Abstract
This study develops and validates a machine learning (ML)–based framework for correlation of Charpy impact properties between sub-sized and full-sized specimens. Reliable correlation of Charpy impact test results across specimen sizes is essential for structural integrity assessments, especially in nuclear applications where limited material volume restricts specimen dimensions. Existing analytical correlation methods exhibit limited generality and accuracy across different materials, heat treatments, and geometries.A comprehensive dataset of Charpy V-notch impact test results was compiled to support the study of size effects in reactor pressure vessel steels and nuclear structural materials. The dataset consists of 4,961 individual test records collected from 109 peer-reviewed publications and includes sub-sized and full-sized specimens across alloys such as SA508, SA533B, 20MnMoNi55, A302B, and SS316. Each record contains material composition, manufacturing and heat treatment history, irradiation conditions, specimen dimensions, and impact energy–temperature data, enabling extraction of upper shelf energy (USE) and ductile-to-brittle transition temperature (DBTT) through curve fitting.
The proposed ML-based framework maps absorbed energy values across the full ductile-to-brittle transition region by applying a temperature shift combined with scaled residual projection to align sub-sized test data with full-sized response. From the resulting temperature–energy profiles, the correlated values for USE and DBTT are extracted via fitting a hyperbolic tangent model. The framework is validated using the collected dataset comprising matched sub-sized and full-sized Charpy impact tests. The ML-based approach demonstrates improved correlation performance relative to conventional analytical methods, achieving R² values of 0.942 for USE and 0.892 for DBTT. The combination of a large, curated dataset and a generalizable data-driven correlation framework support more reliable toughness evaluation of nuclear structural materials across specimen sizes.