipe re-routing planning, using AR presented an integrated method of Scan-to-BIM and Scan-vs-BIM, which
portable devices, satisfy the efficiency requirement of the on-site provides the basis for automated earned value tracking, automated
decision-making process in handling MEP conflicts? percent-built with as-planned measures, and assistance for the delivery
of as-built BIM models from as-designed ones. Kalasapudi et al. [15]
The proposed approach, bearing with the research questions, is further achieved a computationally efficient spatial-change-detection
implemented for further improving the MEP conflict resolution process. approach, which integrates nearest neighbor searching, and relational
Besides, a performance test is conducted on a mobile AR device to graph-based matching approaches. The above research results show the
evaluate the processing time and the related parameters with MEP co­ advance of laser-scanning solutions, however, these approaches implied
ordination. The results can provide a reference for future research to two main limitations. One is the allocation difficulty of scanners on
estimate the efficacy of on-site MEP coordination applications. The pa­ construction sites [18], and the other is still the considerable amount of
rameters, such as path-searching space, also help future research to computational time demanded to process point clouds if a large project
determine an adequate level of details to be considered in a practical is targeted [14].
pipe planning problem. It is expected that such an on-site conflict- Lastly, research is dedicated to refactoring the conventional process
resolving module can provide for future research to integrate with to increase the efficiency of MEP design and construction in practice. For
current BIM tools or automatic modeling techniques on real project instance, they developed frameworks based on the advantage of using
cases. BIM tools as a systematic information delivery format in real projects.

2
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Wang and Leite [19] proposed a formalized clash document schema, management, and construction. It has great potential to reduce the
which can support the management of MEP construction coordination. construction time and cost by increasing the accuracy and quality of
Furthermore, Wang et al. [5] developed a practical BIM framework for information delivery. Further research has also focused on using AR to
integrating the MEP layout from the preliminary design to the con­ enhance pipe inspection and maintenance processes [42,43,45]. The MR
struction stage with different levels of detail, which saved MEP costs that solutions in the market, such as Trimble XR10 with HoloLens 2 [44],
would have been caused by errors in a case study project. However, 78% further embed AR functions in safety helmets for workers to intuitively
of collisions by the automated clash detection functions through existing inspect the construction scene at sites. However, there are only a few
BIM tools were ineffective, which will cost numerous labor time to check cases using AR on MEP system deviation coordination, which could be
and identify them. Furthermore, Singh et al. [3,41], who developed a mainly because the input interface of the AR-embedded portable device
BIM-based automatic pipe installation sequencing and scheduling sys­ is not as complete as the general PC. Such an interface to be used in the
tem, also pointed out that different plans through 4D simulations still field is convenient, but carrying out the detailed coordination process
require human intelligence to identify the best schedule plan through may be tedious. In the MEP system coordination, it is possible to inte­
visualization aids. The research work, therefore, remarked that the grate the automatic planning approaches, such as path-planning algo­
emergence of AR technology would satisfy the goal of enhancing and rithms, to minimize user input and manipulation loading. It would thus
accelerating the process of verifying these deviations at sites. make the practitioners more comfortable to use in practice to achieve
To sum up, the limitations still exist in different methods while the purpose of on-site piping coordination.
addressing the MEP conflict resolution challenges. Among various
research efforts, it is found that fieldwork with deviation inspection and 2.3. Piping path planning
resolution are inevitably vital to produce accurate and efficient MEP
conflict resolution, and there is a trend coming up with an innovative The piping path-planning algorithm is developed to minimize user
visualization approach with the emergence of AR technologies. effort in MEP layout planning and coordination, and speed up the time-
consuming pipe design process to overcome its complexity. The devel­
2.2. Augmented reality opment of piping path planning can be divided into three periods: expert
system, automated design, and dynamic design.
With the support of AR technology, many researchers have overcome Design expert system integrates algorithms for routing problems to
complicated project challenges, mainly from three perspectives: speed up the piping design process, however, it remains a manual
communication, management, and construction. Firstly, AR visualiza­ modeling procedure. Kang et al. [27] developed a ship piping design
tion helps to reduce the complexity of different parties’ communication system based on the configuration knowledge of pipelines and
while keeping the necessary information enhanced. Dong et al. [20] Computer-Aided Design (CAD). By analyzing geometric conditions and
developed a tabletop AR system, ARVita, and accommodated the need assisting decision-making in the process of manual design, it can
for a collaboratively model visualization tool that allows multiple users significantly reduce the time required for the design. However, it can
to observe and interact with dynamic visual simulations of construction only provide the expected result on the geometric model of the main
processes. Furthermore, Lin et al. [21] developed an AR multi-screen piping systems. If the objective function is expanded for including
system, which can significantly reduce the discussion time of project maintainability and available space around pipes in addition to pipe
data findings by 55% and problem prediction by 66%, compared with length and number of bends, it will be challenging to solve such a multi-
the conventional paper-based method. objective function.
AR also helps the project management by providing the managerial To solve such a practical challenge, it accommodates the need for an
information directly on the building entity [7,22,23,43,45]. To illustrate automated design system to support interactive planning and cover
the 3D models at correct positions on actual site building, Bae et al. [22] comprehensive aspects of a piping route path through conceptual design
presented a vision-based mobile AR system. It successfully localized the to a detailed one. Zhu and Latombe [28] described an approach for
user solely based on the on-site photographs taken from mobile devices, designing pipe layouts automatically using a cell decomposition method
without using any external location tracking modules. Also, Meža et al. based on robot path planning techniques. The research also makes the
[23] found out that the use of AR could be significantly beneficial in a approach flexible in order to make it capable of treating a variety of
proposed AR progress management system, which was tested in an other constraints that are typical in practical pipe layout design prob­
actual construction project. Park et al. [7] developed an AR defect lems. Ito [29] proposed the definition of genes to deal with pipe routes.
management system that allows users to easily recognize the deviation The research work used the concept of spatial potential energy: a
between the as-built building and its as-designed one in a mock-up method to generate initial individuals for genetic algorithm (GA) opti­
model test. However, its efficiency and the effectiveness of the imple­ mization to search the pipe path automatically. Park and Storch [30]
mentation in real projects still need to be verified by practical experi­ presented an automatic pipe-routing algorithm accommodating all
ments [22]. major detailed design facets, and a cell-generation method is developed
Lastly, recent research also uses AR during the project construction which satisfies geometric constraints. Sui and Niu [31] proposed a
process. Such AR applications can support the worker-related spatial modified GA-based approach to solving the complex combinatorial
perception abilities to increase the quality of work or to satisfy some branch-pipe-routing problem, which plays fundamental and critical
unique design that is more complicated [24]. Fazel and Izadi [25] pre­ roles in ship-engine design. Liu and Jiao [32] utilized multi-objective
sented an interactive multi-marker AR tool for constructing free-form optimization for pipe routing and vibration problems based on Kriging
modular surfaces. The cost of the system is 5 to 10 times less than model and non-dominated sorting genetic algorithm-II (NSGA-II). Singh
another standard system while keeping the accuracy and construction et al. [3] further compared 3D A* and Fruit fly Optimization Algorithm
time insignificantly different. Furthermore, Chalhoub et al. [26] (FOA) for multiple pipe system layout designs and proved that A*
compared the task performance using a Mixed Reality (MR) approach particularly performed well as the baseline Dijkstra algorithm but saved
and the traditional 2D paper, while doing electrical conduit construc­ more time. Given these, researchers have been committed to exploring
tion. The result shows that MR, as a superset approach of AR, can reduce the possibility of different algorithms, and resulting path planning ap­
the total number of worker mistakes by 75%. More importantly, it proaches helped support the automated MEP design. In practice, though,
reduced the amount of rework required by 72% than that of using the 2D piping engineers still need to manually set and retry the boundary
paper, which shows the potential of using AR-related applications on a condition and other functional parameters to achieve the project goals
complicated MEP system [26]. when doing the design with a certain level of automation.
To date, AR technology has been applied in project communication, Compared to barely automated design, researchers integrate piping

