Machine learning has become an excellent tool for scientists and engineers to predict, design, and fabricate next-generation material. Here, we report the thermal conductivity and thermal rectification of gradient-nano-grained graphene (GNGG) by molecular dynamic simulation with machine learning. It is revealed that the thermal conductivity of GNGG is mainly determined by the average grain size, while its thermal rectification factor varies linearly with the gradient of nanograins. Deep neural network-based machine learning models are developed to estimate the thermal transport properties of GNGG using microstructural signatures, such as the location, number, and orientation of 5|7 pairs. The results stress the pivotal roles of 5|7 defects in the planar thermal transports of graphene and indicate that high-performance 2D thermal rectifiers for heat flow control and energy harvesting can be achieved by bio-inspired gradient structure engineering. The findings are expected to supply a theoretical strategy for the design of bio-inspired materials and create a method to predict the potential properties of the material candidates by using machine learning, which can save the abundant expense of developing the material by using the classical method.

中文翻译：

---

作为二维热整流器的梯度纳米颗粒石墨烯：基于分子动力学的机器学习研究

机器学习已成为科学家和工程师预测、设计和制造下一代材料的绝佳工具。在这里，我们通过机器学习的分子动力学模拟报告了梯度纳米颗粒石墨烯 (GNGG) 的导热性和热整流。结果表明，GNGG的热导率主要由平均晶粒尺寸决定，而其热整流因子随纳米晶粒的梯度线性变化。开发了基于深度神经网络的机器学习模型，以使用微观结构特征（例如 5|7 对的位置、数量和方向）来估计 GNGG 的热传输特性。结果强调了 5|7 缺陷在石墨烯平面热传输中的关键作用，并表明用于热流控制和能量收集的高性能二维热整流器可以通过仿生梯度结构工程实现。研究结果有望为仿生材料的设计提供理论策略，并创建一种利用机器学习预测候选材料潜在特性的方法，从而节省使用经典方法开发材料的大量费用.