Predicting Yield Strength in Multi-Principal Element Alloys

Using data science and machine learning to explore and predict yield strength in multi-principal element alloys based on composition, microstructure, and processing methods

---

Overview

This project explores a curated dataset of Multi-Principal Element Alloys (MPEAs) to:

1. Identify high-strength alloys through exploratory data analysis (EDA)
2. Predict yield strength based on composition, microstructure, and processing methods
3. Interpret key drivers of mechanical performance using SHAP analysis

This analysis focuses on compressive yield strength at room temperature (25 °C) from a curated subset of the full dataset.

Key Results:

• An XGBoost model achieved R² = 0.82, outperforming linear (Ridge) and non-parametric (KNN) models
• SHAP analysis highlighted Mo, Cr, and BCC+secondary phases as critical contributors to strength

---

Repository Structure

|
├── analysis/                # Jupyter notebook + visuals
├── data/
│   ├── raw/                 # Original dataset (MPEA_dataset.csv, compositions.csv)
│   └── processed/           # Cleaned data and top-performing alloys
├── LICENSE.txt              # MIT License terms
├── README.md                # This file
└── environment.yml          # Conda environment configuration

---

Dataset Description

The dataset originates from the study "Expanded dataset of mechanical properties and observed phases of multi-principal element alloys" (Borg et al., 2020) and includes information such as yield strength measurements, chemical composition, microstructure, processing methods, density, grain size, and more.

It is publicly available via Citrine Informatics' GitHub repository and stands as a landmark contribution to open materials science research.

---

Tech Stack

• Python: Pandas, NumPy, Scikit-learn, XGBoost, SHAP
• Visualization: Matplotlib, Seaborn, Missingno
• Workflow: Jupyter Notebook, Git

---

Key Insights

Top-Performing Alloys

• Highest Yield Strength: Cr1 Mo1 Nb1 Ta1 V1 W1 (BCC+Laves+Sec., powder-processed) achieved 3416 MPa.
• Best Strength-to-Weight: Al1 Co1 Cr1 Fe1 Ni1 Si1 (BCC+Sec., cast) reached 423 MPa·cm³/g.
  (See full lists in data/processed/.)

Model Performance

Model	R²	RMSE (MPa)
Ridge	0.485	454.4
KNN	0.690	352.3
XGBoost	0.821	268.3
R² = 0.821 indicates XGBoost explains 82.1% of variance in yield strength.

SHAP Analysis

Key drivers of yield strength:

• ↑ Strength: High Mo/Cr, BCC+Sec. microstructure, powder processing.
• ↓ Strength: FCC phases, casting.

---

How to Reproduce

1. Clone the repository:

   git clone https://github.com/MathRC/mpea.git
   cd mpea

2. Set up the Conda environment:

   conda env create -f environment.yml
   conda activate mpea

3. Run the analysis:

   • Open analysis.ipynb in Jupyter Notebook/Lab
   • All data paths in the notebook are relative to the repo root

---

Acknowledgments

• Citrine Informatics for hosting this transformative dataset and enabling open science.
• Borg et al. (2020) for curating and validating a dataset of over 1,000 MPEA entries through domain expertise and statistical reviews.
• Materials Science Community for pioneering research that bridges metallurgy and data-driven design.

---

References

[1] Borg, C. K. H., Frey, C., Moh, J., Pollock, T. M., Gorsse, S., et al. (2020).
Expanded dataset of mechanical properties and observed phases of multi-principal element alloys. Scientific Data, 7, 430.
https://doi.org/10.1038/s41597-020-00768-9

[2] Borg, C., & Saal, J. (2020).
Expanded dataset of mechanical properties and observed phases of multi-principal element alloys. GitHub repository.
Available at: https://github.com/CitrineInformatics/MPEA_dataset

---

License

This project is licensed under the MIT License.

---

Connect: LinkedIn • GitHub
📧 Email: matt21rc@gmail.com

Open to collaboration and feedback!

---