Towards the Data-Driven Design and Development of Materials Derived from Unstable Mixtures

The field of polymer blending, which aims to combine the properties of polymers to create specialised materials, has the potential to play an important role in helping to tackle some of the major challenges that society faces in the 21st century. However, to realise this in a sustainable and time-effective manner, we must advance our knowledge of structure-property relationships and find ways to leverage these relationships in the design and development process of new materials. One obstacle to developing structure-property relationships is the challenge of characterising complex phase-separated microstructures using accessible experimental data. To overcome this, we recently proposed a method for succinctly describing such microstructures in terms of well-defined and quantifiable characteristics from scattering data using machine learning [1], namely Gaussian process regression. Our findings, which we obtained using numerical simulation data, suggest there is an opportunity for a more complete characterisation of experimental phase-separated microstructures using scattering data. Scattering is a powerful technique for monitoring phase separation in real-time. As such, the careful application of our method to future predictions of the scattering data could allow one to predict the future characteristics of the microstructure as it evolves. In the absence of a satisfactory nonlinear theory describing the time evolution of the scattering data during phase separation, we have demonstrated the ability of dynamic mode decomposition, a data-driven modelling method, to make accurate future predictions of numerical scattering data. We aim to combine these two works to develop a control system to manipulate phase separation to produce new materials with prescribed microstructures and, therefore, properties. We also aim to expand the scope of our work to systems other than polymer blends, for example, metallic alloys and ceramics.<br/><br/>[1] M. Jones and N. Clarke, Machine Learning Real Space Microstructure Characteristics from Scattering Data, Soft Matter, 14 (42) : 9689-9696, 2021