(661c) Development of General-Purpose Machine Learning-Flame Spray Pyrolysis Methodology for Guided Synthesis

Flame spray pyrolysis (FSP) is a common technique for the synthesis of nanomaterials by combustion of their precursors inside flame. It is a continuous process, involving multiple stages during particle formation cutting across multiscale dimensions. However, the numerosity of its synthesis parameters as well as the precise control of the synthesis environment are the key features that allow for development of machine learning (ML) methodology for the accelerated discovery of high performing nanomaterials. In this study, ML-FSP methodology was developed by engineering relevant features from the fundamental understanding of the FSP process. Based on the features relating to synthesis environment provided by the spray flame during FSP, an unsupervised ML technique suggests that two possible values of a certain property of a material can be obtained by FSP with EHA/toluene solvent mixtures and fixed precursor concentration. These properties will be dependent on the materials being synthesized by the FSP technique. However, additional fundamental insights using physics-based model are being developed and incorporated to assist the machine learning algorithms for more accurate prediction. This paradigm could be applied to guide synthesis in order to achieve optimization of the phase of Ce-Zr-Mn systems for maximal activity and stability for oxidation reactions.