AUTCON-02052; No of Pages 12

Automation in Construction xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Automation in Construction

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

4D CAD-based evaluation system for crane deployment plans in

construction of nuclear power plants
Y. Sugimoto a, H. Seki a, T. Samo b, N. Nakamitsu b
a
Research and Development Group, Hitachi Ltd., Hitachi, Japan
b
Global Business Development and Management Division, Hitachi-GE Nuclear Energy Ltd., Hitachi, Japan

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

Article history: To optimize a crane deployment plan is an important problem to reduce construction cost. Previous researches
Received 19 October 2015 focused on developing methods to find an optimal crane deployment plan automatically, which minimizes
Received in revised form 22 March 2016 total operating time under physical constraints such as load capacity. These researches did not focus on evaluat-
Accepted 2 April 2016
ing transitions of crane utilization rates during construction period. Unlike a general tall building, construction
Available online xxxx
schedules of a nuclear power plant (NPP) tend not to take the form of a series of construction cycles in a typical
Keywords:
floor. Therefore, transitions of crane utilization rates need to be evaluated in construction of a NPP.
4D CAD The purpose of this paper is to develop a 4D CAD-based evaluation system, which calculates transitions of utili-
Interference check zation rates automatically, for planning of crane-deployment in construction of a NPP. To calculate utilization
Random forests rates automatically and accurately, two functions were developed. The first function maps data between 3D ob-
Utilization rate jects and carry-in tasks by using a fast geometric interference checking method based on “Axis Aligned Bounding
Crane Box.” The second one estimates installation material quantities in early construction-planning phases by using a
data mining technique called “random forests.” The result of verification tests based on previous construction
data indicates that the proposed system will significantly reduce the time required to create crane-deployment
plans and improve their accuracies.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction
A nuclear power plant (NPP) is composed of a lot of facilities classi-
fied into categories such as mechanical, electrical, and HVAC (heating,
ventilating, and air conditioning). For example, the number of pieces
of equipment reaches several thousand, and the total length of piping
components reaches 100,000 km. These facilities are complicatedly
arranged in several hundred areas surrounded by thick concrete walls.
Unlike a general tall building, floor plans of a NPP differ widely accord-
ing to floor so that construction schedules of a NPP tend not to take the
form of a series of construction cycles in a typical floor. Therefore, as a
project, constructing an NPP is so large and complex that it needs
several years and billions of dollars to complete, and any delay to the
project causes a huge loss.
The “open-top construction method” is adopted for a NPP [1,2]. As
for this method, activities involved in constructing buildings and
installing facilities are carried out concurrently at the construction site
of an NPP. Although this method can reduce the construction period, it
requires cranes to be shared between companies in charge of the two
types of activities, namely, building frame work and equipment work.
Therefore, properly drawing up plans for deployment of cranes, which
E-mail addresses: yohhei.sugimoto.pj@hitachi.com (Y. Sugimoto), hiroshi.seki.mf@
hitachi.com (H. Seki).
contain the number of cranes, their locations, and a schedule to use
them, is vital in regard to preventing delays of construction projects.
In terms of efficiency, the quality of plans for deploying cranes
depends on a planner's know-how based on their experience. If the de-
mands for cranes exceed the cranes' capacities, and delays are forecast,
additional cranes must be deployed and overtime work must be sched-
uled. These measures naturally increase costs. To solve this problem, it is
necessary to develop a system that can create robust plans without the
need for skillful planners on the basis of rules extracted from previous
statistics concerning construction data.
One of the ways to evaluate crane-deployment plans quantitatively
is to estimate the utilization rates of cranes. It requires determining
which cranes are assigned to each material. However, it is a problem
that the number of materials used in constructing a NPP is too large to
determine the assignments of cranes manually. Additionally, it is also
a problem that the quantities of materials required and the number of
facilities to be built is uncertain in early planning phase of crane-
deployment. This is because engineering of a NPP is not complete at
that time owing to the concurrency between engineering and construc-
tion planning.
Researching methods for creating accurate crane-deployment plans
effectively, Huang et al. have developed a planning method (which ap-
plies a mixed-integer linear-programming technique) for automatically
creating deployment plans for tower cranes [3]; Lien et al. have