3
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

path-planning algorithms with existing CAD and BIM software coordination knowledge mainly contains three aspects, which are:
[3,4,33,34], which further created dynamic design systems. They are function, construction, and maintenance (See Fig. 1).
ones with rapidly recursive input and decision-making processes. The function features of MEP systems include type, length, slope,
Asmara and Nienhuis [35] use a deterministic technique to find the cross-section, clearance, insulation, material, and support. These fea­
optimum route. They further used a non-deterministic one to select tures are prescribed by design codes and standards to assure MEP sys­
parameters and form a link to interface with CAD software. Kim et al. tems satisfy performance requirements for specific objectives. A design
[36] developed a practical pipe auto-routing system in a shipbuilding engineer defines the function of each MEP system or component during
CAD environment using network optimization. The result shows that the the design phase. The geometric-related features, which are system type,
system cannot always produce the best pipe route initially or require length, slope, cross-section, and clearance, are considered because they
further decision-making based on offered plans [3]. However, the directly affect the path results of the planning. As for insulation, mate­
designer can input new parameter sets easily and quickly until obtaining rial, and support, which may indirectly affect the pipe path design, are
a satisfactory result in the CAD or BIM environment. out of the scope of this study.
With the current development of pipe path planning algorithms, The construction features contain the fabrication, installation space,
users can get benefits from a few geometries and material parameters and installation sequence during the construction process. They are
input with automatic design processes for solving complex and time- crucial because the resulting constructability determines a practical
consuming piping problems. Researchers also developed the interfaces design for a feasible field operation. Lastly, the maintenance features
linked to existing CAD software as dynamic design systems to make the contain space accessibility and frequency. MEP system maintenance
user easily adjust and fill the gaps between the automated pipe planning must be concerned for the project lifecycle after construction, to reduce
and practical concerns. the derived cost and repairment difficulty during the operation. The
ideal routes should be considered in an intersection area of sufficient
2.4. Research gaps and needs function, construction, and maintenance abilities.

To sum up the review, many researchers have proposed different 3.2. Planning
approaches for MEP system conflict resolution to produce an accurate
and efficient MEP system design and coordination. However, it is still For the path planning module, we propose an automatic pipe plan­
critical to deal with the deviations between reality and MEP design. The ning method to model the multidiscipline MEP system knowledge. The
resolution needs domain knowledge among different MEP systems, such method consists of three essential considerations: pipe priority sorting,
as abilities to construct, function, operate, and maintain, to have a constraint embedding, and MEP A* planner integration. Fig. 2 shows the
reasonable solution for multiple pipe coordination. These have been process of the proposed planning method. The further algorithm, named
seldomly discussed with a systematic organization. Furthermore, the PlanningModule, can be seen in Table 1. It calculates the paths of
trend of AR usage in the field is coming up to support communication, conflicted pipes that are considered by users. Firstly, it sorts out the
management, and construction. Despite that previous research has order (priority) of the considered pipe sets according to their MEP sys­
proved the advantages of using AR to improve the inspection phases of tem attributes. These can be specified by the users at sites or ahead at
MEP system conflict resolution, little research has moved on to the MEP design stages. Once the order is determined, the method starts planning
detailed coordination process in the field. The reason may be the AR- each path of the pipes. It further considers MEP design criteria, function,
embedded user interface on the portable device is not as mature as the construction, and maintenance, for each pipe and treats these criteria as
one on the stationary computer. Using AR to perform the MEP detailed “constraints.” Lastly, the method transfers these constraints into
coordination process, which composites with several MEP systems and weightings for a grid-based A* planner to get the shortest path with
complicated routing tasks, may be tedious. There is rare research using minimized cost. The resulting paths connected and complied with the
AR to solve the problems of the MEP system coordination, especially for constraints. The planning process for each pipe is sequentially separated
on-site piping plan modification, because of limitations on its input but all affected by the previous pipe planning results through an accu­
interface. With the support of automatic path planning algorithms, users mulated set of constraints. It means that after the planner outputs a set of
may design the piping path with a few inputs, which has the potential to paths with higher priorities, the space occupancy and constraints will
make the routing decision at sites. narrow down the solution for following pipes with lower priorities. The
Given these considerations, this research proposes an AR-based on- method then loops the path planning until the last pipe path has been
site piping coordination approach, which integrates the AR and a grid- solved.
based path planning algorithm to overcome the existing MEP coordi­
nation problems. This system supports users to intuitively inspect MEP 3.2.1. Priority sorting
system deviations with the advantages of visualization using AR. And it As shown in Fig. 2, the priority sorting is the initialization to prior­
helps further re-plan the pip layout through a planner with only a few itize each conflicted pipe, further obtaining an ordered list. Korman
inputs required in the field. The following section describes the details of et al. [1] summarize the priority of resolving conflicted systems, and
the proposed approach. each attribute can infer a positive or negative relationship to the
resolving order based on the feature value. Derived from this research
3. Method and practitioners’ feedback, Table 2 shows the priority details of MEP
features. The installation and access spaces, access frequency, length,
The proposed on-site AR pipe conflict resolution approach contains and cross-section represent positive relationships to the priority: the
two parts: 1) a path planning module and 2) a system for the interaction. bigger one first. In contrast, the install sequence represents a negative
The path planner integrates the current MEP design knowledge into an relationship to the priority: the latter one first. The first pipe after the
A* planning algorithm. And the system considers the interaction be­ sorting has the highest priority when resolving its conflict with others.
tween the end-user and the functions for the AR display. The routing of this pipe would not be affected by any other pipes. By
contrast, the last pipe is with the lowest priority, and the most stringent
3.1. MEP knowledge constraint will be applied when finding the routing plan.

The approach is developed based on the MEP knowledge, which is 3.2.2. Constraints
collected from previous related research [1,37], to make the planning Constraints, derived from the MEP design features, help determine
results meet the multidiscipline design criteria. Current MEP the practical concerns of pipe coordination at sites. In this research, five

4
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 1. Design features of MEP knowledge.

Fig. 2. The process of planning the paths for multiple conflicted pipes.

types of pipe constraints are adopted: slope, turning, elevation, relation, the gravity-driven system, because the vertical direction variation drives
and proximity. As shown in Table 3, each design feature can be trans­ the liquid inside to flow, further affecting its function.
ferred to at least one type of constraint among the five as a checking The turning constraint is to make sure the pipe turnings, such as
point to examine whether the path complies with the MEP design rules. elbow components, meet the design criteria. As shown in Table 4 (F.2),
A slope constraint defines a limitation of vertical direction variation the turning constraint contains three parameters: the angle of the elbow
in the horizontal direction. Each slope constraint contains a boundary component, θ, the radius, R, and the diameter, D. When the path en­
value based on the design requirements of different types of pipes. As counters a turning elbow, the two equations in Table 4 (E.2) check
shown in Table 4 (F.1), Δy represents the vertical variation between the whether this turning component violates two conditions. The first one is
start point, s and the end point, e . Δh denotes the horizontal one be­
⇀ ⇀ whether the angle θ is larger than the maximum angle set in the design
tween them, which is the root mean square of the two horizontal vari­ criteria, and the second one is whether the ratio of R divided by D is
ations, Δx and Δz. Such constraint is used to limit the pipe slope, as in larger than the maximum pre-defined. If the checking returns a viola­
Table 4 (E.1). If the slope, which is the Δy value divided by Δh, is smaller tion, there will be a penalty value added during the path selection. Also,
to ensure the cost and the construction difficulty are under an acceptable
than the pre-defined boundary value, the pipe (from s to e ) does not
⇀ ⇀
range, the number of elbow components should be as small as possible.
fulfill the design requirement. Thus a penalty value (extra cost) will be
Thus, a penalty value is applied where the path is turning. This
given during the routing process. This constraint is applied especially on

