Materials Today: Proceedings
Materials Today: Proceedings
Materials Today: Proceedings
a r t i c l e i n f o a b s t r a c t
Article history: Artificial Intelligence has become the backbone of almost every domain of science and engineering.
Received 19 October 2020 Machine learning, the branch of AI adopts probabilistic and statistical methods to learn from the past
Received in revised form 28 October 2020 experience based upon the experimental output data set and detect the possible solution. In this paper,
Accepted 1 November 2020
an overview of various machine learning algorithm used so far for the prediction of various problems
Available online xxxx
such as optimization of process parameters, ranking of materials, validation is discussed. The process
of design and optimization of the fibre reinforcement in polymer composites with distinguished proper-
Keywords:
ties has been redefined by the machine learning approach. This paper also highlights the role of machine
Polymer composite
Machine learning
learning algorithm, solution techniques and their data bases used in the different stages starting from the
Artificial intelligence selection of raw materials to the end user application for the fiber reinforced polymer composites. This
Soft computing paper also supports readers to understand the future course of action to implement for the development
Numerical analysis of new product generation in an industry. At the end, a comparison has been made to understand the
functionality of machine learning with respective to other technical tools used in the real-world problem.
Ó 2020 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Confer-
ence on Advances in Materials Processing & Manufacturing Applications.
1. Introduction and storage gets larger. Data science and machine learning tech-
niques have been rapidly evolving over the past few years, and
Quick advancements in data science, mobile technology, com- these are progressively being implemented in the domain of
puter hardware, space, energy, defense and various different sec- material science.
tor have enforced importance of novel materials and their In today’s era of Artificial Intelligence (AI) and Machine learning
development. With increase in the high-performance parallel (ML), significant advancements are accomplished not only by con-
computing and computational modelling, the use of numerical ventional artificial intelligence researchers, but also by specialists
simulations with acceptable precision can now quantify many in other areas who use these techniques to achieve their own goals
critical material properties. In general, it is much quicker and less and these are currently being explored for a number of
costly to conduct simulations to predict the properties of a mate- applications.
rial than to synthesize, develop, and test the material in a labo- Machine learning, a division of AI, uses a range of statistical and
ratory. In recent days, to predict the material properties probabilistic approaches, allowing computers to learn from experi-
researchers are performing simulation using workstation or com- ence and to identify hidden patterns (input–output correlations)
puter cluster based upon the computational cost. The capacity to from large and often noisy data sets. Machine learning (ML) is
gather and analyze big data sets increases as machines get faster now seen as a successful approach for the design and discovering
of new materials for a wide variety of applications. Machine learn-
ing was introduced in 1959 by Samuel and it is now frequently
⇑ Corresponding author.
employed in the fields of computer vision, gaming, economics,
E-mail address: vikaskukshal@nituk.ac.in (V. Kukshal).
data-mining and bioinformatics.
https://doi.org/10.1016/j.matpr.2020.11.026
2214-7853/Ó 2020 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Conference on Advances in Materials Processing & Manufacturing Applications.
Please cite this article as: P. Pattnaik, A. Sharma, M. Choudhary et al., Role of machine learning in the field of Fiber reinforced polymer composites: A pre-
liminary discussion, Materials Today: Proceedings, https://doi.org/10.1016/j.matpr.2020.11.026
P. Pattnaik, A. Sharma, M. Choudhary et al. Materials Today: Proceedings xxx (xxxx) xxx
In today’s era of artificial intelligence and machine learning, sig- regression, k-nearest neighbor, support vector machine (SVM),
nificant advancements are accomplished not only by conventional Markov process, and Gaussian process (GP). At the last, the over-
artificial intelligence researchers, but also by specialists in other view and potential prospect of soft computing is discussed in the
areas who use these techniques to achieve their own goals. Earlier field of polymer composites.
machine learning was used to detect the solubility of C60 but now
ML has been applied for the prediction of molecular properties of
new materials. Experiments and testing performed on the conven-
2. Soft computing
tional machine plays a vital role in characterization of novel mate-
rials. The tools required to run the machine learning algorithm are
Recent advances in computer hardware have encouraged the
easily available and free to access as it depends upon the open
creation of more efficient frameworks in artificial intelligence tech-
source platform. ML helps in reducing the computational time
niques. A study of the current literature could easily confirm this,
and cost incurred during the experiment.
