Machine Learning-Based Composition Design of Functionally Graded Alloys.

Area of Science:

• Materials Science and Engineering
• Computational Materials Science
• Mechanical Engineering

Background:

• Functionally graded materials (FGMs) mitigate residual stress at dissimilar material interfaces via graded transitions.
• Matching the coefficient of thermal expansion (CTE) is crucial for joint reliability.
• Traditional empirical methods are inefficient for optimizing composition in high-dimensional spaces.

Purpose of the Study:

• To develop a data-driven, machine learning-assisted method for designing FGM compositions.
• To efficiently identify optimal compositions for matching the CTE of dissimilar materials.
• To reduce interfacial thermal expansion mismatch and enhance the reliability of material joints.

Main Methods:

• Systematic evaluation of six regression models using a dataset of 747 CTE data points across 15 elements.
• Utilized a high-precision dataset for training and testing machine learning models.
• Employed the SHapley Additive exPlanations (SHAP) method to identify key physical descriptors influencing CTE.

Main Results:

• The random forest (RF) model demonstrated superior performance with R² of 0.929 and RMSE of 0.658.
• Lattice constant, Young's modulus, and temperature were identified as critical factors for thermal expansion behavior.
• Experimental validation showed a low CTE prediction deviation (0.89%) for Fe-based alloy Norem02.

Conclusions:

• The proposed machine learning framework enables efficient and accurate design of FGM compositions.
• Successful design of a 316L/Norem02 transition layer significantly reduced thermal expansion mismatch.
• Provides a reliable theoretical basis for designing dissimilar material interfaces under extreme conditions.