Related Experiment Video
Updated: Oct 5, 2025
14:16
Fabrication of Schottky Diodes on Zn-polar BeMgZnO/ZnO Heterostructure Grown by Plasma-assisted Molecular Beam Epitaxy
Published on: October 23, 2018
7.8K
Density of Bismuth Boro Zinc Glasses Using Machine Learning Techniques
Shaik Amer Ahmed1, Shaik Rajiya2, M A Samee2
1Department of Physics, Nizam College, Osmania University, Hyderabad, India.
Journal of Inorganic and Organometallic Polymers and Materials
|January 24, 2022
Summary
Machine learning accurately predicts glass density in bismuth-doped borate glasses. Random Forest regression achieved the highest accuracy, confirming the strong correlation between predicted and experimental density values.
Area of Science:
- Materials Science
- Glass Chemistry
- Computational Materials Science
Background:
- The density of glasses is a critical property influenced by their chemical composition and structural arrangement.
- Bismuth oxide (Bi2O3) addition to borate glasses significantly impacts their physical and structural characteristics, including density.
- Understanding these structure-property relationships is essential for designing glasses with tailored properties.
Purpose of the Study:
- To predict the densities of xBi2O3-(70-x)B2O3-20Li2O-5Sb2O3-5ZnO glasses using machine learning.
- To investigate the influence of Bi2O3 content on the density and structural evolution of these glasses.
- To compare the performance of different artificial intelligence (AI) models in predicting glass density.
Main Methods:
- A dataset of 2000 B2O3-rich glasses was utilized, incorporating chemical composition and ionic radius as input features.
- Experimental density measurements were performed on prepared bismuth-doped borate glasses.
- Machine learning models, including Gradient Descent, Random Forest (RF) regression, and Artificial Neural Networks (ANNs), were trained and evaluated.
Main Results:
- Glass density was found to increase with increasing Bi2O3 content, attributed to the conversion of BO3 to denser BO4 structural units.
- Fourier-transform infrared (FTIR) and Raman spectroscopy confirmed structural changes, showing a decrease in B-O-B bonds and a shift in BO4 vibrations with increasing Bi2O3.
- Random Forest regression achieved the highest prediction accuracy with an R² value of 0.983, demonstrating excellent correlation between predicted and experimental densities.
- Artificial Neural Networks also showed effective prediction performance (R² = 0.950) with a Tanh activation function.
- Gradient Descent yielded the minimum cost of 0.018, indicating robust performance.
Conclusions:
- Machine learning, particularly Random Forest regression, provides a highly accurate method for predicting the density of bismuth-doped borate glasses.
- The study confirms the direct relationship between bismuth content, structural transformations (BO3 to BO4), and increased glass density.
- The findings highlight the potential of AI in accelerating materials discovery and optimizing glass composition for desired properties.
