Welcome to Journal of Synthetic Crystals! Today is Share:

Journal of Synthetic Crystals ›› 2026, Vol. 55 ›› Issue (2): 291-300.DOI: 10.16553/j.cnki.issn1000-985x.2025.0207

• Research Articles • Previous Articles     Next Articles

Machine Learning Accelerated Prediction of Mechanical Properties in SiC Nanophononic Heterostructures

DU Yifan1(), LIU Jingsong2, JIANG Ping3(), REN Longjun3   

  1. 1.Intelligent Control Industry Department,Jimei Polytechnic Vocational College,Xiamen 361022,China
    2.Intelligent Systems Electromechanical Department,Xiamen Information School,Xiamen 361009,China
    3.Ma’anshan Engineering Technology Research Center of Advanced Design for Automotive Stamping Dies,Wanjiang University of Technology,Ma’anshan 243031,China
  • Received:2025-09-24 Online:2026-02-20 Published:2026-03-06

Abstract: To systematically investigate the coupled effects of temperature and pore geometry on the mechanical properties of SiC nanophononic heterostructures (NHs),molecular dynamics simulations combined with a random forest machine learning method are employed to analyze the fracture behavior of SiC NHs under uniaxial tension in the temperature range from 50 K to 500 K. The models feature different pore sizes and aspect ratios of rectangular phononic crystal pores,with interfaces constructed along the armchair and zigzag crystal orientations. The results show that,as temperature increases,the fracture strength and fracture strain of SiC NHs decrease by 20%~32% and 26%~35%,respectively. The pore length of the rectangular phononic pores along the loading direction is identified as the dominant geometric parameter controlling the fracture strength. SiC NHs with armchair interfaces exhibit fracture strengths approximately 15%~20% higher than those with zigzag interfaces,and larger phononic pore sizes significantly intensify local stress concentration and thermal softening effects. The random forest model constructed based on molecular dynamics simulation data achieves high accuracy (R2=0.99) in predicting the fracture properties of SiC NHs,while its computational efficiency is about 600 times higher than that of molecular dynamics tensile simulations. This work provides theoretical support and a fast prediction tool for the controllable fabrication and mechanical performance design of SiC NHs in SiC nano-devices.

Key words: SiC NH; mechanical property; molecular dynamics; random forest; machine learning

CLC Number: