Machine learning based prediction of activation energy of pure lithium titanate pebbles fabricated by sol-gel technique

Lithium titanate (Li₂TiO₃) is considered as one of the most promising ceramic breeder materials for fusion reactor blankets because of its favorable tritium release characteristics and low activation energy requirements. In this study, pure Li₂TiO₃ pebbles were fabricated using Sol-Gel technique followed by urea-acetone-assisted spherodization. The thermal and structural characterization was performed using DTA/TG and XRD analyses. Activation energy (Ea) characterizes the temperature dependence of tritium diffusion and release in Li₂TiO₃ breeder materials and is a key kinetic parameter governing tritium extraction efficiency in fusion reactor blankets. Accurate \(\:{E}_{a}\) determination is essential for designing tritium extraction systems. Therefore, a machine-learning model based on a feed-forward Neural Network ( NN ) was developed to estimate activation energy of Lithium titanate (Li₂TiO₃) for tritium release. Unlike conventional Arrhenius fitting approaches that are sensitive to experimental variability, the proposed framework incorporates multiple process variables simultaneously and captures nonlinear relationships. The dataset consisted of 10,000 synthetically generated samples derived from consistency verification of Arrhenius relationships and literature-reported parameter ranges for Li₂TiO₃ breeder materials. The neural network demonstrated high predictive capability (R²=0.94) and estimated activation energy of 1.12 eV, closely matching the Arrhenius-derived value of 1.15 eV. The results demonstrate the potential of machine learning as an efficient tool for kinetic analysis and tritium extraction optimization in fusion energy systems. The present work should be regarded as a methodological demonstration of machine-learning-assisted kinetic analysis using a physically motivated synthetic dataset derived from consistency verified Arrhenius relationships and literature-reported parameter ranges. Independent consistency verification using newly acquired tritium release datasets remains a subject of future investigation.