Artificial Neural Network Based Graphical User Interface for Estimation of Fabrication Time in Rig Construction Project

IJSRD - International Journal for Scientific Research & Development| Vol. 1, Issue 4, 2013 | ISSN (online): 2321-0613
All rights reserved by www.ijsrd.com 1011
Abstract—This paper addresses the problem of estimation of
fabrication time in Rig construction projects through
application of Artificial Neural Network (ANNs) as this is
the most crucial activity for successful project management
planning. ANN is a non-linear, data driven, self adaptive
approach as opposed to the traditional model based
methods, also fast becoming popular in forecasting where
relationship between input and output is not known but vast
collection of data is available. Around 960 data regarding
fabrication activity has been collected from ABG Shipyard
Ltd., Dahej. 3 input parameters have been considered for
estimation of output as fabrication time. 11 Feed Forward
Back Propagation neural networks with different network
architectures were made. Network N10 was able to predict
the output with MSE 1.35337e-2. Coding was done for the
Graphical User Interface (GUI) so that the GUI runs,
simulates network N10, and displays the fabrication time
for different combination of inputs.
Keywords: Fabrication time, Artificial Neural Network,
Graphical User Interface, GUI coding.
I. INTRODUCTION
In the construction projects (ex. Rig building) it is crucial to
minimize risks in the project estimation phase. This is an
early project stage in which different resources are
estimated. One of the important estimations is also the
necessary number of fabrication/construction hours or days.
The estimation phase is commonly a human expert driven
(intuitive method) activity which is sensitive to the expert’s
bias (judgement / experience). This bias can lead to an
underestimation of project resources, when the estimator is
overconfident, or to over-estimation of project resources
when the estimator does not have sufficient confidence that
all aspects of the project can be properly covered. Both
scenarios, based on the expert’s estimation, have a negative
impact on the future business decisions. In case of
underestimation, the project will bring economic loss, and in
case of overestimation, it will most likely be assigned to a
competitive supplier. The estimator’s key competence is to
properly collect and evaluate all the information which is
significant for making the project estimation. The paradigm
lies in the fact that the estimator should spend minimal time
necessary on estimation activity.
One of the main obstacles in this process is to
accurately define the relationship between product
characteristics and the construction/fabrication time
necessary to manufacture the product. Earlier studies
(Zhang and Fu, 2009) showed that the scientific and
reasonable performance evaluation is advantageous to
promote the comprehensive management level of
engineering projects. At present the fields of academia and
engineering had been achieved some results on this issue.
Iranmanesh and Zarezadeh (2008) stated that the researchers
have done lot of research in the area of application of
Artificial Neural Network (ANN) in project success, project
evaluation, project cost forecasting. As there are a few
studies on application of ANN in project estimation, there is
a good opportunity to accurately forecast time of fabrication
in construction projects through applying ANN approach.
The main aim of this paper is to develop the model
that fits into intuitive method for estimating fabrication
time. Three input variables are considered as height of job,
max. plate thickness of job and inspection criteria of job.
Inspection criteria is dependent on the value of max. plate
thickness. From this input value output as productivity
factor is decided by the intuitive method. Fabrication time in
number of days is calculated by dividing quantity of job
(tonnes) by productivity factor. We have developed detailed
step by step neural network model of the expert driven
estimation approach. Through this model, an effort is made
to capture the experience of the data available, which can be
further used to predict new combination of inputs. We have
also developed a Graphical User Interface(GUI) so that the
ANN model can be used without any prior knowledge of
ANN.
II. METHODOLOGY
A. Data
To train the ANN, 960 readings from already completed
jobs were collected from ABG Shipyard Ltd., Dahej,
Bharuch. Out of them 15 sample readings are shown below
(see Table I)
Sr
.
N
o.
Activity/
Job
Hei
ght
(m )
Max
Thick
ness
Of
Plates
(mm )
Inspec
tion
Criteri
a
(%)
Producti
vity
Factor
Quan
tity
Of
Activ
ity
(Ton
nes )
Durat
ion
(
Days
)
1 BOW 8.20 16.00
20.00
%
0.8
108.5
0
136
2 CS1 5.00 63.00
100.00
%
0.5
147.1
8
294
3 CS2P 8.00 80.00
100.00
%
0.5 50.65 101
4 CS3S 7.50 75.00
100.00
%
0.5 71.16 142
5 MD1 1.50 25.40
50.00
%
1 33.22 33
6 LB2P 2.00 22.50
50.00
%
1 26.65 27
7 LB2S 1.00 22.50
50.00
%
1 24.90 25
8 CS7 8.00 29.00
50.00
%
0.5 58.39 117
Artificial Neural Network Based Graphical User Interface for Estimation of
Fabrication Time in Rig Construction Project
D. H. Patel1
A. H. Makwana2
A. A. Mehta3
1, 2
Department of Mechanical Engineering, Government Engineering College, Dahod, India.
3
Department of Civil Engineering, Bhagwan Mahavir College of Engg. & Tech., Surat, India.