http://dx.doi.org/10.1016/j.autcon.2016.04.004
0926-5805/© 2016 Elsevier B.V. All rights reserved.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
2 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx
developed a method that applies the “particle bee algorithm” [4]. These
methods calculate crane deployment plan automatically; however,
these researches mainly target general tall building so that they do not
refer to transitions of crane utilization rates. This is because construction
schedules of general tall building tend to take the form of a series of con-
struction cycles in a typical floor. On the other hand, floor plans of a NPP
differ widely according to floor so that crane utilization rates are possi-
ble to differ widely by floor. Therefore, evaluating transitions of crane
utilization rates is important in NPP construction to reduce deviations
and improve productivity.
In addition, these researches do not refer to how to prepare the input
data such as material quantities carried by each crane. To reduce man-
power to prepare the input data, Wang et al. have developed “Firefly
Algorithm” combined with building information modeling (BIM) [5].
In this research, material quantities carried by each crane is automati-
cally calculated based on Construction BIM(C-BIM), which includes con-
struction support models such as scaffold and formwork. However, in
construction of NPPs, C-BIM is not completely created. For example, it
is often the case that support models are created not as 3D model, but
as 2D model. Also, in early planning phase, quantities of some kinds of
materials such as piping components are not fixed. Therefore, material
quantity take-off is a critical problem for optimal crane deployment
plan.
To evaluate transitions of crane utilization rates based on crane
deployment plans and construction schedules, we have focused on 4D
CAD, which integrates 3D models and a time dimension. 4D CAD was
used to visualize construction schedule as intuitive animation for
validating construction plans [6–9]. As a result of these researches, 4D
CAD has become so prevalent that commercial 3D CAD software
includes a function for it. Recently, Kang L. S. et al. developed a 4D
CAD-based simulation system for evaluating and visualizing risks
involved in a construction project [10].
In this study, a simulation system for evaluating and visualizing the
validity of crane-deployment plan was developed. 4D CAD was used to
calculate transitions of utilization rates of cranes, and a data mining
technique called “Random Forests”(RF) [11] was used to calculate
material quantities from incomplete C-BIM in early planning phase.
2. Planning of crane deployment
2.1. Crane deployment
For constructing facilities in buildings in a NPP efficiently, it is impor-
tant to deploy cranes properly. Two types of cranes are deployed during
construction of a NPP a tower type (which moves up and down while
fixed in a stationary position) and a crawler type (which moves up
and down and can move along the ground by caterpillars). As shown
in Fig. 1, the area in which a crane can operate (“operating area”
building
frame
equip-
ment
crane
operating
area
Fig. 1. Image of an area in which a crane operates.
Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
Crane position
Crane
operating area
Building
Fig. 2. Image of a deployment of cranes.
hereafter) is modeled as a cylinder with radius equal to the operating
radius of the crane. As shown in Fig. 2, all materials and facilities must
be covered by the operating areas of cranes. The number of cranes is
also a key consideration. If it is too low, the utilization rates of the cranes
will exceed their capacities. On the other hand, an excessive number of
cranes incurs large costs and may cause physical interference between
operating cranes. Additionally, the load limits of cranes must also be
considered.
2.2. Carry-in schedule
A sequence of building frame work and equipment work, especially
for carrying equipment and piping components in buildings, is shown in
Fig. 3. In steps 1 and 2, floors and walls of areas are constructed. In step
3, facilities such as equipment and piping components are carried into
those areas from above by cranes. In step 4, the ceilings are constructed.
After step 4, the facilities are installed in those areas. The main reason
for this sequence is that large facilities cannot be carried into areas
through delivery entrances of the building after the ceilings of the
areas are added. Although this sequence of building works and mechan-
ical and electrical works in an area are conducted in series, such works
in several areas are conducted concurrently.
Materials and facilities carried into areas in buildings are categorized
in two types: (a) building frame works such as structures, rebar, molds,
and scaffolding, and (b) mechanical and electrical works such as equip-
ment, piping components, operating platforms, metal components, and
scaffolding.
Each activity in work schedules has attributes such as building name,
floor name, names of NPP systems, and drawing number. These attri-
butes make it possible to automatically link activities to 3D objects,
which is called mapping hereafter. An example of mapping is shown
in Fig. 4. It is supposed that activities have the area names where they
(Step1) (Step1) (Step2)
Construct floors
(Step2)
Construct walls
(Step3) (Step3) (Step4)
Carry-in equipments
and piping spools
(Step4)
Construction of ceilings
Fig. 3. Construction steps performed in an area(left), and image of construction sequences
performed in an area(right).
 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx 3

schedule
Activit Activity Area Start Finish
y ID Name Name Date Date
carry-in pipe
Carry- area B31 carry-in pipe
A1 B31 SD1 FD1
in pipe area B32
Carry- carry-in pipe
A2 B32 SD2 FD2 area B33
in pipe
Carry- carry-in pipe
A3 B33 SD3 FD3 area B34
in pipe
Carry-
A4 B34 SD4 FD4
in pipe

3D model
Object Object Area Upper Lower area B31 area B32
ID Layer Name Point Point P1 P2 P3
P11 P12
P1 Pipe B31 UP1 LP1 P4
P10
P2 Pipe B31 UP2 LP2 caption
P3 Pipe B31 UP3 LP3 P5 P9
Pipe
… … … … … P6 P7 P8
Area
P12 Pipe B34 UP4 LP4 area B33 area B34

Fig. 4. Mapping activities to 3D objects.

