Project Title: Design of Alloys for gaining high strength to weight ratio

Overview:

This repository contains machine learning models developed for the "Design of Alloys with High Strength-to-Weight Ratio" project . The models aim to predict key properties of Aluminum (Al) alloys, specifically minimum density and maximum stress, to facilitate the design of new alloys with improved strength-to-weight ratios.

Features:

• Property Prediction: Machine learning models (Support Vector Regression - SVR) for predicting maximum stress and minimum density of Al alloys.

• Composition-Based Featurization: Utilizes the CBFV (Composition-Based Feature Vectors) library to generate relevant features from alloy chemical formulas.

• Data Preprocessing: Includes steps for data splitting (90% train, 10% test), feature scaling using StandardScaler, and handling of elemental properties.

• Robust Model Training: Employs GridSearchCV with GroupKFold cross-validation for hyperparameter tuning and robust model selection. GroupKFold ensures that samples with the same alloy composition are kept together during cross-validation, preventing data leakage.

• Uncertainty Quantification: Implements a bootstrapping method to estimate the mean and standard deviation of predictions, providing a measure of uncertainty for each prediction.

• Expected Improvement Calculation: Includes functionality to calculate Expected Improvement (EI), an acquisition function used in Bayesian optimization to identify promising new alloy compositions for further investigation.

Repository Structure

• Density_prediction_model.ipynb: Jupyter Notebook containing the code for predicting minimum density of Al alloys.
• maxStress_prediction_model.ipynb: Jupyter Notebook containing the code for predicting maximum stress of Al alloys.
• README.md: This file.

Getting Started

Prerequisites

• Python 3.x
• Jupyter Notebook

Installation

1. Clone the repository:

   git clone [https://github.com/kamaladhi/High-Strength-Lightweight-Alloys.git](https://github.com/kamaladhi/High-Strength-Lightweight-Alloys.git)
   cd High-Strength-Lightweight-Alloys

2. Install the required libraries: The notebooks use CBFV for featurization, along with standard libraries like pandas, numpy, scikit-learn, and matplotlib.

   pip install pandas numpy scikit-learn matplotlib seaborn tqdm scipy
   pip install CBFV

Usage

1. Open the Jupyter Notebooks:

   jupyter notebook

2. Navigate to and open Density_prediction_model.ipynb or maxStress_prediction_model.ipynb.

3. Run all cells in the notebooks sequentially.

   • The notebooks will:
     • Load the respective _target.xlsx files for training and testing.
     • Generate features from the alloy formulas.
     • Train and evaluate the SVR models.
     • Perform bootstrapping for uncertainty quantification.
     • Load Al-Virtual-Samples.xlsx to predict properties for the candidate alloys.
     • Calculate Expected Improvement (EI) for new design candidates.
     • Save the prediction results (including EI) to a CSV file (e.g., Denisty_prediction.csv or max_stress_pred.csv).

Models and Metrics

Maximum Stress Prediction Model

• Model: Support Vector Regression (SVR)
• Training R-squared: ~0.729
• Test R-squared: ~0.668
• Mean Absolute Error (MAE) (Test): ~488.77 MPa
• Root Mean Squared Error (RMSE) (Test): ~597.99 MPa

Minimum Density Prediction Model

• Model: Support Vector Regression (SVR)
• Training R-squared: ~0.947
• Test R-squared: ~0.910
• Mean Absolute Error (MAE) (Test): ~0.323 g/cm³
• Root Mean Squared Error (RMSE) (Test): ~0.519 g/cm³

(Note: For minimum density, the Expected Improvement calculation in the notebook currently identifies "improvement" as values higher than the current maximum observed density. For true minimization, this acquisition function would ideally be adapted to find compositions that minimize the density, or a Lower Confidence Bound (LCB) acquisition function could be used.)