Artificial Neural Network Based Graphical User Interface for Estimation of Fabrication Time in Rig Construction Project
(IJSRD/Vol. 1/Issue 4/ 2013/0047)
9
SPUDCA
NFW
6.00 57.00
100.00
%
0.8
346.7
6
495
10 CSG2P 2.00 89.00
100.00
%
0.2 1.66 8
11 CSG3P 3.30 57.00
100.00
%
0.2 10.52 53
12 CSG4P 2.00 89.00
100.00
%
0.2 1.76 9
13 CSG5P 1.50 32.00
50.00
%
0.2 4.41 22
14 LQ5
10.0
0
19.00
20.00
%
0.8 90.62 113
15 LQ6 6.50 25.40
50.00
%
0.8 54.38 68
Min.(all 960
jobs)
1.00 6.00 20% 0.2 0.50
-
Max.(all 960
jobs)
30.0
0
90.00 100% 1 350
Table. 1: Data on 15 samples of activities from ABG
Shipyard Ltd., Dahej, Bhrauch
B. Application of Artificial Neural Network
Earlier studies (Jha) stated that ANNs are non-linear data
driven self adaptive approach as opposed to the traditional
model based methods. ANN is one of the branches of
Artificial Intelligence (AI). ANNs are powerful tools for
modelling, especially when the underlying data relationship
is unknown. ANNs can identify and learn correlated patterns
between input data sets and corresponding target values.
After training, ANNs can be used to predict the outcome of
new independent input data. ANNs imitate the learning
process of the human brain and can process problems
involving non-linear and complex data even if the data are
imprecise and noisy.
Neural network fitting tool (nftool-MATLAB
(7.8.0) R2009a) was used for creating the network. The
multilayer Feed forward back-propagation neural networks
were selected for the modelling as it’s the most common and
suitable for this study. Different parameters were carefully
selected to achieve the best performance.
11 different Neural Networks (N1 to N11) with
different neural network architectures were created. All the
networks consist of three layers of neurons with three
neurons for three inputs in input layer and one neuron for
one output in the output layer. The number of neurons in the
hidden layer varies in different neural networks (Fig. 1).
Fig. 1: Structure of the neural network
1) Transfer function
Each neuron has its own transfer function ƒ, which produces
the output a of that neuron based on the net input n from the
previous layer.
a = ƒ(n) …(i)
where, n = Σ (piwi +bi) …(ii)
p= scalar input
w=weight
b=bias
i = 1 to N,
N = Number of inputs.
Tan-Sigmoid & Pure-linear transfer functions were used as
an activation or transfer function. Tan-Sigmoid squashes the
output between -1 to 1 by using equation
ƒ (n) = [2 / (1 + e-2n
)] – 1 ...(iii)
and Pure-linear uses the equation
ƒ (n) = n ...(iv)
2) Training
All the networks were trained using Levenberg-Marquardt
back propagation algorithm which uses the following
equation to update weight and bias of the network:
...(v)
Where, xi = matrix of all weights and bias in ith
training
cycle
J = Matrix that contains first derivatives of the
network errors with respect to the weights and biases
μ = scalar value (μ α performance function)
I = Identity Matrix
All the network networks were trained for 1000 epochs
(training cycle). The initial value for µ was taken as 0.001. µ
is multiplied by µ decrease whenever a step would reduce
the performance function and multiplied by µ increase
whenever a step would increase the performance function.
3) Weights and Bias
Each neuron has its bias b and weights (w) equal to neurons
in the previous layer. Initial weights and bias were selected
randomly by the software. After the training, all the weights
and bias were saved for all the networks.
4) Performance function
At the end of each cycle, the performance of the network is
calculated by its performance function. Mean Squared Error
(MSE) was taken as the performance which uses the
following equation.
∑ ...(vi)
Where,
Error, e = actual output - network output
Q = no. of readings used for training = 672
5) Testing and Comparison
After 1000 training cycles, All the networks were simulated
with training set of 672 (70%) readings and unknown
validation and testing set of 144 (15%) and 144 (15%)
reading respectively. Comparison of networks was made on
the bases of MSE and Regression values given by MATLAB
for training and testing dataset (see Table II).
Network set
Network
Architecture
Input-hidden-output
(Transfer Functions)
MSE
Regression
value
N1
train
1
( t ) ( l )
2.62710e-
2
7.81577e-1
test
2.88462e-
2
7.82512e-1
N2 train
1 1
( t ) ( l )
2.68554e-
2
7.79208e-1

