We are currently working to investigate the role of machine learning algorithms in predicting the overall response of non linear elastic and inelastic composite materials, moving from the geometric and constitutive properties of their microstructures. Replacing computationally expensive homogenization techniques with machine learning algorithms could reveal to be extremely promising given the importance of composites in numerous engineering sectors. Composite materials in fact can be tailored to meet specific design requirements and their use is becoming increasingly attractive to fulfill industry needs due to their mechanical and physico-chemical properties.
It is worth emphasizing that the capability to make full use of extensive data is of fundamental importance in the application of machine learning to materials science research. In this regard numerous efforts have made by the scientific community in order to discover ways to overcome the shortcomings that both computational simulations and experimental measurements could involve in terms of time and costs. For example with the introduction of the Materials Genome Initiative (www.mgi.gov) in 2011 and the coming of the big data era, a great work has been done to collect extensive data sets on materials properties in order to provide a fast access to the properties of know materials. Machine learning is a powerful tool for discovering patterns in such a framework and in recent years the modeling of complex relations between physical factors and materials properties has proved to be successful thanks to machine learning techniques. The application of machine learning in materials science concerns mainly material properties prediction, at the macro and micro scales, the discovery of new materials, and numerous other purposes such as process optimization, density functions approximations, monitoring of batteries and prediction of fatigue crack growth rate to cite a few.