講演情報
[14p-K507-14]Investigating Diamond Tool Wear in Iron Machining by Machine Learning Molecular Dynamics Simulation
〇(D)BaoAnh NguyenTrinh1, John Isaac G. Enriquez1, Harry Handoko Halim1, Hiroyuki Ogiwara2, Takahiro Yamasaki2, Masato Michiuchi2, Tamio Oguchi3, Yoshitada Morikawa1 (1.Department of Precision Engineering, Graduate School of Engineering, Osaka University, 2-1 Yamadaoka, Suita, Osaka, 565-0871, Japan, 2.Advanced Materials Laboratory, Sumitomo Electric Industries, Ltd, Itami, Hyogo, 664-0016, Japan, 3.Center for Spintronics Research Network (CSRN), Osaka University, Toyonaka, Osaka 560-8531, Japan)
キーワード：
diamond tool wear、Fe-C interfaces、machine-learning interatomic potential
Diamond tools are commonly used in ultra-precision machining due to their hardness, wear resistance, and ability to produce high-quality surfaces. However, when machining ferrous metals, diamond tool wear can become a significant concern, while its thermo-chemical mechanism remain not fully explained. We aim to clarify the Fe-C interface reactions by machine-learning interatomic potential molecular dynamics, which is a powerful tool compared to the traditional tools like ab-initio or classical molecular dynamics, for investigating surfaces and interfaces structures at the atomic and molecular level.
Our training dataset comprises approximately 6,000 training structures, with a training time of less than one day. The resulting model demonstrates high accuracy, achieving a force RMSE ~130 meV/angstrom and an energy RMSE ~4 meV/atom. The validity of the machine-learning is confirmed by its ability to reproduce the structural and energetic properties of single-element systems (pure bcc Fe and pure C), as well as the Fe-C alloys systems.
Large-scale cutting simulations involving over 8000 atoms were conducted with the timestep 1fs and cutting durations in the order of nanoseconds. Preliminary results reveal that using a diamond surface as the clearance face would cause more wear compared to its use as the rake face. Additionally, the wear rate could be reduced by maintaining low temperature for the cutting environment. Among the low-index diamond surfaces, diamond(100) shows the most resistant to wear. However, during the first few nanoseconds, the calculated wear rate is much higher than the experiment value, by 2-3 orders of magnitude. With extended simulation time, the experimental wear rate is expected to be approached gradually.
Our training dataset comprises approximately 6,000 training structures, with a training time of less than one day. The resulting model demonstrates high accuracy, achieving a force RMSE ~130 meV/angstrom and an energy RMSE ~4 meV/atom. The validity of the machine-learning is confirmed by its ability to reproduce the structural and energetic properties of single-element systems (pure bcc Fe and pure C), as well as the Fe-C alloys systems.
Large-scale cutting simulations involving over 8000 atoms were conducted with the timestep 1fs and cutting durations in the order of nanoseconds. Preliminary results reveal that using a diamond surface as the clearance face would cause more wear compared to its use as the rake face. Additionally, the wear rate could be reduced by maintaining low temperature for the cutting environment. Among the low-index diamond surfaces, diamond(100) shows the most resistant to wear. However, during the first few nanoseconds, the calculated wear rate is much higher than the experiment value, by 2-3 orders of magnitude. With extended simulation time, the experimental wear rate is expected to be approached gradually.