(IJSRD/Vol. 1/Issue 4/ 2013/0047)
test
2.24624e-
2
8.18538e-1
N3
train
1 1
( t ) ( l )
2.48288e-
2
8.07704e-1
test
2.62807e-
2
7.67928e-1
N4
train
2 1
( t ) ( l )
2.21643e-
2
8.26944e-1
test
3.05855e-
2
7.52987e-1
N5
train
2 1
( t ) ( l )
2.43133e-
2
8.07415e-1
test
2.55221e-
2
8.02421e-1
N6
train
1
( t ) ( l )
1.94257e-
2
8.46247e-1
test
3.57585e-
2
7.13734e-1
N7
train
1
( t ) ( l )
2.16734e-
2
8.31115e-1
test
2.24649e-
2
8.15882e-1
N8
train
1
( t ) ( l )
1.76361e-
2
8.65306e-1
test
2.05272e-
2
8.38827e-1
N9
train
1
( t ) ( l )
1.81221e-
2
8.57222e-1
test
2.56533e-
2
7.98515e-1
N10
train
( t ) ( l )
1.35337e-
2
9.0077e-1
test
1.95865e-
2
8.53534e-1
N11
train
1
( t ) ( l )
1.58471e-
2
8.78835e-1
test
2.53753e-
2
7.98257e-1
Table. 2: Performance comparison of 11 different network
after 1000 training cycles
t = tan-sigmoid
l = pure linear
C. Graphical User Interface Development
The GUI was developed using GUIDE (MATLAB's
Graphical User Interface Development Environment).
GUIDE stores GUIs in two files: MATLAB Figure file
(ANN.fig) and MATLAB M file (ANN.m), which are
generated when the GUI is saved or run for the first time.
1) GUI layout
GUI figure file (MATLAB figure file) contains the GUI
figure layout and the components of the GUI. There are 3
inputs and 1 output. For each input, three objects were
created : Static Text, Slider and Edit Text. Static text object
contains label of the input (Fig. 2).
Fig. 2: Objects in GUI
A desired value of an input can be entered in Edit text
object. Slider object can also be used to change the value
with slider. Slider and edit text objects have their value
range. For each slider and edit text, Min and Max value were
selected based on the training dataset. Inspection criteria is
dependent on max. plate thickness of the job so its value is
directly displayed by the tool depending upon the value of
max. plate thickness.
D. GUI Coding
GUI Code file (MATLAB M-file) contains the code that
controls the GU, including the call-backs for its components.
This is referred as GUI M-file. Complete coding was done in
such a manner that the GUI will not run the network N10
and display an error message if the inputs are not in range. A
call-back was assign to analyze the push button. If any of the
input parameters is not in range, an error message will be
displayed, otherwise the call-back will display the final
output as the number of days by simulating the network.
III. RESULTS
Among the 11 different network, network N10 with
architecture 3-50-1 was selected on the bases of least MSE
value (see Table II) 1.35337e-2. The performance graph of
network N10 with respect to the epochs (training cycle) was
generated. Epochs were shown on X-axis and network’s
performance (MSE) was shown on the Y-axis. The best
validation performance was 0.021178 at 93 epoch (Fig. 3).
Fig. 3 : Performance graphs of the network N10
IV. CONCLUSION
In this paper a neural network approach to estimate time of
fabrication in Rig construction project was studied.
Multilayer feed forward back propagation network can be
implemented successfully to estimate number of days of the
fabrication activity. The result show least errors (see Table
II) in 960 sample activities are acceptable and it can be
hopeful for researcher to applying this method with new
hypothesis on network structure and product characteristics.
It is notable that this study is an introductory study and can
be extended in various areas such as changing NN structure,
input variables, variable range and so on. The GUI
developed for the model would proved to be efficient to be
used by the end user without prior knowledge of ANN.
However, this research has some limitations. First,
the fabrication activity dataset is limited to one industry

