STACKING FAULT ENERGY DESIGN MODELING FOR HIGH ENTROPY ALLOY FENICRCOCU USING THERMODYNAMIC CALCULATION AND MACHINE LEARNING

• Main Author: Hans Sebastian T, Leonardus
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85625
• Institution: Institut Teknologi Bandung

Technological development for the sake of fulfilling human necessity gave birth to a new concept that attempts to combine 5 or more elements with high entropy called as high entropy alloy (HEA). Research on HEA becomes a very interesting topic due to HEA’s mechanical properties that are stronger and have better ductility compared to conventional alloy. FeNiCrCoCu is one of the more interesting alloys that is being utilized due to its high wear and irradiation resistance. Understanding the contribution of stacking fault energy (SFE) becomes an integral part due to its contribution on the alloy deformation. Conventional research on HEA requires large expenses and long experiment due to the process of trial and error. Developing technology such as thermodynamic calculations simulation and machine learning becomes an important thing to increase the efficiency of HEA’s research. Hence, this research aims to determine the effect of temperature and composition toward SFE and magnetic contribution, determine the relation between SFE and mechanical properties, determine the best machine learning model, dan provide guidelines for the design of FeNiCrCoCu HEA. The simulation was carried out by determining the parameters used in thermodynamic calculations from various literature. The said parameters and thermodynamic calculations algorithm then are inputted to the MATLAB program, producing the database. The said database is then separated with the ratio of 80:20 as training data and testing data. The training data then being used to train machine learning model consisted of decision tree, random forest, and neural network. The said model then asked to predict SFE from testing data to evaluate the model. The performance and accuracy of the model then being determined by root mean squared error (RMSE) and accuracy calculation to determine the best model. Based on the simulation, the increase of temperature will be followed by the increase of SFE while decreasing the magnetic contribution, the increase of Fe and Ni content will increase the SFE, and the increase of Cr will decrease the SFE. Below the temperature of 250 K, the increase of Cu will increase SFE while above 250 K, the increase of Cu will decrease the SFE. Below the temperature of 1050 K, the increase of Co will increase SFE while above 1050 K, the increase of Co will decrease the SFE. While all machine learning models are already giving good performance with more than 96% accuracy, the best machine learning model is random forest with an RMSE of 1.31 and accuracy of 98.97%. The guideline for FeNiCrCoCu HEA on the temperature of 800 K follows the range of Fe 5 – 16%; Ni 5 – 16%; Cr 22 – 40%; Co 5 – 9%; Cu 26 – 40%.