Symposium NM02: Nanotubes, Graphene and Related Nanostructures

Shun Muroga, National Institute of Advanced Industrial Science and Technology

Multimodal Artificial Intelligence System for Virtual Screening of Complex Nanocomposite Materials

Written by Henry Quansah Afful

Artificial intelligence (AI) has been very instrumental in the search for new materials having properties meeting different application requirements. Conventionally used models employ atoms and chemical bonds as materials descriptors, which cannot be readily extended to complex materials structures such as composites. Composite materials structures involve highly complex interactions between the fillers and matrix resulting from such phenomena as crosslinking, phase separation, and filler orientation amongst others. Shun Muroga introduced a multimodal AI model that integrates conventional AI with other materials descriptors such as the physical and chemical structure. Muroga studied over 80 polymer matrix composites from five matrices, two additives, and three fillers all with different volume fractions. The materials were characterized using optical microscopy, infrared spectroscopy, and Raman spectroscopy and the results fed into the AI model. The model was able to self-generate characterization data and predict the properties of over 100,000 composite materials conditions after the training process. This demonstrates how the multimodal AI approach efficiently accounts for the interplay of complex interactions in composite materials, rendering it more effective at materials prediction than the conventional models.