الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 96 |  |  |
| | | from | 96 | | |
Abstract
Accurate estimation of thermophysical properties of nuclear materials is crucial for simulation codes. We developed an ML-based for predicting thermal conductivity of Uranium Dioxide nuclear fuel. The model predicts thermal conductivity as a function of: temperature, irradiation temperature, annealing temperature, burnup, density, and deviation from stoichiometry. We used a data set of 927 measurement points collected from eight different studies for model training and testing. A model selection algorithm was developed to evaluate six different ML algorithms based on their performance on a testing data set. The candidate models were selected to cover the commonly used ML regression models and included: K-nearest neighbors, random forest, support vector machine, artificial neural networks, group method data handling, and exterme gradient boosting regression. The exterme gradient boosting model outperformed the other five models with: R2 score 0.964, MSE of 0.035 and training time of 0.275 seconds. The six models were verified against the recommended correlations in literature for estimating UO2 thermal conductivity. Global sensitivity analysis was conducted using Fourier amplitude sensitivity test (FAST) to rank the model input parameters according to their effect on the output. The analysis showed that the main parameters in determining UO2 thermal conductivity are: temperature, burnup, and annealing temperature respectively. Finally, we conducted three parametric studies to investigate the changes in the predicted thermal conductivity with changing various model parameters.