Skip to main navigationSkip to main content
The University of Southampton
Engineering

Research project: Applications of artificial neural networks (ANNs) in materials property correlations exploration

Currently Active: 
Yes

Artificial neural networks (ANNs) are currently one of the most powerful data mining techniques that have been widely applied in many fields, including chemistry, biology, materials science, etc. The general aim of this work is to explore correlations that might exist between different properties in materials through a neural network approach, and employ such correlation to solve problems that are not accessible to conventional methods or to better understand the underlining phenomena. The way of using ANNs to extract knowledge from collection of data and support analysis, is expected to accelerate progress in science and engineering research.

1) To prove that ANNs are capable of capturing meaningful property correlations without any prior knowledge or any assumptions of the form of the relationship made in advance, a combinatorial search has been conducted to explore cross property correlations in a prepared database that contain 37 pure metals with 24 properties available. Several universally recognized binary and ternary correlations have been successfully identified, which can be explained in a number of physical models.

2) Property correlations captured by ANNs can be used to detecting and correcting errors in handbooks and databases. An improved method has been proposed to apply to a complex situation, where ANNs are employed to capture and utilize the ternary relationships between metal's elastic properties. In addition, a revised table of metal's elastic properties data is provided.

3) Property correlations captured by ANNs between multiple/unknown numbers of variables can be used to highlight the variables' impacts on a target property. This is exemplified by using ANNs to predict the increment of hardness of pure metal due to high pressure torsion (HPT) from a limit set of parameters automatically chose from a large number of physical properties that may explain the changes.

4) Combined with genetic programming (GP), mathematic equations can be developed to quantitatively describe property correlations captured by ANNs from a statistical point of view. This is illustrated by determining the enthalpy of vaporisation at boiling point for 175 organic and inorganic compounds. ANNs are applied to spot the important input parameters and GP is employed to derive the equations representing these relationships.

Share this research project Share this on Facebook Share this on Twitter Share this on Weibo
Privacy Settings