Data-driven modelling for predicting DBTT in low-alloy steels for naval application

Abstract

Ductile-to-brittle transition temperature (DBTT) is an important material property for ascertaining the structure integrity and reliability of naval steel, particularly under low-temperature and high-strain-rate conditions experienced in maritime conditions. Lower DBTT provides greater fracture resistance and minimizes the probability of disastrous brittle failure, rendering it an important naval performance parameter. Conventional strategies for optimizing DBTT are based significantly on experimental protocols and empirical models, which are frequently time-consuming, expensive, and not very scalable. To meet these demands, the current work investigates a data-driven approach employing machine learning (ML) methods for prediction and optimization of the DBTT curve in naval steels, thus streamlining material design and performance analysis. The work starts from the acquisition and preparation of a detailed dataset including compositional data, microstructural and thermodynamic descriptors (e.g., grain size, phase fractions, and dislocation energy), and mechanical properties obtained both experimentally and by thermodynamic calculations. The central aim of the project is to reduce the DBTT while increasing the Upper Self Energy—a measure of a material's resistance to deformation and defect creation. These two targets were chosen because they were most relevant to impact toughness as well as structural behavior under dynamic loading conditions.