5
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Table 1
Algorithm: Planning Module.

total clear space above the clear height requirement and cproper is the
Table 2
proper clearance. For the functional purposes mentioned above, the
Priority sorting rules considering different MEP design features.
constraint help to check if the clear space of the pipe is less than the
Feature Description maximum, keeping the pipe from placing under the clear height. If c is
● Firing systems have the highest priority far from cproper, a penalty value is added on the neighboring nodes of the
System ● Gravity-driven systems have a higher priority candidate path node. This constraint can optimize the vertical usage for
than pressurized systems
specific MEP systems. For example, it is better to have such a constraint
Function ● The pipe with longer length from start to end
Length
point has higher priority on the firing system to keep its elevation low and close to the clear
Cross- ● The pipe with larger cross-section has higher height, because the firing spryer should not be placed on top of any other
Section priority system. The pipe at the first installation place has an elevation constraint
● The pipe which requires larger installation space to place just below the ceiling to make other subsequential pipes be
Space
has higher priority
Construction
● The pipe at the prior install sequence has higher
easily installed. While for the pipe with the highest access frequency, it
Sequence should be placed just above the clear height so that this pipe can be
priority
● The pipe which requires larger access space has easily accessed.
Space
Maintenance
higher priority The relation constraint is to guarantee a specific type of pipe not
● The pipe with higher access frequency has higher
Frequency being placed above or below another specific one. As shown in Table 4
priority
(F.4), the start and end points of one pipe (pipe A) are marked as s and e ;
⇀ ⇀

′ ′

the start and end points of another one (pipe B) are marked as s and e .
⇀ ⇀
constraint is necessary, especially for MEP pressurized systems. The ′ ′
elbow component would reduce the effectiveness of the pressurized pipe To constraint s and e not placing above s and e , the equations as
⇀ ⇀ ⇀ ⇀

and be at risk if the pressure at the elbow exceeds the design limits to ⇀
Table 4 (E.4) are determined. The direction, d, equals to (0,1,0). The
cause irreversible damage to the pipe. ⇀
The elevation constraint limits pipe paths to be designed above a decisive point, i˙ , is the intersect point of pipe B with the plane deter­
clear height. Clear height is a common space requirement that makes ⇀ ⇀
mined by pipe A and direction d. To solve point i˙ , we first solve the x
sure the space under the pipe is sufficient for functional usage. The
and z components of it by the linear equation of the two pipes and then
elevation constraint specifies the preferred height of each pipe to opti­ ⇀
⇀
mize the usage of vertical space. As shown in Table 4 (F.3), c represents solve the y component. If the dot result of d with the vector from s to i˙ is
⇀

the clear space below the ceiling and above the pipe presence, cmax is the larger than zero, pipe B is above pipe A and violates the constraint.

Table 3
Design features considered by different constraints.
Feature Constraint

Slope Turning Elevation Relation Proximity

System ● ● ●
Function Slope ●
Clearance ●
Space ●
Construct
Sequence ● ●
Space ●
Maintenance
Frequency ● ●

6
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Table 4
Overview of the constraints.
Constraint Illustrative Figure Equation Example

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Δh = Δx2 + Δz2
Δy
Slope > slopemin ● Gravity-driven systems to keep the flowing fluid achieve the design criteria
Δh
(E.1)

θ < θmax
R
Turning > ratiomin ● The number of turning on pressurized systems should be minimized
D
(E.2)

c < cmax
● Some components on firing systems have elevation preference to place just
Elevation c ≃ cproper
above the clear height
(E.3)

a1 = ez − sz
a2 = ez′ − sz′
b1 = sx − ex
b2 = ex′ − sx′
c1 = a1sx + b1sz
c2 = a2sx′ + b2sz′
⃒ ⃒
⃒ a b1 ⃒
d = ⃒⃒ 1 ⃒
a2 b2 ⃒
⃒ ⃒
⃒ c1 b 1 ⃒
⃒ ⃒
⃒ c2 b 2 ⃒ ● Electric systems to add restrictions to other water systems above it
Relation ix = ● Restrictions to be added on pipes with later install sequence
d
⃒
⃒ a1 c1 ⃒
⃒ ● Restrictions to be added on pipes with the lower access frequency
⃒ ⃒
⃒ a2 c2 ⃒
iz =
d
( )
iy = sy + ey − sy *
ix − s x
ex − sx
(⇀ ) ⇀
i − s •d<0
⇀

(E.4)
v = s− e
⇀ ⇀ ⇀

⇀ ⇀ ⇀
v =s − e
′ ′ ′

⇀
c ross = v × v
⇀ ⇀ ′

c ross
⇀
● Restrictions to be added on other pipes within the install space
n =
⇀
Proximity
‖ c ross‖
⇀
● Restrictions to be added on other pipes within the access space
⃦⇀⃦ ⇀
⃦ ⃦
⃦d⃦ = (s − s ) • n
′ ⇀ ⇀

⃦⇀⃦〈
⃦ ⃦
⃦d⃦ r
(E.5)

Relation constraint is an essential one for specific MEP systems. For order. It is to make sure that the prior pipe would not (or minimally)
instance, an electrical system should be born with the relation constraint affect the installation of subsequent pipes. Another example is the pipe
to all other water systems above it to prevent the functionality problems with a higher access frequency has the relation constraint to other pipes
due to leaking. Relatively, a water system should also take care of the with lower access frequency below it to assure that the operation and
relation constraint to other electrical systems below it. Moreover, the maintenance would not be affected.
relation constraint is also important from the aspects of construction and The last constraint is a proximity constraint, which is to keep a
maintenance. The pipe at a prior installation order should consider the specific pipe in the distance away from other specific pipes. As shown in
relation constraint to other pipes above it or those at later installation

7
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Table 4 (F.5), the start and end points of pipe A are marked as s and e ;
⇀ ⇀ following subsequent pipe planning.
′ ′

those of the pipe B are marked as s and e . Firstly we need to cross-

⇀ ⇀

′
3.2.3. MEP A* planner
product the v and v , which are the two direction vectors of the two
⇀ ⇀
We propose the MEP system layout planner through an MEP A* al­
pipes, to get the cross vector, c ross. This vector points along the shortest
⇀ gorithm to search the shortest path in a short time. The original A* al­
gorithm is designed to process a pathfinding problem [38]. It is similar
line from pipe A to pipe B. After normalizing c ross, we get the unit
⇀
⃦⇀⃦ to Dijkstra’s algorithm but applies heuristics, which obtains a solution
⃦ ⃦
vector, n. Lastly, the distance between the two pipes, ⃦d⃦, can be solved more efficiently.
⇀