are carried out, and 3D objects have as the area names where they are
carried or installed. It is thus possible to automatically create mapping
between activities and 3D objects according to the area names.
2.3. Assignment of material quantities to cranes
The number of cranes is limited by the space available in a construc-
tion site and by the delivery and operational costs of the cranes.
Companies in charge of building works and mechanical and electrical
works must therefore share cranes.
Operating areas of cranes must overlap as much as they can so that
the dependency of each crane decreases and risks of delay can be
reduced.
To calculate utilization rates of cranes, it is necessary to assign a
crane to each 3D object. Fig. 5 shows an example of the assignments
in the case that operating areas of two cranes overlap. Pipe 1 can be car-
ried by crane 1, and pipe 3 can be carried by crane 2. On the other hand,
pipe 2 can be carried by both cranes 1 and 2. In this research, the quan-
tity of pipe 2 is equally divided to the two cranes because its assignment
is not so important in early planning phase. In detailed planning phase,
pipe 2 should be assigned to either crane 1 or crane 2 depending on
some indices such as distances, utilization rates, and specs of cranes.
2.4. Quantity of piping components
When utilization rate of a crane is calculated, the quantities of mate-
rials and facilities to be carried in are uncertain because detailed design
of a NPP is not complete at that time due to the concurrency between
design and construction planning. As for a piping spool which means
a unit of prefabricated piping components, the design process is
made considering workability and cost through manufacturing, trans-
portation, installation, and welding. Therefore, it takes a long time to
complete the design process. Fig. 6 shows an example of piping compo-
nents and piping spools.
Quantities have been predicted on the basis of ones in a reference
plant and the differences between the size of buildings (such as number
of floors and floor space) in the current plant being constructed and the
reference one. This method can precisely predict quantities when they
are calculated in floor units. However, the prediction accuracy gets
worse when quantities are predicted in more detailed units such as
area. This is a problem in regard to predicting utilization rates of cranes
because a crane operating area is smaller than the floor space.
3. System for simulating crane deployment
The purpose of this system is to improve the quality of crane-
deployment plans. To solve the aforementioned problems, 4D simula-
tion technology and a data mining technique, called “random forests,”
is applied in the developed crane-deployment simulation system. The
functions of the system are described as follows.
The first function maps data between 3D objects, carry-in tasks, and
cranes by using a method called “fast geometric interference checking.”
The calculation time is a significant problem described as following
subsection. The second one estimates quantities of installation materials
in the early phases of construction planning by using a data mining
technique, called “random forests,” which infers data from the results
of previous projects.
3.1. System configuration
The configuration of the developed system is shown in Fig. 7. The
data input into the system include spec and position data of cranes,

pipe1 pipe2 pipe3 Before After

Sp 1
P1 P2 P3 P1 P2 P3 Sp 4
P11 P12 P11 P12
P4
crane1 crane2 P4
P10 P10
Sp 2
P5 P9 P5 P9

P6 P7 P8 P6 P7 P8

Fig. 5. Assignment of piping components to cranes. Fig. 6. Image of piping components (left) and piping spools (right).

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
4 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx

(Step1) Table 1
Spec & Position Create 3D models of Data format for crane deployment.
of Cranes cranes
Basic unit Crane name X Y Z Max radius (m) Spec
carry-in time
(Step2) (m) (m) (m) (t ∙ m)
Mapping: TC-1 −45 2 0 60 900
Plant 3D
model crane–plant 3D TC-2 −15 −20 0 50 400
(Step5)
TC-3 30 100 0 55 450
Calculate crane
(Step3) utilization rate
Construction Mapping:
Schedule schedule–plant 3D

(Step4) (Step6)
carry-in time, which means the operating time taken by a crane to
carry one of materials or facilities into a building, are decided based on
Predict quantities Visualize
the results in previous projects. These values are defined for each kind
of material and facility.
Fig. 7. System configuration and flowchart. Finally, in step 6, the function called “visualization” visualizes the
predicted utilization rates of cranes in the form of 2D or 3D graphs.

plant 3D model, a construction schedule, and basic units of working
time for carry-in of each material. The spec and position data include
the operating radius of cranes and the coordinates of points at which
cranes are deployed. The plant 3D model includes at least building
frames, large equipment, and large diameter pipes (even if they have
not been split in spools). The 3D-object data include parameters of
shapes and attributes such as system name, drawing number, and
weight. The construction-schedule data includes name, start date, finish
date, and attributes for specifying the details of the work (such as
working area).
In step 1, 3D models of cranes are created in the plant 3D model
according to the spec and position data.
In step 2, the function described as “mapping: crane–plant 3D”
connects cranes and 3D objects in the plant 3D model that they can
carry into buildings.
In step 3, the function described as “mapping: schedule–plant 3D”
connects activities in the construction schedule and 3D objects in the
plant 3D model by matching their attributes.
In step 4, the function called “predict quantities” calculates quanti-
ties of concretes, rebar, equipment, and piping spools on the basis of
the plant 3D model. In addition, before engineering of piping spools is
finished, this function predicts the quantities of piping spools on the
basis of 3D models in previous projects.
In step 5, the function called “calculate crane utilization rate” calcu-
lates utilization rates of cranes. Utilization rate means operating time of
cranes in a particular period in units of months or weeks. It is calculated
based on quantities and basic units of carry-in time. The basic units of
3.2. Create 3D models of cranes
This system has an interface that enables users to deploy cranes
on a 3D model easily. An image of the interface is shown in Fig. 8.
Users can add and delete cranes with their name and specs (such
as operating radius) and modify their positions and specs. Cranes
can be deployed one by one manually using this interface or auto-
matically using an input function for text data (whose format is
shown in Table 1). As for the 3D models of cranes, some templates
have been created and will be attached to the 3D model of a NPP at
each position of the cranes.
3.3. Mapping between cranes and 3D model
To determine which 3D objects are located within the operating area
of each crane, interference-checking functions, provided by commercial
3D-CAD software, can be used. For example, operating areas of cranes
can be modeled as cylinders, and interference checking between the
cylinders and the 3D objects composing a NPP can be performed.
However, this checking takes a long time to calculate possibility of
interferences. This is because the interference-checking function con-
siders precisely detailed shapes of 3D objects. This calculation time is a
significant problem when several deployment plans must be evaluated.
Therefore, a fast interference-checking method using an “axis-aligned
bounding box” (AABB) was developed.
An image of an interference check between 3D objects in a crane
operating area and piping components is shown in Fig. 9.