but could also show the increasing focus of the scientific commu-
Role of artificial intelligence and its application in the area of
nity from the engineering point of view on the relatively new area
composite materials have been conducted by many researchers
of soft-computing. Dr. Zadeh introduced the concept of Soft Com-
as presented in table 1. A detailed review was conducted on the
puting (SC) in the 1990 s. Soft Computing is an emerging set of arti-
evolutionary optimization techniques especially genetic algorithm
ficial intelligence methods to use the toll of inexactness and
(GA) for the composites [1]. The mechanical properties of short coir
complexity inherent in human thought and real-life problems to
fiber reinforce vinyl ester composite were optimized using the
provide stable, reliable, and optimal. The principal methodology
application of GA [2]. The tensile properties of natural fiber bio
of soft computing includes artificial intelligence, fuzzy computing,
composites were optimized by the use of three-factor-level Box-
neural network, evolutionary computing, and machine learning. SC
Behnken experimental design [3]. The investigation was also car-
techniques can be introduced as separate methods or embedded in
ried on the elastic modulus of polymer nano composites by the
hybrid and simplified architectures. The integration of SC tech-
application of artificial intelligence [4]. The prediction of cutting
niques induces a revolutionary change in engineering and science
force and delamination on CFEP composites was done by the use
fields by providing the solution to the problem that could not be
by Back propagation neural networks (BPNN) [5]. The geometrical
solve using the traditional computing methods [17].
parameters of satin reinforced multi-layer composites were opti-
mized using the genetic algorithm [6] and optimal wear settings
for minimum wear performance of polymer composites were also
simulated by the use of artificial intelligence techniques [7,8]. Mul- 3. Machine learning
tiple investigation on the finding and simulating the properties of
polymer composites by the use of different soft computing tech- Machine learning is branch of artificial intelligence (AI), which
niques. While optimization methods are computationally effective, enables systems to learn and develop experience automatically
the optimal solution depends upon the initial arrangement i.e. without being specifically programmed. Machine learning focuses
design variables during the optimization process. Due to this the on computer programmes that access and use data for themselves.
solutions derived from these optimization approaches [9–11] The learning process starts with feedback or information along
therefore not only differ from initial set up to initial setting but with examples, practical experience or direction to search for
may also be caught in local minima or critical points in certain cir- trends in data and, based on the examples, to make informed deci-
cumstances. It is therefore important that alternative methods are sions in the future. The main objective of machine learning is to
explored to allow the reverse design of materials. allow computers to automatically learn and adapt actions without
The main focus of this paper is to discuss the possible use of human intervention or assistance. Machine learning can be applied
machine learning in the area of fiber reinforced polymer compos- to all the domains where relationship between input and output is
ites. This paper emphasis on the applications of machine learning dependent. Machine learning algorithms learn from data. It is
models to predict physical, mechanical, tribological, thermal, therefore very much important to choose the right data and pre-
thermo-mechanical properties of fiber reinforced polymer com- pare them accordingly to enable the problem to be solved effec-
posites as well as applying machine learning models to optimize tively. The different machine learning models used in forecasting
the design of experiments of the said composites with desired based upon the input data are highlighted under the classification
properties. The present work also includes the review of type of of supervised learning, unsupervised learning and reinforcement
machine learning with some basic ML algorithms including linear learning represented in Fig. 1.
Table 1
Overview of various optimization techniques used in the field of polymer composite.
2
P. Pattnaik, A. Sharma, M. Choudhary et al. Materials Today: Proceedings xxx (xxxx) xxx
3
P. Pattnaik, A. Sharma, M. Choudhary et al. Materials Today: Proceedings xxx (xxxx) xxx
3.2.2. Gaussian mixture rewards. Different software and machines used it to find the best
This is the probabilistic model used for normal distribution behaviour or path possible in a particular situation. This learning
from an overall population by soft clustering approach. It is most varies from supervised learning, in that the answer is the key to
widely used for speech recognition and feature recognition [28]. the supervised learning so that the model is equipped with the
right answer, while in reinforcement learning there is no response,
3.2.3. Hidden Markov but the reinforcement agent decides on what to do to perform the
The Hidden Markov (HMM) models form a class of statistical given task. In the absence of a training dataset, he will learn from
models that assumes that the system is a Markov process with hid- his experience. The practical applications of reinforcement learning
den states. From the observed output sequences produced by the are robots in industrial automation [30].
Markov process, dynamic programming methods can be used for
both the output emission probabilities and the transition probabil- 4. Integration of Machine learning algorithms in polymer
ities between the hidden states [29]. composites
3.3. Reinforcement learning Experiments have generally played a central role in identifying
new materials and characterising them. For an incredibly small
Reinforcement learning is a different machine learning field. It number of materials, experimental testing must be performed over
is about taking sufficient action in a given situation to maximise a long period of time, since it creates high demands in terms of
4
P. Pattnaik, A. Sharma, M. Choudhary et al. Materials Today: Proceedings xxx (xxxx) xxx
5
P. Pattnaik, A. Sharma, M. Choudhary et al. Materials Today: Proceedings xxx (xxxx) xxx
behavior of polymer matrix composites. Materials (Basel) 2020;13. Doi: [22] Decision Tree in Machine Learning | by Prince Yadav | Towards Data Science n.