′ ⃦⇀⃦
⃦ ⃦ The MEP A* algorithm is derived from the original one. As shown in
by dot product the vector from s to s and n. If ⃦d⃦ is less than the
⇀ ⇀ ⇀
Table 5, MEP A* algorithm declares a closeSet and an openSet to store the
proximity requirement r, pipe B is too close to pipe A as a violation. node depending on whether it is discovered or being considered. The
Proximity constraints should be considered in most systems. For current node is the node with the lowest f-cost in the openSet of each loop.
instance, there should be a proximity constraint between heating water Also, same as A*, we define g-cost as the distance from the start node
supply and others. From the construction perspective, all pipes should along to the current node. f-cost is the distance from the current node to
contain proximity constraints to any other ones to keep in the distance the target node. Then the algorithm continuously discovers the neighbor
larger enough for accessibility, in which the worker can easily perform node from the start point and selects the node with the lowest f-cost as
the installation. The constraints with similar ideas for maintenance the current node until reaching the target point. The main difference
purposes are necessary, too. between the proposed MEP A* and the original A* algorithm is that we
As shown in Table 1, all constraints listed are considered in an MEP have steps (at lines 15–16) to obtain penalty, influenced by constraints
A* planner with a specific order. For slope, turning, and elevation, they of pipes, to the f-cost.
are considered at the beginning of all pipe planning. It is because the As in an indoor building environment, we remaster the path
user or the design codes determine the configurations of these con­ searching space into a grid in the algorithm. The search initialization
straints in advance. By contrast, for relation and proximity, these con­ consists of two steps: nodes creation and collision detection with ob­
straints cannot be considered until the planning path is preliminarily stacles. Firstly, nodes creation is to treat individual grid cells as constant-
decided, because these consider the suitability of nodes on the path. sized nodes. The number of nodes is equal to the multiplication result of
After the planner determines the candidate path of a pipe, the planning the length, width, and height divided by the diameter of pipes. Each
module assigns corresponding relation and proximity constraints based node we specify its x, y, and z indices in the grid for speeding up the
on the path nodes and other parameters pre-determined, such as clear­ process to find its neighbor nodes. The second step is to perform collision
ance of installation and maintenance. Then these two constraints will be detection to identify if the existing grid cells are occupied by the existing
accumulated into a constraint list and will affect the consideration of the obstacles. If there is an obstacle overlapped with a particular grid cell

Table 5
Algorithm: MEP A* planner.

8
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

(node), we mark it as a non-walkable cell (node). After the two steps, the 3.3. Interaction
environment is ready for pipe path searching.
The five constraints, turning, slope, elevation, relation, and prox­ To deal with on-site pipe conflict resolution, we have designed an
imity, will be checked whenever the MEP A* planner explores a AR-based mobile system for the interaction, as shown in Fig. 4. The
neighbor node from the current node during the sequential planning system connects the proposed planning module as a processing core. It
process in the grid. If the neighbor node violates a constraint, the allows the user to perform the following tasks: space ranging, model
planner adds a penalty value to the f-cost of the neighbor node. Fig. 3 inspection, conflicted pipes selection, constraint setting, and priority
shows the original path generated by the A* algorithm and the refined setting.
results after applying the five constraints to the planner. To synchronize the user location and virtual scene for the AR display,
Firstly, due to the turning constraint, it reduces the turnings of the we need to obtain the user location related to a known position through
path from six to one only, which is much more reasonable and cost- space ranging. During such a task, the user has to align the mobile device
saving. Secondly, if the pipe has the slope constraint, it evaluates the with a known tag (mark) in the real world. Then the system aligns the
slope, and then penalties to the horizontal neighbor nodes are added to virtual view to the tag position to complete the localization. It is a crucial
make the slope larger to fit the requirement. Thirdly, if the neighbor step because the deviation between the real world and the virtual one
node is under clear height, the planner adds penalties to yellow cells, as would directly affect the effectiveness of the following works. After
shown in the top-right hand of Fig. 3. It makes the path preferably pass localization, the system displays the virtual information (in this case, a
over clear height. As for the cases of relation and proximity constraints BIM MEP model) to the user.
in Fig. 3, the triangle notation represented other existing pipes. If the The second task is to inspect the existing BIM model to find out if
neighbor node is in the affection area of constraints, which are the there are any pipe conflicts. After determining the conflicted pipes, the
yellow cells shown, the penalty will be added to increase the difficulty user can start setting the input for the conflict resolution through pipes
for the path passing over. After applying the five constraints to the A* selection, constraint setting, and priority setting. The user selects the
algorithm, we can obtain paths considered of different MEP system particular conflicted pipes and changes their constraint settings,
design requirements. including whether or not to add specific constraints with corresponding
Furthermore, the penalty value of each type of constraint is different parameters, if the original setting is inappropriate. Then the user can
from each other. Some of the penalty values are relatively higher if the specify a priority sorting approach whenever it is suitable to fit the
MEP design code has a stringent requirement, making the penalty values design requirements. Also, the user can have an inclusive priority
of other constraints are lower. For instance, the penalty values of slope assignment considering other non-conflicted pipes to increase the pos­
and elevation constraints are higher than others because the two con­ sibility of finding an ideal solution. Finally, the MEP A* planner de­
straints are based on defined values in the design code. In contrast, the termines the paths of pipes and display the result for another round of
penalty values of relation and proximity constraints are less while they model inspection. The whole interaction is a recursive process to allow
are still affecting other planning paths. The turning constraint has the the user to interact with the AR system and try out an optimized solution
lowest penalty value in general because it only affects the cost of design. for solving the conflict issues.
However, for pressurized systems, the turning constraint may have a
high penalty value to ensure the pressure of the liquid inside the system 4. Implementation
complies with safety requirements.
This section explains the architecture and functions of the

Fig. 3. Cases of MEP A* algorithm considering different types of constraints.

9
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 4. The process of the interaction between the user and the proposed AR system.

implemented conflict resolution system. The system is based on a visu­ and intuitive function for pipeline coordination.
alization platform, Unity3D, to link the virtual models from a BIM
software, Revit, and connects to mobile devices for AR display by using a
toolkit, ARKit. To accomplish the conflict resolution mission, we 4.2. System architecture
developed the functions, including localization, model inspection, and
coordination. Following the system requirements above, the system architecture
can be seen in Fig. 5. It is mainly built on the Unity3D platform, which is
a game engine with compatibility on various 3D model formats and off-
4.1. System requirement the-shelf functions for the interactions. The MEP BIM model is thus easy
to be imported from Autodesk Revit, BIM design software, to Unity 3D to
Before the actual system implementation, we interviewed general establish the virtual building environment. To update the user’s location
managers and engineers of Continental Engineer Corporation to deter­ and superimpose the virtual objects with real building components, we
mine the system requirements for on-site MEP system conflict resolu­ use ARKit as a plugin on top of the Unity3D platform. ARKit encapsu­
tion. In practice, the construction company or the general contractor lated a concept called Visual Inertial Odometry (VIO) [39] and was
conducts the MEP system conflict resolution through the Combined developed by Apple Inc. It receives the camera image and the IMU
Service Drawing (CSD) and Structure, Electric and Mechanic integration sensor data from the iOS mobile device and extrapolates its position as
(SEM) process in BIM software after the design review. And current output. To deal with the AR positioning accumulative error over time, it
practice still needs to determine the MEP system conflict solution and calculates the different camera images from a period to get the feature
update the BIM model from the information in the field. Therefore, the points to calibrate the device location estimated from the IMU sensor.
system requirements are fourfold: clear BIM model input, accurate on- Finally, we implement the proposed conflict resolution system on top. It
site model inspection, user-friendly interface for conflict resolution, determines the conflicted pipes, along with constraint and priority

Fig. 5. System architecture of the conflict resolution system.