(IJSRD/Vol. 1/Issue 4/ 2013/0047)
ABG Shipyard only. Therefore it is necessary to collect
more sample datasets from various-sized industries. Second,
the variables for estimating fabrication time are used with
restricted project inputs i.e. height, max. plate thickness and
inspection criteria of job that too within given range (see
Table I). Therefore, it is necessary to conduct more studies
using more input parameters and a wide range of each input
parameters. Third, GUI can be run by MATLAB's GUIDE
tool only. An independent software tool can be developed
which can run the network and the GUI which does not need
MATLAB to run the network.
REFERENCES
[1] Anish Kumar K.N, 'ENGINEERING,
PROCUREMENT AND CONSTRUCTION OF
JU2000A JACK-UP DRILLING RIG', ABG Shipyard
Ltd. Offshore Division training, 27th June 2008.
[2] Seyed Hossein Iranmanesh & Mansoureh Zarezadeh,
'Application of ANN to forecast actual cost of a project
to improve earned value management system', World
academy of science, engineering & technology, 2008.
[3] Jha Girish Kumar, 'Artificial Neural Networks and its
applications', I.A.R.I. New Delhi.
[4] Mehta A. A., Dr. Desai A. K., 'Artificial Neural
Network based graphic user interface for steel fibre
reinforced concrete flexural member', SVNIT, Surat,
2009.
[5] Qinghua Zhang and Qiang Fu, 'Performance evaluation
model of engineering project management based on
improved wavelet NN", J. Serv. & Management
(scientific research publishing) 2009.
[6] Karl Kuzman and Blaz Florjanic, 'Estimation of time for
manufacturing of injection moulds using ANN based
model', 2012.
[7] Chaudhari V R., 'Application of artificial neural
network in manufacturing in prediction of process
parameters and comparison with regression analysis',
Department of mech. Engineering, SVNIT, 2011.
[8] Zhigang Ji & Yajing Li, 'The application of RBE NN on
construction cost forecasting', Second international
workshop on knowledge discovery & data mining,
2009.
[9] Ismaail Elsawy, Hossain Hosny & Mohammed Abdel
Razek, 'A neural network model for construction
projects site overhead cost estimating in Egypt', IJCSI
international journal of computer science issue, vol. 8,
issue 3, no.1,2011.
[10]Seyed Hossein Iranmanesh & Mansoureh Zarezadeh,
'Application of ANN to forecast actual cost of a project
to improve earned value management system', World
academy of science, engineering & technology, 2008.
[11]Gwang-Hee Kim, Sung-Hoon An & Kyung-In Kang,
'Comparison of construction cost estimating model
based on regression analysis, NN & case-based
reasoning', Building and environment (science direct)
2004.
[12]Jing Xu & Jianguo Chen, 'Application study on the
performance evaluation for engineering project
management based on BP NN', International forum on
information technology and applications, 2009.
[13]Yu-Ren Wang & Chun-Yin Yu, 'Predicting project
success using ANN-Ensemble classification models',
IEEE,2011.
[14]Se Hun Lim & Kyungdoo Nam, 'ANN modelling in
forecasting successful implementation of ERP system',
International journal of computational intelligence
research, 2006.
[15]Qinghua Zhang & Qiang Fu, 'Performance evaluation
model of engineering project management based on
improved wavelet NN', J. Serv. & Management
(scientific research publishing), 2009.
[16]Ivo M.L. Ferreira & Paulo J. S. Gil, 'Application and
performance analysis of NN for decision support in
conceptual design', Experts systems with applications,
2012.
[17]Olanrewaju O.A., Jimoh A.A & Kholopane P.A.,
'Comparison between regression analysis and ANN in
project selection", IEEE, 2011.
[18]Krishnamoorthy C.S., Rajeev S., 'Artificial Intelligence
and Expert Systems for Engineers', CRC Press LLC
ISBN: 0849391253 Pub Date:08/01/96.

Artificial Neural Network Based Graphical User Interface for Estimation of Fabrication Time in Rig Construction Project

Recommended

More Related Content

What's hot (19)

Viewers also liked (18)

Similar to Artificial Neural Network Based Graphical User Interface for Estimation of Fabrication Time in Rig Construction Project (20)

More from ijsrd.com (20)

Artificial Neural Network Based Graphical User Interface for Estimation of Fabrication Time in Rig Construction Project