Click here to load reader
Upload
phamminh
View
220
Download
4
Embed Size (px)
Citation preview
Abstract—This paper addresses the issues and technique
predicting increasing cement resistance using machine learning
Techniques. Machine learning is a subfield of artificial intelligence
that deals with the construction and study of systems that can learn
from data, rather than follow only explicitly programmed
instructions. Artificial neural networks are known as intelligent
methods for modeling of behavior of physical phenomena. This
study, then, aimed to analyze the factors affecting the recovery of
neural network model to predict the resistance of cement and used
them to improve the resistance of cement.
Keywords—Machine learning, cement resistance, neural
network.
I. INTRODUCTION
ACHINE learning can be considered a subfield of
computer science and statistics. It has strong ties to
artificial intelligence and optimization, which deliver
methods, theory and application domains to the field. Machine
learning is employed in a range of computing tasks where
designing and programming explicit, rule-based algorithms is
infeasible. Example applications include spam filtering, optical
character recognition (OCR)[1]search engines and computer
vision. Machine learning is sometimes conflated with data
mining,[2] although that focuses more on exploratory data
analysis.[3] As a scientific endeavor, machine learning grew
out of the quest for artificial intelligence. Already in the early
days of AI as an academic discipline, some researchers were
interested in having machines learn from data. They attempted
to approach the problem with various symbolic methods, as
well as what were then termed "neural networks"; mostly
Perceptrons and other models that were later found to be
reinventions of the generalized linear models of statistics.
Probabilistic reasoning was also employed, especially in
automated medical diagnosis.[4]
An artificial neural network (ANN) learning algorithm,
usually called "neural network" (NN), is a learning algorithm
that is inspired by the structure and functional aspects of
biological neural networks. Computations are structured in
terms of an interconnected group of artificial neurons,
1 Mehrnoosh Ebrahimi was with Industrial Engineering Department
university of Science and Research Branch, Islamic Azad,(MA student) Iran,
Kerman([email protected]) 2 Ali_Akbar Niknafs is with computer department ,university of Bahonar
(assistant PhD) Iran, Kerman ([email protected])
processing information using a connectionist approach to
computation. Modern neural networks are non-linear statistical
data modeling tools. They are usually used to model complex
relationships between inputs and outputs, to find patterns in
data, or to capture the statistical structure in an unknown joint
probability distribution between observed variables.[4]
Previous researchers on chemical and physical resistance of
cement and cement quality, ingredients, and appropriate
physical or chemical influencing factors have been studied.
Many sophisticated mechanisms have been for building
strength, but none of them are obtained from numerical
methods. Artificial intelligence (AI) is the intelligence
exhibited by machines or software. It is also an academic field
of study. Major AI researchers and textbooks define the field
as "the study and design of intelligent agents,[5] where an
intelligent agent is a system that perceives its environment and
takes actions that maximize its chances of success.[6] John
McCarthy, who coined the term in 1955,[7] defines it as "the
science and engineering of making intelligent machines". The
Main articles: Neural network and Connectionism
Fig. 1 ANN dependency graph
Artificial neural network types vary from those with only
one or two layers of single direction logic, to complicated
multi–input many directional feedback loops and layers. On
the whole, these systems use algorithms in their programming
to determine control and organization of their functions. Most
systems use "weights" to change the parameters of the
throughput and the varying connections to the neurons.
Artificial neural networks can be autonomous and learn by
input from outside "teachers" or even self-teaching from
written-in rules.[8]
II. BACKGROUND
A study proposed a method of strength improvement of
cement–sand mortar by the microbiologically induced mineral
precipitation[9]The Ordinary Portland Cement (OPC) is one of
Predicting Increasing Cement Resistance Using
Machine Learning Techniques
Mehrnoosh Ebrahimi1, and Ali Akbar Niknafs
2
M
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
http://dx.doi.org/10.15242/IIE.E0115032 71
the main ingredients used for the production of concrete and
has no alternative in the civil construction industry.
Unfortunately, production of cement involves emission of
large amounts of carbon-dioxide gas into the atmosphere, a
major contributor for greenhouse effect and the global
warming, hence it is inevitable either to search for another
material or partly replace it by some other material.[10] The
search for any such material, which can be used as an
alternative or as a supplementary for cement should lead to
global sustainable development and lowest possible
environmental impact. Substantial energy and cost savings can
result when industrial by products are used as a partial
replacement of cement. Fly ash, Ground Granulated Blast
furnace Slag, Rice husk ash, High Reactive Met kaolin, silica
fume are some of the pozzolanic materials which can be used
in concrete as partial replacement of cement.[11]
III. INFERENCE FROM LITERATURE
According to the studied literature one of the most
important factors considered by Quality control experts and
analysts, is cement strength. Experts in quality control
laboratories have always been looking for ways to increase the
strength of the resulting cement production to increase
productivity and product quality.
