DeepFusion-SMAs: Predicting Phase Change Temperature in Shape Memory Alloys

📌 Overview

This repository provides a hybrid deep learning model to predict the martensitic transformation start temperature (Ms) in multicomponent shape memory alloys (SMAs). The DeepFusion model combines neural networks to analyze alloy composition, processing conditions, and material properties, achieving high accuracy. It includes a comprehensive dataset of 32 alloy characteristics to support predictions despite limited data.

🗂 Contents

• 📂 Dataset/ – Experimental and literature data on SMA transformation temperatures.
• 📓 COMMAT-D-25-00345.ipynb – Jupyter Notebook with the DeepFusion model implementation.
• 📜 requirements.txt – List of Python dependencies for running the project.
• 📜 README.md – This file explaining the project structure and goals.

🏗 Model Architecture

The DeepFusion model includes:

• Dense Branch: Processes alloy properties and processing conditions with neural layers.
• Self-Attention Branch: Captures interactions among alloy elements (e.g., nickel, titanium).
• Regression Head: Combines outputs to predict Ms temperature accurately.

📊 Dataset Details

The dataset includes 32 characteristics:

• Alloy composition: Nickel, titanium, hafnium, zirconium, copper, palladium, cobalt, niobium.
• Processing conditions: Homogenization temperature/time, aging temperature/time, applied stress.
• Material properties: Atomic mass, bulk modulus, melting temperature, thermal conductivity.
• Thermodynamic properties: Mixing entropy, mixing enthalpy, valence electron concentration.

🛠 Requirements

Install dependencies using the provided requirements.txt:

pip install -r requirements.txt

🚀 How to Use

1. Clone the repository:

   git clone https://github.com/your-username/DeepFusion-SMAs.git
   cd DeepFusion-SMAs

2. Install dependencies (see Requirements section).
3. Run the notebook:

   jupyter notebook COMMAT-D-25-00345.ipynb

📈 Results

• Achieved a test Mean Absolute Error of 17.22°C and R² of 0.94 (full dataset).
• Analysis shows nickel group (nickel, palladium, copper, cobalt) and aging temperature as key predictors.
• Model excels in complex alloys but needs more high-temperature data for further gains.

🎯 Future Work

• Expand dataset with more alloy samples to boost accuracy.
• Explore graph-based neural networks for alloy interactions.
• Develop a web tool for interactive Ms predictions.

📝 Citation

Huynh, T.N., Raji, H., Saedi, S. and Nguyen, K.D., 2025. Deep learning framework for accurate prediction of phase transformation in multicomponent shape memory alloys. Computational Materials Science, 258, p.114037.

🤝 Contributions & Contact

We welcome contributions! Open issues or pull requests for improvements. For questions, contact nhuynh2023@my.fit.edu.