Stress-strain curve predictions by crystal plasticity simulations and machine learning

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Materials Science

Background:

• Predicting stress-strain curves (SSC) for additively manufactured metals via laser powder bed fusion (LPBF) is computationally intensive.
• Traditional methods involve complex microstructure reconstruction and crystal plasticity simulations, which are time-consuming.
• Machine learning (ML) offers a potential solution to accelerate SSC prediction.

Purpose of the Study:

• To develop and evaluate various ML methods for predicting the temperature-dependent SSCs of LPBF-fabricated Hastelloy X.
• To establish a direct link between experimental conditions and predicted SSCs, bypassing microstructure reconstruction.
• To identify the most effective ML approach for this specific application.

Main Methods:

• Utilized several machine learning (ML) methods to predict SSCs.
• Focused on linking experimental conditions directly to SSCs, excluding microstructure data.
• Trained and compared different ML models, including decision tree-based regressors and artificial neural networks (ANNs).

Main Results:

• Decision tree-based ML regressors demonstrated superior performance compared to other popular ML methods.
• The developed models successfully predicted the temperature dependence of SSCs for Hastelloy X.
• The approach of directly linking experimental conditions to SSCs proved effective, especially with smaller datasets.

Conclusions:

• Decision tree-based ML models are highly effective for predicting temperature-dependent SSCs of LPBF Hastelloy X, particularly when using experimental conditions as input.
• This ML approach offers a computationally efficient alternative to traditional simulation methods.
• The study highlights the potential of ML in accelerating materials characterization and design for additively manufactured alloys.