Scribd
Upload
43 views

Machine Learning in Materials Science

Uploaded by

CLINDON BRITTO (SHIBIN)
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
43 views

Machine Learning in Materials Science

Uploaded by

CLINDON BRITTO (SHIBIN)
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 7

Machine Learning In Materials Science

Need for Machine Learning In Materials Science


 Rapid developments in information science, energy, national defense, and other fields have vital diverse

requirements for materials.

 However, traditional methods for discovering new materials, such as the empirical trial, error method, density

functional theory (DFT)-based method and numerical simulations, typically require a long research and
development cycle.

 Such methods are of high cost with low efficiency and have difficulty keeping pace with the development of

materials science today.

 Machine learning can substantially reduce the computational costs and shorten the development cycle.

 Hence, it is one of the most efficient ways of replacing DFT calculations or even repetitive laboratory

experiments.

 As the resources and tools for machine learning are abundant and easy to access, the barrier to entry for

applying machine learning in materials science is lower than ever.


Modelling of materials using Machine Learning
• Algorithms are available for implementing the four types of machine learning methods can be divided into two
types
 Shallow learning
 Deep learning
 Algorithms in shallow learning
 Support vector machine
 Naive Bayes classifier
 Decision tree
 Artificial neural network
 Algorithms in Deep learning
 Convolutional neural network
 Recurrent neural network
Applications of ML in Materials Science
Material property analysis
• Degradation detection
 Machine learning is more accurate and convenient than human judgment in material analysis for the detection of
metal corrosion and asphalt pavement cracking and the determination of concrete strength.
• Nanomaterials analysis
 With the development of artificial intelligence, machine learning plays an increasingly important role in the field of
nanomaterials.
 As early as 1993, the use of machine learning to study the solubility of C 60 was proposed.
 Machine learning has been widely used to predict the toxicity of nanomaterials, to discover new nontoxic
nanoparticles, to develop multi structure/single-property relationships of nanoparticles, to study quantum-mechanical
observables of molecular systems, to analyze chemical reactions of nanomaterials and to solve kinetic systems.
• Molecular property prediction
 High-throughput density functional calculations for molecular property prediction are highly time-consuming.
 As an alternative, machine learning is a feasible approach for the fast prediction of structures or properties of
molecules, compounds and materials.
 In addition, it can realize high accuracy.
Applications of ML in Materials Science (Contd)
Discovering new materials
Structure-oriented design
Many classification and regression algorithms can be applied to predict the chemical
composition of a material from its structure.
Perovskite is an important crystal structure in many fields.
Shuaihua used various regression algorithms (Gradient boosting regression, kernel ridge
regression [KRR], support vector regression, Gaussian process regression, DT regression,
and multilayer perceptron regression) to predict stable lead-free perovskite compounds
(C2H5OInBr3, C2H6NInBr3, NH3NH2InBr3, C2H5OSnBr3, NH4InBr3, and C2H6NSnBr3)
Oliynyk proposed a Heusler discovery engine that is based on the random forest
algorithm for identifying new full-Heusler compounds. This model realized a high true-
positive rate of 0.94 and successfully predicted 12 novel gallides, namely, MRu 2Ga and
RuM2Ga (M = Ti − Co), as Heusler compounds.
Applications of ML in Materials Science (Contd)
Element-oriented design
 Machine learning can be used to predict new compounds and their structures from an input composition.
 A probabilistic model that was based on an experimental crystal structure database was used to identify
209 new ternary oxides.
 Meredig constructed a machine learning model that was based on thousands of DFT calculation results
for predicting the thermodynamic stability of arbitrary compositions.
 The researchers built a large database of DFT calculation results and two predictive formation energy.
 Then, they used these two models to scan 1.6 million candidate ternary compositions, ranked the most
probable results by combining the two models and finally identified 4500 new stable materials.
 In a similar study, machine learning was used to study binary compounds.
 First, researchers used an unsupervised learning algorithm to separate 67 octet compounds into distinct
classes according to their crystal structures
 Second, the supervised learning algorithm was applied to identify the correct crystal structures of 55
compounds
 Finally, a regression algorithm was used to predict the melting points of 44 sub octet compounds.
Applications of ML in Materials Science (Contd)
Inverse design
 Deep learning has high potential in the inverse design of materials.
 Inverse design begins from the required functionality and searches for the ideal molecular structure that exhibits this
functionality.
 The method takes a functionality as input and outputs a molecular structure.
 Reconstructing molecule structures directly from molecule descriptors, the inverse design model translates molecule
descriptors into molecule identifiers.
 If the molecular structure of the ML output is stable and the molecular properties are consistent with the target, the
inverse design of molecules is successful
Drug design
 Drug design is one of the mature fields in which machine learning is utilized.
 Machine learning approaches that are modeled on small molecules can handle the structural complexity of
proteins and can predict structure-activity relationships accurately, which facilitates the discovery of target drugs.
 A typical process of machine learning for drug discovery begins with compounds that have already been tested.
 Then, batches of compounds are repeatedly designed and selected for parallel testing.
 The activity model is refined at each step to select the most promising compound for the next batch and the
process is repeated until an active drug has been identified

You might also like