Machine Learning for Optimizing and Disrupting Thermal Transport Science

Machine learning has emerged as a promising tool for science and engineering, and in this talk we show our recent work in developing machine learning approaches to optimize or even disrupt thermal transport science.<br/> <br/>In the optimization case study, we demonstrate that an intuition-based manual search for aperiodic superlattice structures (random multilayers or RMLs) with the lowest thermal conductivity yields only a local minimum, while a genetic algorithm (GA) based approach can efficiently identify the globally minimum thermal conductivity by only exploring a small fraction of the design space. Our results show that this minimum value occurs at an average RML period that is, surprisingly, smaller than the period corresponding to the minimum SL thermal conductivity. Moreover, the lower limit of the thermal conductivity occurs at a moderate rather than maximum randomness of the layer thickness.<br/> <br/>In the disruption case study, we recognize that discovering exceptions has been a major route for advancing sciences but a challenging and risky process. Machine learning has shown effectiveness in high throughput search of materials and nanostructures, but using it to discover exceptions has been out of the norm. In this example we demonstrate the use of genetic algorithm to discover unexpected thermal conductivity enhancement in disordered 2D nanoporous graphene and 1D aperiodic superlattices, as compared to their periodic counterparts. Through structural analysis, we proposed that such unusual enhancement in 2D is due to the effect of shape factor and channel factor dominating over that of the phonon localization. In 1D, it is due to the coherent interfaces that suppress thermal resistance. Such findings not only provide insights in thermal transport in disordered materials but also demonstrate the effectiveness of machine learning to discover small probability events and the intriguing physics behind.