Machine learning and a computational fluid dynamic approach to estimate phase composition of chemical vapor deposition boron carbide

Abstract

Abstract Chemical vapor deposition is an important method for the preparation of boron carbide. Knowledge of the correlation between the phase composition of the deposit and the deposition conditions (temperature, inlet gas composition, total pressure, reactor configuration, and total flow rate) has not been completely determined. In this work, a novel approach to identify the kinetic mechanisms for the deposit composition is presented. Machine leaning (ML) and computational fluid dynamic (CFD) techniques are utilized to identify core factors that influence the deposit composition. It has been shown that ML, combined with CFD, can reduce the prediction error from about 25% to 7%, compared with the ML approach alone. The sensitivity coefficient study shows that BHCl2 and BCl3 produce the most boron atoms, while C2H4 and CH4 are the main sources of carbon atoms. The new approach can accurately predict the deposited boron–carbon ratio and provide a new design solution for other multi-element systems.

Keywords

• machine learning (ML)
• computational fluid dynamic (CFD)
• chemical vapor deposition
• boron carbide
• B/C ratio
• kinetic mechanisms