A comprehensive Ai-driven framework for predictive design of grained materials: Case study on nanoglass

A Edalatmanesh and M Mahnama, ACTA MATERIALIA, 296, 121292 (2025).

DOI: 10.1016/j.actamat.2025.121292

Designing materials with specific mechanical properties requires a comprehensive understanding of the processstructure-property relationships. Traditional methods, such as experimental approaches and computational simulations, tend to be resource-intensive, time- consuming, and often lack scalability. To address these challenges, this study develops a novel artificial intelligence based framework aimed at the predictive design of grained materials, with a particular focus on nanoglasses (NGs), which offer unique tunable microstructures compared to conventional metallic glasses (MGs). The framework incorporated a novel microstructure quantification technique, dimensionality reduction, and machine learning models to enable both mechanical property prediction and inverse design. A key component is the angular 3D chord length distribution (A3DCLD), a microstructure characterization method that effectively captures the spatial features of NGs, overcoming the limitations associated with 2D approaches. Moreover, the integration of conditional variational autoencoders allows for inverse design by generating optimal process-structure combinations for desired mechanical properties, thus providing a data-driven approach to material design. Evaluations show that the framework not only predicts mechanical properties with high accuracy but also optimizes processing parameters and resulting microstructures to achieve specific mechanical behaviors. While the framework has been demonstrated using NGs, its adaptability allows it to be applied to various types of grained materials, significantly broadening its potential impact beyond this particular material system.

Return to Publications page