10
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

settings as input from users, and output the resolution results through length of pipe layout, and total elbows consumption. These planning
the AR visualization. The connectivities and further functionalities of results are compared under different sizes of the grid cells and sorting
the system are programmed as scripts using C# and further deployed as orders. The cases are categorized into six groups based on different cell
an iOS app for the on-site usage. sizes: 0.1, 0.15, 0.2, 0.25, 0.3 and 0.35 m. Also, there are six different
sorting considerations, which are those determined by pipe length,
4.3. System demonstration cross-section, access space, access frequency, install space, and install
sequence.
Fig. 6 shows a use case of the implemented system. Snapshots
through a mobile device are presented as Fig. 7. The case is a simple 5.1. Test settings
pipeline routing problem in the general office of a university building.
The user has to obtain a solution without conflicts. The first task is the The performance tests are conducted on iPhone 6S. The mobile de­
localization for determining where to start the AR function and model vice is equipped with an Apple A9 CPU, which is a 64-bit 1.85 GHz ARM-
inspection. Fig. 7 (a) shows the complete AR building model from based system. The A9 in the iPhone 6S has 2 GB of LPDDR4 RAM. It also
Autodesk Revit. After the user chooses the current floor, the system embeds a motion coprocessor, an accelerometer, a gyroscope, and a
displays only the model of the selection, as Fig. 7 (b), for the user to pick compass. The camera on the device can perform video recording with a
a specific start point. As shown in Fig. 7 (c), the user needs to align the 720p resolution at 30 fps. The minimum configuration for running such
virtual frame with the real tag of the start point and confirms by tapping AR functions requires the A9 processor hardware and iOS 11 software. It
the “localize” button. Then the localization is complete, and the virtual will get better performance on the state-of-the-art A12 processor and iOS
model is augmented on the actual building elements, as shown in Fig. 7 15 or any later configurations, including potential Android solutions
(d). with equivalent modules. To a better illustration, the following scenarios
Next, the model inspection is performed, the user can check the for the performance test are shown in a virtual environment instead of
current model and configure the input for the planning module, an AR scene for a clearer presentation.
including conflicted pipe selection, constraint, and priority settings. The first test scenario is a general indoor room space with a size of 13
After selecting the considered or conflicted pipe, which is the yellow m in length, 10 m in width, and 4 m in height. As shown in Fig. 8, there
pipe of this case, Fig. 7 (e) to Fig. 7 (g) show the interfaces for further are eight columns with a 0.9 by 0.9 m cross-section and four walls
configurations. Firstly, the user can view the property of the selected around the room. Also, there are two beams with a 0.4 by 0.6 m cross-
pipe, as Fig. 7 (e). Afterward, the user can determine whether to set the section. They placed on the long edges over the room space, which
detailed setting of each constraint, as Fig. 7 (f). Before starting re- would block the path of pipes in the scene. Moreover, considering the
planning the conflicted pipe, the user can specify which sorting con­ room for human activities, we set a clear height at 2.8 m. Therefore, the
siderations to apply in the planning module. The user can either select proper height for all pipes should be around 2.8 to 4 m above the floor.
all considerations by default to obtain complete results, as Fig. 7 (g). Finally, the planning targets are twelve pipes: six firing pipes, two
Finally, as the resulting path of the yellow pipe is shown in Fig. 7 (h), the electrical pipes, two pressurized water pipes, and two gravity-driven
user can thus view the planning results based on each consideration and water pipes.
determine the best one as the final solution. The detailed demonstration The second test scenario is an S-shape aisle (as shown in Fig. 9), with
is captured as an attached video in Appendix 1 for further reference. a size of 13 m in length, 10 m in width, and 4 m in height. There are also
eight 0.9 by 0.9 m columns in the enclosed space. Two walls splitting the
5. Evaluation area into the S-shape aisle are the main obstacles to this scenario. We
also set a clear height at 2.8 m. Therefore, the proper heights for all pipes
To validate the performance and feasibility of the MEP A* planner in are around 2.8 to 4 m. Finally, the planning targets are 13 pipes,
the proposed system, we conducted tests under two virtual scenarios. including five firing pipes, four electrical pipes, and four pressurized
The goal of the tests is to evaluate the planner through different aspects, water pipes.
including processing time needed to plan the pipes, feasibility, total For the firing pipes under the two scenarios, there are turning

Relocation
Localization
Information Retrieval
User selects
current space Model Review
Conflicted pipe
identification

Path Planning
Localization
with tags
Model review
through AR
Pipe priority and
constraint setting
and inspection
Manual Pipe model creation from
calibration start point to end point with
inputs

Repeat

Fig. 6. The use case of the proposed system.

11
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 7. The implementation result of the MEP conflict resolution process.

constraints applied to reduce the number of elbow components. Because in Fig. 10. The trend of time consumption is rapidly reducing along with
the sprayers should not be placed over any other pipes to assure the the increase of grid cell size (diameter). The average time for planning a
spraying function, also on firing pipes, there are elevation constraints to single pipe in the grid with 0.1 m cell size is 28 s, which means users
keep the pipes near to the clear height, and relation constraints that have to wait for nearly 2 min and 48 s to obtain all six types of pipe
affect all pipes below them. Besides, proximity constraints are applied to results with different sorting orders. For other groups with cell sizes
keep the clearance for pipe accessibility and installation. For the elec­ <0.15 m, the average time consumed to obtain a planning result is <7 s.
trical pipes, there are also turning and proximity constraints to meet the The feasibility of planning under the room space is shown in Table 6.
design requirement. However, the relation constraints only affect other We classified the levels of feasibility into three degrees. The highest
water pipes above them that are not classified as the electrical system. degree, marked solid circle, is used if all pipes have found their paths
For the pressurized pipes, there are turning and proximity constraints as and complied with all constraints that have been set in advance. The
well, and relation constraints affecting electrical pipes below them. middle degree, marked hollow circle, means all pipes have found their
Finally, for the gravity-driven pipes, most constraints are the same as the paths, but some paths violated the constraints. The lowerest degree,
pressurized pipes, but there are slope constraints to ensure their marked dash, means that some pipes cannot find the paths, which failed
functionality. in avoiding conflicts. Under the room space, most of the solutions were
found and met all constraints. However, in some groups sorted by
5.2. Test results – Room space considering pipe length and access frequency, some paths violated the
relation constraints. In the group under 0.3 m of cell size and sorted
The average pipe planning times under the room scenario are shown according to access frequency, the result showed that the planner could

12
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 8. Test scenario 1: room space.

Fig. 9. Test scenario 2: aisle space.

not find solutions for pressurized pipes. Table 8. The groups sorted by considering cross-section consumed the
To evaluate the cost of the planning result, we sum up the total least elbows and had the least standard deviation under different sorting
length of pipes along the corresponding planned paths. Table 7 shows considerations. Also, the groups with 0.1 m of cell size average have the
the total pipe length under different groups considering different sorting highest elbow consumption.
priorities in room space. The average results are disturbed from 138.3 to
142.3 m, and there is little difference in the groups with different grid 5.3. Test results – Aisle space
cell sizes. In most of the groups, the results sorted by the cross-section
consideration are with the shortest paths compared to other groups The average planning time results in the aisle scenario are shown in
under the same grid cell size. Fig. 11. Among all groups, similarly, the computational time is rapidly
To evaluate the functionality of the planning results, we sum up the decreased along with the increasing cell size. For groups with grid cell
total number of elbows along pipes under different groups. The average sizes <0.15 m, the average time to obtain a result for a single pipe is
elbow consumption in room space is between 38.33 and 50, as shown in <10.05 s.

