From Small Data Modeling to Large Language Model Screening: A Dual-Strategy Framework for Materials Intelligent Design

Abstract

Small data in materials present significant challenges to constructing highly accurate machine learning models, severely hindering the widespread implementation of data-driven materials intelligent design. In this study, the Dual-Strategy Materials Intelligent Design Framework (DSMID) is introduced, which integrates two innovative methods. The Adversarial domain Adaptive Embedding Generative network (AAEG) transfers data between related property datasets, even with only 90 data points, enhancing material composition characterization and improving property prediction. Additionally, to address the challenge of screening and evaluating numerous alloy designs, the Automated Material Screening and Evaluation Pipeline (AMSEP) is implemented. This pipeline utilizes large language models with extensive domain knowledge to efficiently identify promising experimental candidates through self-retrieval and self-summarization. Experimental findings demonstrate that this approach effectively identifies and prepares new eutectic High Entropy Alloy (EHEA), notably Al₁₄(CoCrFe)₁₉Ni₂₈, achieving an ultimate tensile strength of 1085 MPa and 24% elongation without heat treatment or extra processing. This demonstrates significantly greater plasticity and equivalent strength compared to the typical as-cast eutectic HEA AlCoCrFeNi_2.1. The DSMID framework, combining AAEG and AMSEP, addresses the challenges of small data modeling and extensive candidate screening, contributing to cost reduction and enhanced efficiency of material design. This framework offers a promising avenue for intelligent material design, particularly in scenarios constrained by limited data availability.

Keywords: adversarial domain adaptation; experimental candidates screening; material intelligent design; small data modeling.

Figure 1

Overview of the Dual‐Strategy Materials Intelligent Design (DSMID) Framework. Adversarial domain Adaptive Embedding Generative network (AAEG) enhances material property prediction by aligning feature representations across domains using domain‐adaptive techniques. Automated Material Screening and Evaluation Pipeline (AMSEP) utilizes NSGA‐II and large language models to generate and identify materials with superior mechanical properties. Synthesis and Validate confirms the model's predictive accuracy and supports the reverse design of materials.

Figure 2

Adversarial domain adaptive embedding generative network structure. F: Feature representation network, whose output of the last hidden layer is used as the generated embedding of materials composition. C: Source domain bucketing network, used to provide gradients for F to perform source domain bucketing task. G: Generation network, used to provide gradients for F to confuse the source and target domain. D: Discriminator network is designed to perform two critical tasks: discriminate between the source and target domains for classification and source domain target properties bucketing. The network is adversarially trained with G and F and is directly transferred to F utilizing residual connections to minimize domain shift between the source and target domains.

Figure 3

Automated Screening and Evaluation of Inverse Design Candidates via Joint Data and Rule Stream Approach.

Figure 4

(Top row) Enhanced prediction accuracy in HEAs through AAEG: a tenfold validation analysis. (Bottom row) Comparative analysis of predictive and actual mechanical properties in OOD HEAs with the AAEG.

Figure 5

Overview of the Inverse Design Results Evaluated Using AMSEP. a) Comparison of experimental candidates with existing data points in the dataset. b) Evaluation of experimental schemes without additional processing. The mean UTS of this design outcome is 1071.36 MPa, while the mean EL is 16.36%.

Figure 6

The microstructure and true stress versus true strain curves of as‐cast design alloy: a) Electron backscatter diffraction (EBSD) phase image, b) X‐ray diffractometer (XRD) pattern. c) True stress versus true strain curves of the design HEA with a strain rate of 10⁻³ s ⁻¹, the inset is the geometrically necessary dislocations (GND) density of design HEA before and after tension. Detailed microstructural changes in Al₁₄(CoCrFe)₁₉Ni₂₈ EHEA under tension can be found in Section S7 (Supporting Information).