10.3390/ma13061381. d. https://towardsdatascience.com/decision-tree-in-machine-learning-
[13] V. Infante, JFA Madeira, R.B. Ruben, F. Moleiro, S.T. de Freitas, Characterization e380942a4c96 (accessed 19 October 2020).
and optimization of hybrid carbon–glass epoxy composites under combined [23] Weinberger KQ, Saul LK. Distance Metric Learning for Large Margin Nearest
loading, J. Compos. Mater. 53 (18) (2019) 2593–2605, https://doi.org/10.1177/ Neighbor Classification. vol. 10. 2009.
0021998319834673. [24] Machine Learning — Hidden Markov Model (HMM) | by Jonathan Hui |
[14] P.R. Pati, Prediction and wear performance of red brick dust filled glass–epoxy Medium n.d. https://medium.com/@jonathan_hui/machine-learning-hidden-
composites using neural networks, Int. J. Plast. Technol. 23 (2) (2019) 253– markov-model-hmm-31660d217a61 (accessed 19 October 2020).
260, https://doi.org/10.1007/s12588-019-09257-0. [25] Unsupervised Machine Learning: What is, Algorithms, Example n.d. https://
[15] P. Antil, S. Singh, A. Manna, Analysis on effect of electroless coated SiC p on www.guru99.com/unsupervised-machine-learning.html (accessed 19 October
mechanical properties of polymer matrix composites, Part. Sci. Technol. 37 (7) 2020).
(2019) 791–798, https://doi.org/10.1080/02726351.2018.1444691. [26] Unsupervised Learning - MATLAB & Simulink n.d. https://www.mathworks.
[16] A.A. BASEER, D.V. RAVI SHANKAR, M.M. HUSSAIN, INTERFACIAL AND TENSILE com/discovery/unsupervised-learning.html (accessed 19 October 2020).
PROPERTIES OF HYBRID FRP COMPOSITES USING DNN STRUCTURE WITH [27] T. Kanungo, D.M. Mount, N.S. Netanyahu, C.D. Piatko, R. Silverman, A.Y. Wu, An
OPTIMIZATION MODEL, Surf. Rev. Lett. 27 (02) (2020) 1950099, https://doi. efficient k-means clustering algorithms: Analysis and implementation, IEEE
org/10.1142/S0218625X19500999. Trans. Pattern Anal. Mach. Intell. 24 (2002) 881–892, https://doi.org/10.1109/
[17] Soft Computing. In Rescue when Conventional Algorithms. . . | by Ayang TPAMI.2002.1017616.
Laishram | Towards Data Science n.d. https://towardsdatascience.com/soft- [28] D. Reynolds, Gaussian Mixture Models, Encycl. Biometrics (2009) 659–663,
computing-6cef872f7704 (accessed 19 October 2020). https://doi.org/10.1007/978-0-387-73003-5_196.
[18] What are the types of machine learning? | by Hunter Heidenreich | Towards [29] Krogh A, È rn Larsson B, von Heijne G, L Sonnhammer EL. Predicting
Data Science n.d. https://towardsdatascience.com/what-are-the-types-of- Transmembrane Protein Topology with a Hidden Markov Model: Application
machine-learning-e2b9e5d1756f (accessed 26 October 2020). to Complete Genomes n.d. https://doi.org/10.1006/jmbi.2000.4315.
[19] R.H. Inman, H.T.C. Pedro, C.F.M. Coimbra, Solar forecasting methods for [30] Kaelbling LP, Littman ML, Moore AW. Reinforcement learning: A survey. J.
renewable energy integration, Prog. Energy Combust. Sci. 39 (6) (2013) 535– Artif. Intell. Res. 1996, 4, 237–85.
576, https://doi.org/10.1016/j.pecs.2013.06.002. [31] C.-T. Chen, G.X. Gu, Machine learning for composite materials, MRC 9 (02)
[20] Linear Regression Analysis - George A. F. Seber, Alan J. Lee - Google Books n.d. (2019) 556–566, https://doi.org/10.21203/rs.3.rs-97500/v1.
https://books.google.co.in/books?hl=en&lr=&id=X2Y6OkXl8ysC&oi=fnd&pg= [32] J. Wei, X. Chu, X.-Y. Sun, K. Xu, H.-X. Deng, J. Chen, Z. Wei, M. Lei, Machine
PR5&dq=linear+regression+scholarly+articles&ots=sehQE6nPgr&sig=rzQBN learning in materials science, InfoMat 1 (3) (2019) 338–358, https://doi.org/
RaKVTLAiPS5T5nKrNEYwHE#v=onepage&q&f=false (accessed 26 October 10.1002/inf2.12028.
2020). [33] Machine Learning Explained: Understanding Supervised, Unsupervised, an n.d.
[21] 1.4. Support Vector Machines — scikit-learn 0.23.2 documentation n.d. https:// https://datafloq.com/read/machine-learning-explained-understanding-
scikit-learn.org/stable/modules/svm.html (accessed 19 October 2020). learning/4478 (accessed 26 October 2020).