13
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 10. Planning time under different grid cell sizes of the room space.

paths.
Table 6
Table 10 shows the total pipe length consumption under configura­
Feasibility of planning under different pipe priorities and cell sizes in room
tions in the aisle space. They are disturbed from 254.2 to 263.2 m. The
space.
shortest paths under the different groups vary. It is hard to tell which
Grid cell size (m) 0.1 0.15 0.2 0.25 0.3 0.35
sorting consideration suits the most. Though, the planning results with
Length ● ○ ● ○ ○ ○ considering install space returned the longest paths in all groups.
Cross Section ● ● ● ● ● ● The average elbow consumption in the aisle space is disturbed from
Access Space
127.3 to 136.5 m, as shown in Table 11. The groups sorted by consid­
● ● ● ● ● ●
Access Frequency ● ● ● ○ – ○

Install Space ● ● ● ● ● ● ering install sequence have come out with the least elbows required for
Install Sequence ● ● ● ● ● ● the paths but with some constraint violations, as shown in Table 9. The
second best sorting approach is to consider pipe cross-section.
●: All pipes have paths and obey constraints, ○: Some pipes violet constraints, − :
Some pipes cannot find paths

5.4. Discussion
Table 7
Total pipe length under different pipe priorities and cell sizes in room space. This section describes the discussion regarding test results, including
computational time, planning feasibility, proper priority settings, and
Grid cell size (m) 0.1 0.15 0.2 0.25 0.3 0.35
appropriateness of the planner configurations under the two test sce­
Length 140.6 139.65 140.6 137.75 137.1 140.0
narios. Firstly, in both scenarios, the smaller size of the grid cell returns
Cross Section 139.6* 137.55* 140.6 136.25* 134.7* 137.9*
Access Space 144.8 138.15 140.2* 140.75 139.5 144.9
the result with higher accuracy but requires more computational time.
Access The computational times under the 0.1-m-cell-size groups in both sce­
142.2 140.85 140.6 138.25 140.0
narios are averagely 28 and 45 s per pipe, which may be too long for the
–
Frequency
Install Space 142.6 140.85 141.0 139.25 138.9 142.1 on-site decision-making process. On the contrary, the times required for
Install Sequence 144.0 140.55 141.8 140.75 141.3 143.5
groups with the grid cell size larger than 0.15 m are <10.05 s per pipe,
Average 142.3 139.6 140.8 138.8 138.3 141.4
which means in both scenarios, the user needs to wait less than one
*
The shortest result in the groups with the same cell size minute to obtain the results and are thus more preferable.
Regarding the planning feasibility, there are pipe plans that violated
the proximity or the relation constraints in both scenarios. Usually, they
Table 8
occurred in the 0.25, 0.3, and 0.35-m-cell-size groups. As shown in
Total elbows needed under different pipe priorities and cell sizes in room space.
Fig. 12 (a), one violation is in the 0.35-m-cell-size group sorted by
Grid cell size (m) 0.1 0.15 0.2 0.25 0.3 0.35
considering pipe length. The highlighted node, which is a node of the
Length 36* 41 37 30 31 35 path of the yellow pressurize pipe, violates the relation constraint of the
Cross Section 35 28* 35* 25* 24* 27* red firing pipe, which limits the path of the yellow one to go above. And
Access Space 70 32 37 55 55 54
Access Frequency 43 42 35* 33 34 39
it is because of the large cell size making there is no solution for the
Install Space 53 46 42 47 45 46 pressurized pipe to pass over from the top of the firing pipe. This could
Install Sequence 63 45 44 57 54 52 affect the function of the firing sprayer, but on the contrary, reduce the
Average 50 39 38.33 41.17 40.5 42.17 total pipe length consumption.
*
The result using the least elbows under the same cell size groups The reasons that the planner cannot find the solutions for some
groups are usually because of pipeline-blocking issues. These happen
The feasibility of planning under the aisle space is shown in Table 9. because the cell size is designed too large, and the planner overlooks
In most cases, paths are found and comply with all constraints. However, potential clearance in-between pipes given their actual sizes. Fig. 12 (b)
in the cases sorted by install sequence consideration, all groups cannot showed the snapshot of a conflict case under the 0.3-m-cell-size group
find good results. Besides, paths cannot be found no matter what sorting and room space scenario. For the last pipe in the group, we cannot find a
approach is adopted under the groups with 0.35 m of cell size. For the collision-free path for it because the paths of other planned pipes block
groups with cell sizes larger than 0.25 m, some of them cannot find all neighbor cells to the target end point highlighted. Such a situation
also occurred sometimes in the 0.25, 0.3, and 0.35-m-cell-size groups in

14
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 11. Planning time under different grid cell sizes of the aisle space.

planned firstly. However, the pressurized pipes require space signifi­

Table 9
cantly more than those of the firing systems. Therefore, planning was
Feasibility of planning under different pipe priorities and cell sizes in aisle space.
compromised, given the limited space. The consideration of the pipe
Grid cell size (m) 0.1 0.15 0.2 0.25 0.3 0.35 install sequence may not be appropriate in such a complicated scenario.
Length ● ● ● ● – – We suggest sorting approaches other than based on install sequence.
Cross Section ● ● ● – ● – In the two testing scenarios, we observed that the groups sorted by
Access Space ● ● ● ● – –
considering cross-section returns short and fewer elbow-demanded re­
Access Frequency ● ● ● ● ● –
Install Space ● ● ● ● ● – sults in all groups. Also, most of the planning results are feasible and
Install Sequence ○ ○ ○ ○ ○ – comply with all constraints. A possible reason is a fact that the cross-
section of a pipe directly affects space occupation. It suggests that
●: All pipes have paths and obey constraints, ○: Some pipes violet constraints, − :
users considering the cross-section firstly to sort the pipes is appropriate
Some pipes cannot find paths
before the planning process.
Given the results discussed above, appropriate cell sizes in the room
Table 10 scenarios are 0.15 and 0.2 m. Firstly, the two groups demand <7 s to
Total pipe length under different pipe priorities and cell sizes in aisle space. obtain conflict resolved results, which are reasonable for the user
Grid cell size (m) 0.1 0.15 0.2 0.25 0.3 0.35
making decisions on-site. Secondly, although they did not come out with
the shortest path among the six groups, they both obtained the paths of
Length 250.3 252.2 255.4 259.8*
the twelve pipes. And the paths are without violations to all constraints.
– –
Cross Section 250.1 244.1* 257.4 – 252.3* –
Access Space 257.7 262.1 260.6 282.3 – – Finally, yet importantly, the average numbers of elbows used are 39 and
Access Frequency 253.1 246.2 253.4* 265.8 261.9 – 38.33 under the two scenarios, which is relatively low among the six
Install Space 260.7 267.5 272.6 275.8 275.7 – groups. Excluding the groups with constraint violations, the best result is
Install Sequence 249.3* 248.9 257.0 264.3 254.1 –
sorted by considering cross-section in the 0.15-m-cell-size group, as
Average 253.5 254.2 259.4 263.2 261.0 –
shown in Fig. 13 (a). In the 0.2-m-cell-size group, the best result is sorted
*
The shortest result in the groups with the same cell size by considering access space, as shown in Fig. 13 (b).
In the aisle space scenario, the appropriate sizes of the cells are also
0.15 and 0.2 m. Similar reasons as in the room space scenario can be
Table 11
concluded: required <10.05 s, the short total pipe length (254.2 m in the
Total elbows needed under different pipe priorities and cell sizes in aisle space.
0.15 m-cell-size group), and the least total number of elbow consumed
Grid cell size (m) 0.1 0.15 0.2 0.25 0.3 0.35 (135.8 and 127.3). Other than the groups with constraint violations, the
Length 131 141 132 142 – – best result came from those sorted by considering cross-section in the
Cross Section 119 123 111 – 112 – 0.15 m-cell-size group, as shown in Fig. 13 (c). In the 0.2 m-cell-size
Access Space 164 146 131 161 – – group, the best result is planned by considering the access section, as
Access Frequency 135 140 124 140 156 –
Install Space 176 182 164 152 167 –
shown in Fig. 13 (d).
Install Sequence 94* 97* 91* 107* 106* – To sum up, the proposed on-site MEP system coordination approach
Average 136.5 135.8 127.3 135.2 135.3 – for conflict resolution shows its feasibility in terms of on-site usages
*
The result using the least elbows under the same cell size groups through a field demonstration. It can help decision-making as long as a
balance of computational time, and constraint/considerations are ach­
ieved as in the tests. These reflect the answers to the research questions
the aisle space scenario.
raised in the study. For a comparative view with the latest automatic
In the groups sorted by considering the install sequence under the
pipe coordination studies, the proposed approach achieved a gap-filling
aisle space scenario, almost every group returns a constraint violation.
to tackle on-site pipe routing adjustment or re-planning problems, while
We think the reason is the complexity: In such a scenario, we initially let
others are focused on comprehensive and macro solutions at design
the firing systems, with the highest priority in the actual installation, be
stages. The proposed approach can further compensate for the potential

15
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

Fig. 12. The constraint violation and pipeline blocking issues.