Fig. 8. Crane-deployment interface.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx 5

[Isometric view] [Top view] learning steps

learning create bootstrap samples
operating radius AABB (axis aligned data
R bounding box) Operating radius
AABB sample 1 … sample N
d1
d1 learning decision tree 1 … learning decision tree N
R
d2 prediction steps
d2 test prediction by decision tree 1 … prediction by decision tree N
data
result 1 result N
Fig. 9. Image of interference checking using an axis-aligned bounding box.
take a vote (or an average)

When the distances between the center axis of a cylinder and the result
intersection points of the edges of the AABB are smaller than the crane
radius, the 3D object and the crane are considered to be interfering. Fig. 11. Procedures of random forests.

min dk b R
k∈Ei
Ei: set of intersection points between edges of AABB for object i
dk: distance between center axis of a cylinder and an intersenction
point k in Ei
3.4. Mapping between schedule and 3D model
To link activities of construction schedules with 3D objects, the
matching of their attributes can be used. For example, they can be linked
on the basis of matching names of areas in which activities are
performed with names of areas in which 3D objects are located.
Additionally, a function for attaching an attribute such as an area
name on the basis of simple input data, in the case that 3D objects
have not been given an attribute, was developed. The input data are a
text data in which area names are manually assigned to each voxel sep-
arated by baselines of a building. Then, the voxels with their area names
are linked to 3D objects by a fast interference check using AABB. As a
result, each 3D object is given the area name where it is located.
ð1Þ
length of the piping objects, which have not been spools, and other
attributes such as diameter. This prediction method is described in
detail in the next section.
As for equipment, the quantities are counted on the basis of attri-
butes such as equipment ID, names of parts of equipment, and volumes
of 3D objects composing equipment. As for materials for building works
(such as rebar), their quantities are predicted on the basis of the space
and volume of objects composing building frames.
Regarding to materials which need to be carried into buildings by
cranes, the quantity of rebar is calculated based on concrete volume.
The concrete volume is calculated by the volume of 3D objects compos-
ing of building frames. In addition, the quantity of steel frame is calculat-
ed based on the number and volume of 3D objects.
To consider quantities of other materials such as scaffolding, which
are not designed in plant 3D model and therefore whose quantities
are difficult to be calculated directly, we use surface area of wall and
ceiling, and volume of each area in a building calculated from 3D objects.
3.6. Prediction of quantity of piping spools

3.5. Calculate quantity of 3D objects The data mining technique for predicting quantities of piping spools
that are not grouped by spools is described in detail as follows.
It is often the case that 3D objects and quantities cannot be linked The input parameters for this method are length of the piping object,
easily because of a lack of attributes for construction works due to un- length of the piping line to which the pipe object belongs, pipe diameter,
completed detailed design of an NPP. As for piping components, before type of building in which it is located, the name of the NPP system that it
detailed design involving welding points and piping spools are complet- belongs to, and number of bends in the piping line. The output parame-
ed, it is impossible to count the number of spools, which is equivalent to ter is average length of spools in a piping line. To determine the quantity
the quantity of piping components to be carried into buildings. In this of a pipe object, the actual length of the object is divided by the predict-
case, it is necessary to predict the number of spools on the basis of the ed average length of spools. “Random forests” (RF) [10] is applied for
determining the relationship between the input parameters and the
output parameters. RF is an “ensemble learning technique” that uses
several learning algorithms as weak learners and merges their
all lines predictions to predict something accurately. It uses decision trees (or
regression trees) as weak learners.
An image of a decision tree (regression tree) is shown in Fig. 10. This
number of number of technique classifies input data (training data) into multiple groups by
bends < 5 bends > 5 using a tree structure. The root node of the tree holds all the training
data. The child nodes of the root node are separated by the number of
bends (which is one of the parameters of the training data). The thresh-
system system diameter diameter olds using the separation are determined by an algorithm such as C4.5
A&C B < 200mm > 200mm and CART, which is based on entropy of information. Leaf nodes hold
the distribution parameters such as a mean value and a variance of
training data classified in each node. The decision tree allows both
qualitative data and quantitative data.
avg. µ1 avg. µ2 avg. µ3 avg. µ4 Sequences of learning and prediction by RF are shown in Fig. 11.
dev. σ1 dev. σ2 dev. σ3 dev. σ4 First, bootstrap samples are created from learning data by random sam-
pling with replacement. Second, decision trees are created from each
Fig. 10. Example of a decision tree (regression tree). bootstrap sample. In this step, RF does not necessarily use all the

