Effective thermal conductivity evaluation of SPS-nanoscale W powder sintered at different temperature by microstructural-simulation-based Machine Learning

Tungsten (W) stands out as one of the most promising matrix materials for cermet fuels in nuclear thermal propulsion (NTP) systems, with its effective thermal conductivity (ETC) being intricately linked to the economic viability and safety of reactors. This study investigated the influence of sintering temperature on the microstructure of nanoscale W powder using Spark Plasma Sintering (SPS). Based on the microstructural evolution observed at various temperatures, both a finite element (FE) model and an Artificial Neural Network (ANN) model were employed to assess the ETC of the W matrix, and the ETC was experimentally measured using the frequency-domain thermoreflectance (FDTR) method. The FE simulation results reveal a 7.5 % average relative error between modeled and measured ETC values at high sintering temperatures, demonstrating good consistency between simulation and experiment. Further analysis reveals that the porosity of the W matrix significantly affects its ETC. As the sintering temperature increases from 1273 to 1973 K, the porosity decreases from 32.48 % to 3.86 %, while the ETC increases from 46.07 to 160.71 W/(m·K). The ANN results demonstrate that the trained ANN model could accurately assess the ETC of the W matrix, with a regression coefficient of 0.9983 for the trained dataset and 0.9922 for the tested dataset. Based on the relationship between porosity and ETC, an optimized formula has been developed for predicting the ETC of the W matrix with an R² value of 0.9954 and an RRMSE of 2.37 %. The ANN model established in this study provides an accurate method for predicting the ETC of W and offers new insights and references for predicting the ETC of other materials.