Thermal Conductivity Modeling for Liquid-Phase-Sintered Silicon Carbide Ceramics Using Machine Learning Computational Methods

Silicon carbide is a covalently bonded engineering material and structural ceramic with excellent mechanical properties, high resistance to oxidation, corrosion, and wear, and tunable thermal conductivity. The exceptional thermal conductivity of silicon carbide ceramic promotes its candidature in many industrial applications, such as nuclear fuel capsule materials, substrate materials employed in semiconductor devices, heater plates, and heaters for processing semiconductor and gas seal rings employed in compressor pumps, among others. The synthesis of polycrystalline silicon carbide through the liquid-phase sintering approach results in lower thermal conductivity due to the presence of structural defects associated with grains, lattice impurities, grains' random orientations, and the presence of secondary phases in polycrystalline silicon carbide ceramic. The conventional experimental method of enhancing thermal conductivity is laborious and expensive. This present work modeled the thermal conductivity of liquid-phase silicon carbide ceramic via intelligent approaches involving genetic algorithm-optimized support vector regression (SVR-GA), an extreme learning machine with a sine activation function (ELMS), and random forest regression (RFR). The descriptors for the models included the nature of sintering additives as well as their weights, sintering conditions, applied pressure, sintering temperature, and time. Using the mean absolute error (MAE) and root mean square error (RMSE) for performance assessment, it was observed that the ELMS outperformed the RFR and SVR-GA models with improvements of 40.50% and 25.76%, respectively, using the MAE metric and improvements of 16.57% and 24.43%, respectively, using the RMSE metric. The developed models were further used to investigate the effect of the weight of sintering additives and sintering time on the thermal conductivity of silicon carbide ceramic. The precision of the developed models facilitated a comprehensive investigation of the effect of sintering factors on thermal conductivity while hidden connections that exist between the factors are uncovered for enhancing application domains for silicon carbide ceramics.