IV. NEURAL NETWORK STRUCTURE
Neural networks are models of biological neural structures.
The starting point for most neural networks is a model neuron,
as in Figure 2. This neuron consists of multiple inputs and a
single output. Each input is modified by a weight, which
multiplies with the input value. The neuron will combine these
weighted inputs and, with reference to a threshold value and
activation function, use these to determine its output. This
behavior follows closely our understanding of how real
neurons work.
Fig. 2 Neural network structure
While there is a fair understanding of how an individual
neuron works, there is still a great deal of research and mostly
conjecture regarding the way neurons organize themselves and
the mechanisms used by arrays of neurons to adapt their
behavior to external stimuli. There are a large number of
experimental neural network structures currently in use
reflecting this state of continuing research.
To build a back propagation network, proceed in the
following fashion. First, take a number of neurons and array
them to form a layer. A layer has all its inputs connected to
either a preceding layer or the inputs from the external world,
but not both within the same layer. A layer has all its outputs
connected to either a succeeding layer or the outputs to the
external world, but not both within the same layer.
Fig. 3 Sample of neural network
Next, multiple layers are then arrayed one succeeding the
other so that there is an input layer, multiple intermediate
layers and finally an output layer, as in Figure 3. Intermediate
layers, that is those that have no inputs or outputs to the
external world, are called >hidden layers. Back propagation
neural networks are usually fully connected. This means that
each neuron is connected to every output from the preceding
layer or one input from the external world if the neuron is in
the first layer and, correspondingly, each neuron has its output
connected to every neuron in the succeeding layer.
V. TRAINING
Training was conducted with 70% of the input - output.
Weight training internal connection between neurons is set as
to minimize the mean square error training. After a complete
education system, the best weights obtained have been used for
the main experiments. The method used in the network is
described below:
Data is input neurons:
Ii={∑IJWji+bi} (1)
Before Training, all weights are randomly selected. Training
mode made the output achieves network weighs. The output is
compared to the target output and for all output neurons the
weights are updated.
VI. TRAINING AND ESTIMATE NETWORK
As for the operational experience, the variables listed in
Table 1 were selected. In order to achieve/approach the model
year, the data work on cement plant operation were collected.
After removing outliers attention to the steady state, operation
of the data were obtained from 244 data, selected for training
and testing neural network. Seventy percent of the data was
used for training, and 35% for the main test.
VII. SUMMARY SETTINGS USED IN THE NEURAL NETWORK
Neural networks predict a categorical target based on one or
more predictors by finding unknown and possibly complex
patterns in the data.
• Target: 28 days
• Prediction inputs: the parameters shown in table 1
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
http://dx.doi.org/10.15242/IIE.E0115032 72
• Neural networks models: Multilayer Perceptron (MLP)
• Hidden layers: 3
• Use maximum training time (per component model) : true
• Number of component models for boosting and / or
bagging:10
• Over fit prevention Set: 35%
• Replicate results: true
• Random seed: 229176228
• Records used in training: 244
VIII. ANN RESULTS ON THE STRENGTH OF CEMENT
TABLEI
RESULT OF AUDIT
FIELD MIN MAX MEAN STD.DV VALID
LOSS% 1.01 1.74 1.242 0.106 244
SIO2% 21.63 22.55 22.07 0.173 244
AL203% 3.51 5.3 4.608 0.641 244
FE203 % 3.61 4.72 4.018 0.277 244
CAO % 62.25 64.23 63.189 0.529 244
MGO % 1.6 2.1 1.805 0.108 244
S03 % 1.66 2.45 2.013 0.147 244
FREE
CAO %
0.62 1.98 1.287 0.295 244
SM 2.39 2.85 2.565 0.134 244
AM 0.8 1.4 1.162 0.226 244
LSF 85.9 92 88.464 1.387 244
C3S 38.1 57.4 47.07 5.725 244
C2S 19.6 34.5 27.781 4.173 244
C3A 1.8 7.6 5.414 2.137 244
C4AF 11 14.4 12.23 0.847 244
MESH
(170)
0.1 4.3 1.339 0.741 244
BLAIN
CM2/GR
2820 3554 3066.57 172.496 244
INIT MIN 110 190 155.246 15.835 244
FINAL
MIN
150 220 190.943 16.609 244
28 DAY 340 550 412.787 40.648 244
Initial value Training model was used to obtain the initial
weight. The mean absolute error (MAE) at 10% in the
prediction Training cements and improves the output
resistance.
