IMPLEMENTATION OF ATOMIC POTENCTIAL BASED ON MACHINE LEARNING TO OBSERVE STRUCTURE DEFECT ON TIO2

• Main Author: Rausyanfikr, Fadhil
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78243
• Institution: Institut Teknologi Bandung

With the development of the times, computational simulations are widely used to find possible combinations of molecules with efficient positions until a material with optimal properties is found. In this way, it shortens the time for finding new materials compared to looking for the right combination only through experiments in the laboratory. The most popular method in material processing is the Density Functional Theory (DFT) calculation. But DFT has huge processing costs so it is necessary to find a different solution. One of them is the potential of machine learning. In this study, using an Artificial Neural Network (ANN) based on the Behler-Parinello approach to observe structural defects in TiO2. Titanium Dioxide (TiO2) is used because of its versatility. TiO2 has applications in various fields including electronics, energy, environment, health & medicine, sensors, and catalysts. This research was started by looking for the potential of ANN which has the highest accuracy for the TiO2 structure. It was found that the potential NN which was constructed using the bfgs method and the hyperbolic tan activation function with 2 hidden layers and 20 nodes in each layer had the smallest error between the training data and the test data. The potential ANN is used for the structure optimisation and NVT simulation of the TiO2 structure that has been given defects. We get very good results because there are no overlapping TiO2 bonds in the final structure.