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
6 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx

object – crane object – carry-in object – quantity basic unit of carry-

mapping schedule mapping table in time table
object crane start finish Quantity basic unit
object object Q(O j )
object
date date (O j )
O1 TC-1
O1 SD1 FD1 O1 Q1 O1 1
O1 TC-2
O2 TC-2 O2 SD2 FD2 O2 Q2 O2 2

(step1) unite tables

obje quan Carry-in schedule
crane SD1 SD2 FD1 FD2
ct tity
TC-1
O1 Q1
TC-2

O2 TC-2 Q2
Q(O j ) (O j ) 1
R(O j , t )
(step2) calculate operating time of each crane WT ( SDO j , FDO j ) NC (O j )
( SDO j t FDO j )
crane resource graph
Q1 1 1
TC-1 R1 R1 R(O1 , t )
WT ( SD1, FD1) 2
R1+R2
TC-2 R2 Q2 2 1
R1 R2 R(O2 , t )
WT ( SD2, FD 2) 1

Fig. 12. Image of the process for calculating operating time of cranes.

parameters that are used to classify learning data in nodes of a tree. That where
is, RF can randomly select parameters to be used when it creates a
decision tree from each bootstrap sample. Finally, it merges the results Q(Oj): Carry-in quantity of Oj,
of predictions by all decision trees (as weak learners) on the basis of β(Oj): Basic unit of time to carry a unit quantity of Oj,
accepting a majority or taking an average. SDOj: Start date of carrying Oj,
FDOj: Finish date to carrying Oj,
3.7. Calculation of crane-utilization rate WT(SDOj, FDOj): Number of working days between SDOj and FDOj,
NC(Oj): Number of cranes that can carry Oj.
To calculate utilization rates of cranes, the following data are used:
schedule data, mapping data between schedules and 3D objects and be- After the assignments of all 3D objects are completed, the utilization
tween 3D objects and cranes, quantities to be carried into buildings, and rate of a crane is calculated from the sum of the quantities assigned to
basic units of operating time of cranes to carry each material and facility the crane described as the following equation:
into buildings.
X


An image of the algorithm for calculating utilization rates is shown in U ðck ; t Þ ¼ δ ck ; O j  R O j ; t
Fig. 12. In step1, the input data is united on the basis of object ID as a key. 
j

 ð3Þ
In step2, the carry-in quantity of each 3D object is equally assigned to 1 If ck and O j are mapped:
δ ck ; O J ¼
cranes that are linked to the 3D object. 0 If ck and O j are not mapped:
The carry-in quantity of a 3D object, “Oj” hereafter, assigned to each
crane, is defined as follows: where U(ck,t): the utilization rate of crane “ck” in time t.

8 3.8. Visualization of utilization rates

< Q ðOjÞ  βðOjÞ 1


 SDOj ≤ t ≤ FDOj
RðOj; t Þ ¼ WT SDOj ; FDOj NC ðOjÞ ð2Þ
:
 This function is to visualize the utilization rates by graphs represent-
0 t b SDOj ; FDOj b t ed by 2D and 3D. 2D graphs of crane-utilization rates are shown in

