Prediction of elastic and optical properties of binary glass system using Artificial Intelligence approach

• Main Author: Prasad, Soundrarajan
• Format: Final Year Project / Dissertation / Thesis
• Published: 2024
• Subjects: QA75 Electronic computers. Computer science; QA76 Computer software; QC Physics; QD Chemistry; TP Chemical technology
• Online Access: http://eprints.utar.edu.my/6688/1/Prasad_AL_Soundrarajan_1805217.pdf; http://eprints.utar.edu.my/6688/
• Institution: Universiti Tunku Abdul Rahman

Binary glass systems are a rising prospect in the industrial field and have piqued interest in research and the electronic field. The unique properties and distinctive structure of the binary glass systems provide a wide range of applications in the electronic field such as optical fibers, optical switching devices, laser hosts, and more. Additionally, during the manufacturing process of the binary glass, certain simulations is necessary to predict the characteristics of the glass before the pure materials of oxide are melted. Previous research has suggested and implemented the usage of artificial neural network models as instruments to simulate and predict the optical and elastic properties of binary glass series ZnO-TeO2 glasses. Based on previous results, MATLAB software was used to predict the properties of the glasses, and sufficient results were produced for different types of ZnO-TeO2 glass compositions. However, there was a drawback using MATLAB where the perfect fit correlation value, R2 which represents the proportion of variance in the dependent variable that is predictable from the independent variable in a regression model is satisfactory as the correlation value R was all between 0.90361 and 0.99985. Nevertheless, the research established that the use of the ANN model is a good approach to be used in future research. In this project, python software with deep learning libraries such as PyTorch and scikit-learn was utilized to predict the elastic and optical properties of several glass series with different compositions. Thus, the results produced had a better and consistent R2 value within a range of 0.97 to 0.99 which indicates a high degree of predictability in the relationship between the independent and dependent variables in the model. Furthermore, the training total loss on binary glass characteristics data set, graphs of predicted values, and real values were visualized, discussed, and studied in this report.