Optimal design of CH4 pyrolysis in a commercial CVD reactor using support vector machines and Nelder-Mead algorithm

• Published in: Chemical engineering research & design 2022-02, Vol.178, p.124-135
• Main Authors: Gupta, Anand, Mudgal, Abhisek, Shinde, Vijay M., Kumar, Harish, Prasad, N. Eswara
• Format: Article
• Language: English
• Subjects: Algorithms; CFD modelling; Chemical reactors; Chemical vapor deposition; Design optimization; Flow velocity; Gas mixtures; Inlet flow; Machine learning; Methane; Methane pyrolysis; Model accuracy; Nelder-Mead algorithm; Optimization; Process parameters; Pyrocarbon; Pyrolysis; Regression models; Support vector machines; Temperature dependence; Thickness; Transport properties
• ISSN: 0263-8762; 1744-3563
• DOI: 10.1016/j.cherd.2021.12.015

•Development of comprehensive 3D CFD model for commercial hot wall reactor.•Understanding of the critical parameters affecting the film performance.•Correlation of growth patterns and operating conditions using support vector machine.•Optimal design using machine learning combined with Nelder-Mead algorithms. Chemical vapour deposition (CVD) of pyrocarbon (PyC) effectively fabricates advanced carbon materials. Controlling the nanotexture of PyC is critical for the desired application. Reactor operating conditions, including temperature, pressure, inlet flow rate, reactant concentration, govern thickness, and film uniformity. The optimal film performance can be obtained by selecting appropriate process conditions. However, the optimisation of the CVD reactor is challenging due to the highly nonlinear and multi-variable nature of the process. In this study, the support vector machine, a robust supervised learning algorithm and Nelder-Mead algorithm are coupled to optimise the CH4 pyrolysis in a commercial CVD reactor. To this end, a comprehensive CFD model for CH4 pyrolysis in a commercial CVD reactor is first constructed. The model accuracy is improved by considering temperature-dependent transport properties of gas mixture and incorporating the detailed gas and surface chemistry (14 species and 32 reactions). The deposition rate and film uniformity are then obtained using a parametric study. Subsequently, the support vector machine (SVM) is employed to deduce the correlation between the PyC growth rate/film uniformity and operating variables. It has been found that the accuracy of SVM is better than the linear regression model. Finally, SVM coupled with the Nelder-Mead algorithm is proposed to optimise the CVD process parameters to maximise the PyC film quality.