10
Utilization rate [h/day]

6 TC-1

4
TC-2
2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

Fig. 13. Graphs of utilization rates of cranes.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx
Z (utilization rate)
(position)
0 X
current
position
Fig. 14. Bar graph showing utilization rates at each position of a crane.
Fig. 13. The horizontal axis means date, and the vertical axis means
crane-utilization rates in units of [hours/day] or [hours/month]. In this
figure, the two line graphs represent transitions of utilization rates of
two cranes, called “TC-1” and “TC-2.” This graph confirms the validity
of the crane-deployment plan. If the peaks of utilization rates of cranes
exceed the limits, the plan of crane deployment should be modified
from the viewpoint of number of cranes, their positions, and carry-in
schedules.
To support manual modification of crane positions, we developed a
function for visualizing the utilization rates of a crane around its current
position by using bar charts. The image is shown in Fig. 14. The z-axis
represents the deference of utilization rates between the current posi-
tion and each position. The z value is positive if the utilization rate is
larger than that at the current position. This graph is useful to determine
local optimization of a crane, although the effects between other cranes
are not shown in it.
An actual visualization of this function is shown in Fig. 15. The 3D
objects of bar charts show utilization rates of a crane around its current
position. This visualization is useful when a planner modifies the
position of a crane whose utilization rate is too high (or low).
Additionally, if the range of positions is expanded to the whole site of
a NPP, as shown in Fig. 16, the visualization helps not only modification
of crane positions but also initial planning of crane positions.
An image of the steps for determining deployments of cranes by
using this visualization is shown in Fig. 17. Step (1) shows a scene in
which the position of crane “TC-1” is being considered by visualizing
its utilization rates. Step (2) shows a scene in which the position
of “TC-2” is being considered by visualizing its utilization rates after
Fig. 15. 3D visualization of utilization rates of a crane at several positions around the current one.
Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
7
Fig. 16. 3D visualization of utilization rates of a crane at the whole site of a NPP.
“TC-1” has been deployed. In this manner, the height of bar graph be-
comes lower around the positions of cranes that have been deployed.
To solve overload of cranes, a function to level utilization rates of
several cranes was also developed. This function modifies the assign-
ments of materials and facilities which are shared between several
cranes whose operating radiuses overlap as described in 2.2. For this
purpose, the operating time required for carrying shared quantities,
which is described as “shared operating time” hereafter, is calculated
from Eq. (3). Fig. 18 shows breakdown of the operating time of “TC-1,”
which includes the shared operating time with “TC-2.”
The modification of the assignments of shared operating time
between two cranes, which are described as c1 and c2, in time t is
given by the following equation:
U nþ1 ðc1 ; t Þ ¼ U n ðc1 ; t Þ−U ̂n ðc1 ; c2 ; t Þ
U nþ1 ðc2 ; t Þ ¼ U n ðc2n; t Þ þ U n̂ ðc1 ; c2 ; t Þ
̂
U n ðc1 ; c2 ; t Þ ¼ min γ  U SH n ðc1 ; c2 ; t Þ; ð4Þ
U ðc2 Þ−U n ðc2 ; t Þ;
1
fU n ðc2 ; t Þ−U n ðc1 ; t Þg ð0 ≤γ ≤1Þ
2
This equation means the n-th leveling of the operating time of the
two cranes, where n is a natural number. Un̂ ðc1 ; c2 ; tÞ means the value
of operating time which shifts from c1 to c2 for the (n + 1)th time.
This value is normally defined by the product of constant value: γ and
shared operating time between c1 and c2:USH n (c1, c2, t). However, in
this shift, the upper limit of operating time of c2, which is described as U
ðc2 Þ, is necessary to be considered. Therefore, the shift value has to be
8 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx

(1) utilization rates of a crane TC-1 (3) utilization rates of a crane TC-3

TC-1 TC-3

TC-2
TC-1

(2) utilization rates of a crane TC-2 (4) utilization rates of a crane TC-4

TC-2 TC-3
TC-4

TC-2
TC-1 TC-1

Fig. 17. Graphs showing utilization rates of cranes.

less than Uðc2 Þ  U n ðc2 ; tÞ which means the remaining capacity of
operating time of c2.
Additionally, to avoid shifting shared operating time too much,
1
2 fU n ðc2 ; tÞ  U n ðc1 ; tÞg is also set as a upper limit of the shift value.
If U n̂ ðc1 ; c2 ; tÞ ¼ 1 fU n ðc2 ; tÞ  U n ðc1 ; tÞ g, then Un+1(c1,t)=Un+1(c2,t)
2
holds.
Fig. 19 shows the image of shift of operating time for two cranes.
4. Experiments and discussion
The simulation system was developed as an add-on program for
Navisworks (2014 Manage), and it was validated through computation-
al experiments performed in the following environment: CPU: Intel
Core i7-3930K 3.20 GHz; OS: Windows 7 Professional; and Graphics:
NVIDIA Quadro 2000D. The targets quantitatively evaluated were the
speed and accuracy of calculation by a function for mapping data be-
tween a crane and 3D objects and the accuracy of the function for
predicting quantities of piping spools. Data of previous projects were
used for the experiment.
4.1. Mapping between a crane and 3D objects
We evaluated calculation time for mapping between a crane and 3D
objects though the test case as shown in Fig. 20. In this case, the number
of 3D objects which are within the operating area of the crane was
64,156. The 3D objects are equipments, piping spools, and building
frames.
We used the native interference checking function in Navisworks as
a comparative method. This function calculates geometrical interfer-
ence strictly between the crane operating radius and materials of the
building. The result is listed in Table 2. In this table, the proposed meth-
od based on AABB reduces the calculation time from 527 s (by the native
function) to 0.62 s.
As for this result, considering the number of cranes in an NPP is
between 10 and 15, the proposed method takes about 10 s to calculate
for all cranes. This is significantly quicker than the 2 hours taken by the
native function.
Also we evaluated the time for calculating distribution of utili-
zation rates of a crane at whole construction site as shown in
Fig. 17. We compared serial and parallel computing (multi-thread
programming on CPU) to calculate a utilization rate at each posi-
tion. The results are shown in Fig. 21. The horizontal axis repre-
sents the mesh size by which the site was split. The size of the
construction site was about 250 × 200 m. In the case that the
mesh size is 5 m, the number of the crane positions was about
2000, and the parallel method takes about 27 min to calculate. As
for calculating a distribution of utilization rates in a whole site, it
seems necessary to introduce GPU programming for further reduc-
tion of calculation time.

10
Utilization rate [h/day]

6
TC-1

4
shared with
2 TC-2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

Fig. 18. Breakdown of utilization rates of TC-1.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx 9

10

Utilization rate [h/day]

8
TC-1
6 before
TC-1
after
4
TC-2
before
2
TC-2
after
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

Fig. 19. Leveling of utilization rates of the two cranes (TC-1 and TC-2).

