Towards robust prediction of material properties for nuclear reactor design under scarce data -- a study in creep rupture property

TL;DR

The paper tackles the challenge of predicting creep-rupture life for nuclear reactor materials under scarce data by introducing a probabilistic meta-learning framework based on a conditional neural process (CNP). This approach encodes prior knowledge from related cast codes into meta-parameters $\theta^{*}$ to enable fast, uncertainty-aware adaptation to new, data-poor tasks, producing a distribution over predictions $p(y|x,C)$. Empirical results on a large repository of creep data show superior generalization over Larson–Miller baselines, achieving metrics such as $R^{2}=0.89$ and $P_{95}=0.88$, and providing credible extrapolations in sparse data regimes. The method offers transferable, trustable AI analytics for nuclear materials design, with implications for safer and more robust reactor components.

Abstract

Advances in Deep Learning bring further investigation into credibility and robustness, especially for safety-critical engineering applications such as the nuclear industry. The key challenges include the availability of data set (often scarce and sparse) and insufficient consideration of the uncertainty in the data, model, and prediction. This paper therefore presents a meta-learning based approach that is both uncertainty- and prior knowledge-informed, aiming at trustful predictions of material properties for the nuclear reactor design. It is suited for robust learning under limited data. Uncertainty has been accounted for where a distribution of predictor functions are produced for extrapolation. Results suggest it achieves superior performance than existing empirical methods in rupture life prediction, a case which is typically under a small data regime. While demonstrated herein with rupture properties, this learning approach is transferable to solve similar problems of data scarcity across the nuclear industry. It is of great importance to boosting the AI analytics in the nuclear industry by proving the applicability and robustness while providing tools that can be trusted.

Figures (4)

• Figure 1: Comparison of the trained model along with two baselines on an arbitrary cast code in the validation set. The green line suggests the LM model with $d=1$ while the yellow curve suggests the LM model with $d=2$. The blue curve denotes the conditional mean produced by the trained model and the blue shades denote the range of 2 standard deviation
• Figure 2: Performance comparison of varying context points. top row represents the predictions from the a pretrained neural network model; middle row predictions from the conditional neural process model; bottom row predictions from a Gaussian Process model. From the first to the last column, more context points are available to the models
• Figure 3: The prediction performance of the conditional neural process model over the whole testing set. The horizontal bar represents 2 times of standard deviations
• Figure :