Fig. 13. Best results under room space and aisle space scenarios.

issues when deviations of planned pipe operations in practice are 6. Conclusion and future work
encountered. Nevertheless, experienced engineers could take around 2 h
to form a plan for nine pipe systems [3], and the current pipe coordi­ In this research, an AR-based approach for MEP system conflict
nation systems at the design stages can shorten the time to about 5 min resolution is developed to achieve the need for on-site pipeline inspec­
[12]. The proposed approach can then guarantee an efficient re- tion and coordination. The contribution comes from its practical usage
planning within 1 min as the scenarios can be narrowed down to to accomplish an efficient pipe re-planning ability and intuitive visual­
certain areas like the room and aisle spaces mentioned above. ization on-site, helping the inspection practitioners swiftly react to any

16
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

deviations discovered at a late stage of construction or maintenance [6] L. Deliang, L. Huibiao, Interfere-check applying to 3D automatic pipe route
arrangement, in, in: Proceedings of International Conference on Computational
processes. Little research before has moved on to the MEP real-time
Intelligence and Software Engineering, 2009, pp. 11–13. https://ieeexplore.ieee.
coordination process at the actual field construction or inspection org/document/5365920.
stages. In the planning module of the system, we follow the previous [7] C.S. Park, D.Y. Lee, O.S. Kwon, X. Wang, A framework for proactive construction
research works to consider the MEP knowledge from three aspects: defect management using BIM, augmented reality and ontology-based data
collection template, Autom. Constr. 33 (2013) 61–71, https://doi.org/10.1016/j.
function, construction, and maintenance, and accordingly develop an autcon.2012.09.010.
automatic grid-based pipe path planning algorithm for pipe coordina­ [8] C.H. Lee, M.H. Tsai, S.C. Kang, A visual tool for accessibility study of pipeline
tion. The developed MEP A* planner, a planner based on the A* shortest maintenance during design, Visualiz. Eng. 2 (2014), https://doi.org/10.1186/
s40327-014-0006-y, pp. 2:6.
path search algorithm, is able to deal with MEP system constraints based [9] H.P. Tserng, Y.L. Yin, E.J. Jaselskis, W. Hung, Y. Lin, Modularization and assembly
on design criteria and resolve the conflicts with new plans in a reason­ algorithm for efficient MEP construction, Autom. Constr. 20 (2011) 837–863,
able short time. In terms of user interaction, we designed a user interface https://doi.org/10.1016/j.autcon.2011.03.002.
[10] F. Leite, A. Akcamete, B. Akinci, G. Atasoy, S. Kiziltas, Analysis of modeling effort
for visualizing AR-based pipe planning results at sites. The system and impact of different levels of detail in building information models, Autom.
implementation is built on a mobile application using Unity 3D and Constr. 20 (2011) 601–609, https://doi.org/10.1016/j.autcon.2010.11.027.
ARKit package. [11] G. Lee, J.W. Kim, Parallel vs. sequential cascading MEP coordination strategies: a
pharmaceutical building case study, Autom. Constr. 43 (2014) 170–179, https://
To evaluate the performance and the planning results of the doi.org/10.1016/j.autcon.2014.03.004.
approach, we designed two virtual scenarios and conducted tests on the [12] J. Singh, J.C.P. Cheng, Automating the generation of 3D multiple pipe layout
planner accordingly to different configurations: pipe planning priority design using BIM and heuristic search methods, Lecture Notes Civil Eng. 98 (2021)
54–72, https://doi.org/10.1007/978-3-030-51295-8_6.
and grid cell sizes for path searching. The result shows that the groups
[13] F. Bosché, M. Ahmed, Y. Turkan, C.T. Haas, R. Haas, The value of integrating scan-
with 0.15 and 0.2 m grid cell sizes both achieved an efficient time to-BIM and scan-vs-BIM techniques for construction monitoring using laser
consumption, within 10 s, to get the reasonable pipe path plans. This scanning and BIM: the case of cylindrical MEP components, Autom. Constr. 49
result indicates that the approach is feasible, effective, and efficient in (2015) 201–213, https://doi.org/10.1016/j.autcon.2014.05.014.
[14] A. Dimitrov, M. Golparvar-Fard, Segmentation of building point cloud models
solving the on-site MEP conflict resolution issues. It shows an advantage including detailed architectural/structural features and MEP systems, Autom.
compared to other existing pipe planning approaches that may only Constr. 51 (2015) 32–45, https://doi.org/10.1016/j.autcon.2014.12.015.
focus on global coordination of the MEP systems at an entire project [15] V.S. Kalasapudi, P. Tang, Y. Turkan, Computationally efficient change analysis of
piece-wise cylindrical building elements for proactive project control, Autom.
level and lack on-site re-planning flexibilities. Furthermore, the devel­ Constr. 81 (2017) 300–312, https://doi.org/10.1016/j.autcon.2017.04.001.
oped planner considers the MEP system constructability and maintain­ [16] Y. Kim, C.H.P. Nguyen, Y. Choi, Automatic pipe and elbow recognition from three-
ability. And it can help the on-site works avoid further issues caused by dimensional point cloud model of industrial plant piping system using
convolutional neural network-based primitive classification, Autom. Constr. 116
deviations between ideal designs and actual site conditions. The system (2020), 103236, https://doi.org/10.1016/j.autcon.2020.103236.
further complements existing automatic MEP system planning ap­ [17] M.-K. Kim, J.P.P. Thedja, H.-L. Chi, D.-E. Lee, Automated rebar diameter
proaches that apply at the design or planning stages. classification using point cloud data based machine learning, Autom. Constr. 122
(2021), 103476, https://doi.org/10.1016/j.autcon.2020.103476.
This research has limitations in the MEP systems planning process. [18] F. Li, H. Li, M.-K. Kim, Registration-free point cloud generation technique using
Currently, the approach only allows us to obtain path solutions from rotating mirrors, Com. aided Civil Infrastruct. Eng. 37 (2) (2022) 204–226, https://
single priority sorting consideration. It is to achieve the sequential path doi.org/10.1111/mice.12707.
[19] L. Wang, F. Leite, Formalized knowledge representation for spatial conflict
planning process instead of a brute-force search. Further, the optimal
coordination of mechanical, electrical and plumbing (MEP) systems in new
solution may not appear in the discussed priority sorting approaches. building projects, Autom. Constr. 64 (2016) 20–26, https://doi.org/10.1016/j.
The user may want to consider the sorting based on multiple consider­ autcon.2015.12.020.
ations with different MEP design knowledge features. Therefore, we [20] S. Dong, A.H. Behzadan, F. Chen, V.R. Kamat, Collaborative visualization of
engineering processes using tabletop augmented reality, Adv. Eng. Softw. 55
suggest future research focus on the multi-criteria sorting method and (2013) 45–55, https://doi.org/10.1016/j.advengsoft.2012.09.001.
determine its weighting mechanism. Future research can also dedicate [21] T.H. Lin, C.H. Liu, M.H. Tsai, S.C. Kang, Using augmented reality in a multi-screen
to the balance of the grid cell size and the searching range using not only environment for construction discussion, J. Comput. Civ. Eng. 29 (6) (2015)
04014088, https://doi.org/10.1061/(ASCE)CP.1943-5487.0000420.
the heuristic path-planning algorithm but finding the optimization so­ [22] H. Bae, F.M. Golparvar, J. White, High-precision vision-based mobile augmented
lution when possible. reality system for context-aware architectural, engineering, construction and
Supplementary data to this article can be found online at https://doi. facility management (AEC/FM) applications, Visualiz. Eng. 1 (2013), https://doi.
org/10.1186/2213-7459-1-3, pp. 1:3.
org/10.1016/j.autcon.2022.104400. [23] S. Meža, Z. Turk, M. Dolnec, Component based engineering of a mobile BIM-based
augmented reality system, Autom. Constr. 42 (2014) 1–12, https://doi.org/
10.1016/j.autcon.2014.02.011.
Declaration of Competing Interest [24] K.C. Yeh, M.H. Tsai, S.C. Kang, On-site building information retrieval by using
projection-based augmented reality, J. Comput. Civ. Eng. 26 (3) (2012) 342–355,
https://doi.org/10.1061/(ASCE)CP.1943-5487.0000156.
The authors declare that they have no known competing financial [25] A. Fazel, A. Izadi, An interactive augmented reality tool for constructing free-form
interests or personal relationships that could have appeared to influence modular surfaces, Autom. Constr. 85 (2018) 135–145, https://doi.org/10.1016/j.
the work reported in this paper. autcon.2017.10.015.
[26] J. Chalhoub, S.K. Ayer, Using mixed reality for electrical construction design
communication, Autom. Constr. 86 (2018) 1–10, https://doi.org/10.1016/j.
References autcon.2017.10.028.
[27] S.S. Kang, S. Myung, S.H. Han, A design expert system for auto-routing of ship
pipes, J. Ship Prod. 15 (1999) 1–9, https://doi.org/10.5957/jsp.1999.15.1.1.
[1] T.M. Korman, M.A. Fischer, C.B. Tatum, Knowledge and reasoning for MEP
[28] D. Zhu, J.C. Latombe, Pipe routing-path planning (with many constraints), in:
coordination, J. Constr. Eng. Manag. 129 (6) (2003) 627–634, https://doi.org/
Proceedings of 1991 IEEE International Conference on Robotics and Automation 3,
10.1061/(ASCE)0733-9364(2003)129:6(627).
1991, pp. 1940–1947. https://ieeexplore.ieee.org/document/131911.
[2] Z.Z. Hu, J.P. Zhang, F.Q. Yu, P.L. Tian, X.S. Xiang, Construction and facility
[29] T. Ito, A genetic algorithm approach to piping route path planning, J. Intell. Manuf.
management of large MEP projects using a multi-scale building information model,
10 (1999) 103–114, https://doi.org/10.1023/A:1008924832167.
Adv. Eng. Softw. 100 (2016) 215–230, https://doi.org/10.1016/j.
[30] J.H. Park, R.L. Storch, Pipe-routing algorithm development: case study of a ship
advengsoft.2016.07.006.
engine room design, Expert Syst. Appl. 23 (2002) 299–309, https://doi.org/
[3] J. Singh, J.C.P, C.J. Anumba, BIM-based approach for automatic pipe systems
10.1016/S0957-4174(02)00049-0.
installation coordination and schedule optimization, J. Constr. Eng. Manag. 147
[31] H. Sui, W. Niu, Branch-pipe-routing approach for ships using improved genetic
(11) (2021) 04021143, https://doi.org/10.1061/(ASCE)CO.1943-7862.0002077.
algorithm, Front. Mech. Eng. 11 (3) (2016) 316–323, https://doi.org/10.1007/
[4] Q. Lu, Y.H. Wong, A BIM-based approach to automate the design and coordination
s11465-016-0384-z.
process of mechanical, electrical, and plumbing systems, HKIE Trans. 25 (4) (2018)
[32] Q. Liu, G. Jiao, A pipe routing method considering vibration for aero-engine using
273–280, https://doi.org/10.1080/1023697X.2018.1537813.
kriging model and NSGA-II, IEEE Access 6 (2018) 6286–6292, https://doi.org/
[5] J. Wang, X. Wang, W. Shou, H.Y. Chong, J. Guo, Building information modeling-
10.1109/ACCESS.2018.2789361.
based integration of MEP layout designs and constructability, Autom. Constr. 61
(2016) 134–146, https://doi.org/10.1016/j.autcon.2015.10.003.