The evaluated accuracies of the mapping function are listed in of trees created in RF was 300, and the number of parameters selected
Table 3. The accuracy was evaluated in terms of the difference between randomly in each node was 3.
the mapping results calculated by the native function and the proposed As for comparative methods, a method using average length of
method. Cover rate (CR) and wrong-detection rate (WDR) were all spools in a building, a method using average length of all spools
determined from the following equations: composing each NPP system, and the proposed method were experi-
mentally compared.
A Two types of results on accuracy of prediction of pipe quantities are
CR ¼ ð3Þ listed in Table 4. The left column is the average error rate between total
AþB
pipe quantities in each building in terms of actual ones and predicted
ones. The error rate of the proposed method is 2.1%. The right column
C is the average error rate of a pipe quantity in each area. The error rate
WDR ¼ ð4Þ
AþC of the proposed method is 20.7%.

Here, A is the number of common mapping results given by

both the native function and the proposed method, B is the number
of mapping results given by the native function excluding A, and C
is the number of mapping results given by the proposed method
excluding A. Note that larger CR is better, and smaller WDR is
better. 160 160
calculation time [min]

As shown in Table 3, CRs were confirmed to be over 99% and WDRs 140
were confirmed to be under 0.22% in the case of each method. 120
As for the influences of these errors, the error in utilization rate of 100
each crane is calculated to be lower than 30 min/day, which is consid- 80
80
ered to be small enough to be ignored. serial
60
40 parallel
4.2. Prediction of quantities of piping spools 40 26.6
13.3 20
20 6.6
0 3.3
To validate the function for predicting quantities of piping spools, a
5 10 20 40
previous project was used for learning data, and another one was used mesh size [m]
for test data. The input parameters were length of a piping line, pipe
diameter, type of building, the name of a NPP system, and number of Fig. 21. Graphs showing calculation time between the serial method and the parallel one.
bends in the piping line. The output parameter was average length of
spools of a piping line. CART was used to create tree in RF. The number

crane-2
Building crane-3
64,156 objects
-Equipment
-Piping spool
-Building frame
crane
50m crane-1
crane-4

Fig. 20. Test case to evaluate calculation time for mapping. Fig. 22. Positions of cranes used in the test.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
10 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx

crane-1
12

Utilization rate[hours/day]
10

6
actual
4 simulation
2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

crane-2
12
Utilization rate [hours/day]

10

6
actual
4 simulation
2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

crane-3
12
Utilization rate [hours/day]

10

6
actual
4
simulation
2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

crane-4

14
Utilization rate [hours/day]

12
10
8
6 actual
4 simulation

2
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

Fig. 23. Utilization rates of each crane between actual and simulated ones.

4.3. Calculation of crane utilization rates The construction schedule used in this test was the early version of
the previous project, in which detailed activities and their schedules
To validate the accuracy of the calculated utilization rates of cranes, were not planned.
we have compared predicted utilization rates with actual ones in a pre- Fig. 22 shows the positions and sizes of operating radius of cranes
vious construction project. used in this test.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx 11

crane-4 Table 2
14 Calculation time of mapping.
Ulitization rate [hours/day]

12
Native function of Navisworks Proposal method
10 shared with
8 crane-3 Calculation time (s) 527 0.62
shared with
6
crane-2
4 shared with
crane-1 Table 3
2
simulation Accuracy of mapping.
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 Wrong detection
Cover rate (%)
Month rate (%)

Equipment 99.1 0.11

Fig. 24. Utilization rates of crane-4 and its shared ones with other cranes. Pipe 99.6 0.05
Building frame 99.8 0.22

The results are described in Fig. 23. This figure shows monthly
transition of utilization rate for each crane, which means an average
Table 4
operating time a day [hours/day] in each month.
Accuracy of prediction of pipe quantities.
Regarding to the result of crane-4, the difference was bigger than
other cranes. This came from two reasons. First one was that the Average error rate for Average error rate for
Method type
buildings (%) areas (%)
operating radius of crane-4 was bigger than the ones of other cranes.
Second one was that we used average resource assignment, i.e. operat- Average length in each building 3.8 35.0
ing time required for carrying a 3D object, which is carried by multiple Average length in each NPP system 6.5 33.8
Proposed method 2.1 20.7
cranes, was assigned equally to each crane without considering their
load capacity as defined by Eq. (2).
To reduce the overload of crane-4, we tried to level the assignment
of shared operating time (i.e. utilization rate). Fig. 24 shows the shared utilization rates of the two cranes was leveled and the difference
operating time included in the operating time of crane-4. From this between actual and simulated utilization rate was reduced.
figure, the shared operating time between crane-4 and crane-1 was Regarding to the results of crane-1 and crane-3 in the last some
confirmed to be the biggest. months, the errors between simulated utilization rates and actual ones
Therefore, we modified the utilization rates of the two cranes by was bigger than other months. This is because removal of the cranes,
leveling the shared operating time using Eq. (4). The result is shown which become no longer needed, was not considered in this simulation.
in Fig. 25. In this figure, utilization rates before and after the modifica- Excluding the errors in these months, the absolute mean error of all
tion and actual utilization rate are described. It was confirmed that the cranes was 1.48 hours/day.

crane-1
10
Utilization rate [hours/day]

8 actual

6
simulation
4
simulation with
2 levelling
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

crane-4
14
Utilization rate [hours/day]

12
actual
10
8
simulation
6
4 simulation with
2 levelling
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Month

Fig. 25. Utilization rates after leveling of crane-1 and crane-4.

Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
12 Y. Sugimoto et al. / Automation in Construction xxx (2016) xxx–xxx
Level of detail (LOD) of the construction schedule, which was used
in this experiment, was rough. In case where more detailed schedule
(i.e. close to actual schedule) is used, results of simulation become
more accurate. Therefore, in an actual project, simulation should be
done several times according to refinement and modification of
construction schedule.
5. Conclusion
A support system for planning of crane deployment, which can eval-
uate transitions of utilization rates, was developed. It applies 4D CAD
technology to calculate utilization rates from combinations of 3D
models, construction schedules, and crane deployment plan. To calcu-
late automatically and accurately, two functions were developed. The
first function maps data between 3D objects, cranes, and schedules by
using a method called “axis-aligned bounding box.” The second function
estimates quantities of installation material required in the early
construction-planning phases by using a data mining technique called
“random forests.” The simulation system including these two functions
was developed as an add-on program for a commercial 3D model
viewer named “Navisworks” (produced by Autodesk).
The proposed system was confirmed to be effective by verification
tests. According to the results of these tests, the calculation time of the
first function for linking data between a crane and 3D objects in a NPP
model was 0.62 s, compared with 527 s for the native function of
Navisworks. As for the second function, the average error rate of a
total quantity of piping spools in each building was 2.1%, compared
with 3.8% by a method using average length of all piping spools in
each building (data of a previous project). The average error rate of a
quantity in each area was 20.7%, compared with 35.0% by the method
using average length. Finally, overall verification test, which compared
the utilization rates calculated by the proposed method with the actual
ones in a previous project, was conducted. As a result, the absolute
mean error of all cranes was 1.48 hours/day.
Please cite this article as: Y. Sugimoto, et al., 4D CAD-based evaluation system for crane deployment plans in construction of nuclear power plants,
Automation in Construction (2016), http://dx.doi.org/10.1016/j.autcon.2016.04.004
These results indicate that the proposed simulation system will sig-
nificantly reduce the time required to create crane-deployment plans
and improve their accuracies.
The company names and product names described in this paper are
trademarks or registered trademarks of their respective companies.
References
[1] K. Morita, K. Akagi, R. Miyahara, J. Miura, M. Ishiwata, S. Yamanari, Y. Yoshimoto,
Advanced Construction Technology for Shika Nuclear Power Station Unit No. 2 of
the Hokuriku Electric Power Company, Proceedings of the 15th Pacific Basing
Nuclear Conference (PBNC), 2006.
[2] K. Akagi, K. Morita, R. Miyahara, K. Murayama, The latest application of Hitachi's
state-of-the-art construction technology and further evolution towards new build
NPP projects, Proceedings of the 29th Annual Conference of the Canadian Nuclear
Society (CNS), 2008.
[3] C. Huang, C.K. Wong, C.M. Tam, Optimization of tower crane and material supply
locations in a high-rise building site by mixed-integer linear programming,
Autom. Constr. 20 (2010) 571–580.
[4] L.C. Lien, M.Y. Cheng, Particle bee algorithm for tower cranes layout with materials
quantity supply and demand optimization, Proceedings of the 30th ISARC 2013,
pp. 66–74.
[5] J. Wang, X. Zhang, W. Shou, X. Wang, B. Xu, M.J. Kim, P. Wu, A BIM-based approach
for automated tower crane layout planning, Autom. Constr. 59 (2015) 168–178.
[6] D. Heesom, M. Mahdjoubi, Trends of 4D CAD applications for construction planning,
Constr. Manag. Econ. 22 (2004) 171–182.
[7] J. Zhang, Y. Zhang, Z. Hu, M. Lu, Construction management utilizing 4D CAD and -
operations simulation methodologies, Tsinghua Sci. Technol. 13 (2008)
241–247.
[8] L.S. Kang, S.Y. Park, H.S. Moon, C.H. Kim, Improvement of basic functions for visual-
izing schedule and 3D object in 4D CAD system, CIB World Building Congress 2007,
vol. 13 2008, pp. 241–247.
[9] Z. Mallasi, Identification and visualisation of construction activities’ workspace
conflicts utilising 4D CAD VR tools, 1st ASCAAD International Conference 2004,
pp. 235–253.
[10] L.S. Kang, S.K. Kim, H.S. Moon, H.S. Kim, Development of a 4D object-based system
for visualizing the risk information of construction project, Autom. Constr. 31
(2013) 186–203.
[11] L. Breiman, Random forests, Mach. Learn. 45 (1) (2001) 5–32.