TABLEII
RESULTS FOR OUTPUT FIELD 28 DAY
The Initial value Training model was used to obtain the
initial weights. The mean absolute error (MAE) at 10.059% in
the prediction Training improves cement resistance and the
output resistance.
TABLE III
PREDICTED AND ACTUAL OUTPUTTING
28_day $N 28_day
360 359.40881
365 382.26258
385 383.1868
360 366.22391
340 350.59334
365 362.70057
370 378.0902
370 367.39055
360 367.35613
370 390.49933
Fig 2 Importance input
The importance factors in this model as shown in fig 2.
IX. RESULT
Using the neural network model combined with the high
industrial data for network training can predict the strength of
cement. MPL neural network was used successfully to predict
cement resistance of 19 input variables. A good agreement
was observed between predicted and actual value of the
cement strength .Using the neural network model combined
with the high industrial data for network training can predict
the resistance of cement. MPL neural network was used
successfully to predict cement resistance of 19 input variables.
A good agreement was observed between predicted and actual
value of the cement resistance.
According to the analysis carried out, the following results
were obtained:
• Blaine is effective in increasing the amount of cement
resistance.
• Reduction of iron oxide will increase the resistance of
cement.
• Increases in Silicon oxide increases the resistance of
cement provided that the Blaine and the mesh are also declined
Exploration: In the first step of data exploration data is
cleaned and transformed into another form, and important
variables and then nature of data based on the problem are
determined.
REFERENCES
[1] Wernick, et al., Machine Learning in Medical Imaging. IEEE Signal
Processing Magazine, 2010. 27(4): p. 25_38.
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
http://dx.doi.org/10.15242/IIE.E0115032 73
[2] Heikki, M., Data mining: machine learning, statistics, and databases, in
Int'l Conf. Scientific and Statistical Database Management1996, IEEE
Computer Society.
[3] Friedman and H. Jerome, Data Mining and Statistics: What's the
connection. Computing Science and Statistics, 1998. 29(1): p. 3_9.
[4] McGraw, T. and H. Edition, Neural Networks in Computer Intelligence,
(2003): New Delhi.
[5] [Russell, et al., The intelligent agent paradigm. 2003 1998: p. 27, 32_55
7_21
[6] Russell, S. and P. Norvig. Artificial Intelligence: A Modern Approach
(2nd ed.). 2003. New Jersey: Prentice Hall.
[7] McCarthy, et al., A Proposal for the Dartmouth Summer Research
Project on Artificial Intelligence. (1955 26 August 2007.
[8] Maind, B.S. and P. Wankar, Research Paper on Basic of Artificial
Neural Network. International Journal on Recent and Innovation Trends
in Computing and Communication, 2014. 2(1): p. 96_100.
[9] Ghosh, P., et al., Use of microorganism to improve the strength of
cement mortar. Cement and Concrete Research, 2005. 35(10): p. 5.
http://dx.doi.org/10.1016/j.cemconres.2005.03.005
[10] Babu, D.V. and S. Nateshan, Investigations on silica fume concrete. The
Indian concrete Journal, 2004: p. 57_60.
[11] elaty, M.a.a.A., Compressive strength prediction of Portland cement
concrete with age using a new model. HBRC Journal, 2013.
[12] Mikut, et al., Data Mining Tools. Data Mining and Knowledge
Discovery, 2011)(1): p. 431_445.
[13] Perzyk, m., et al., comprarion in applications to process fault diagnosis.
information sience, 2013: p. 1_13.
[14] Tesema, T.B., A. Abraham, and C. Grosan, Rule Mining And
Classification of Road Traffic Accidents Using Adaptive Regression
Trees. On Simulation, 2008. 6.
[15] Thotappa, C., K. Ravindranath, and Data mining Aided Proficient
approach for optimal inventory control in supply chain management.
International Journal of Computer Science and Information Security,
2010. 8(2).
[16] S.Shankar and T.Purusothaman, Utility Sentient Frequent Itemset
Mining and Association Rule Mining: A Literature Survey and
Comparative Study. International Journal of Soft Computing
Applications, 2009(4): p. 81_95.
[17] Guyon, I., Elisseeff, and André, An Introduction to Variable and Feature
Selection. JMLR 3, 2003.
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
http://dx.doi.org/10.15242/IIE.E0115032 74