17
L.-T. Tsai et al. Automation in Construction 141 (2022) 104400

[33] D.L. Liu, H.B. Lu, Interfere-check applying to 3D automatic pipe route [39] E.S. Jones, S. Soatto, Visual-inertial navigation, mapping and localization: a
arrangement, in: International Conference on Computational Intelligence and scalable real-time causal approach, Int. J. Robot. Res. 30 (4) (2011) 407–430,
Software Engineering, 2009, pp. 1–3. https://ieeexplore.ieee.org/documen https://doi.org/10.1177/0278364910388963.
t/5365920. [40] C. Wang, X. Sun, L. Sun, T. Yuan, A method based on PSO for pipe routing design,
[34] W.Y. Jiang, Y. Lin, M. Chen, Y.Y. Yu, A co-evolutionary improved multi-ant colony in: Proceedings of the 6th Annual IEEE International Conference on Cyber
optimization for ship multiple and branch pipe route design, Ocean Eng. 102 Technology in Automation, Control and Intelligent Systems, 2016, pp. 422–427,
(2015) 63–70, https://doi.org/10.1016/j.oceaneng.2015.04.028. https://doi.org/10.1109/CYBER.2016.7574862.
[35] A. Asmara, U. Nienhuis, Automatic piping system in ship, in: Proceedings of [41] J. Singh, M. Deng, J.C.P. Cheng, Implementation of mass customization for MEP
International Conference on Computer and IT Application (COMPIT), 2006, layout design to reduce manufacturing cost in one-off projects, in: Proceedings of
pp. 8–10. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.105. 26th Annual Conference of the International Group for Lean Construction (IGLC),
354&rep=rep1&type=pdf. 2018, pp. 625–635, https://doi.org/10.24928/2018/0519.
[36] S.H. Kim, W.S. Ruy, S.B. Jang, The development of a practical pipe auto-routing [42] Y.-J. Chen, Y.-S. Lai, Y.-H. Lin, BIM-based augmented reality inspection and
system in a shipbuilding CAD environment using network optimization, Int. J. maintenance of fire safety equipment, Autom. Constr. 110 (2020), 103041, https://
Naval Architect. Ocean Eng. 5 (2013) 468–477, https://doi.org/10.2478/IJNAOE- doi.org/10.1016/j.autcon.2019.103041.
2013-0146. [43] F. Baek, I. Ha, H. Kim, Augmented reality system for facility management using
[37] C.B. Tatum, T.M. Korman, Coordination building system: process and knowledge, image-based indoor localization, Autom. Constr. 99 (2019) 18–26, https://doi.org/
J. Archit. Eng. 6 (4) (2000) 116–121, https://doi.org/10.1061/(ASCE)1076-0431 10.1016/j.autcon.2018.11.034.
(2000)6:4(116). [44] Trimble XR10 with HoloLens 2, Trimble Field Technology, Trimble. https://fieldte
[38] P.E. Hart, N.J. Nilsson, B. Raphael, A formal basis for the heuristic determination of ch.trimble.com/en/products/mixed-reality/trimble-xr10-with-hololens-2, 2022
minimum cost paths, IEEE Trans. Syst. Sci. Cybernet. 4 (2) (1968) 100–107, (accessed January 25).
https://doi.org/10.1109/TSSC.1968.300136. [45] B. Schiavi, V. Havard, K. Beddiar, D. Baudry, BIM data flow architecture with AR/
VR technologies: use cases in architecture, engineering and construction, Autom.
Constr. 134 (2022), 104054, https://doi.org/10.1016/j.autcon.2021.104054.

18