DESIGN OF HIGH ENTROPY SUPERALLOY FENICRALCU WITHSTACKING FAULT ENERGY MODELLING USING THERMODYNAMICCALCULATION AND MACHINE LEARNING

• Main Author: Faiz Baskara, Farrel
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/72789
• Institution: Institut Teknologi Bandung

High temperature material demands are becoming more and more challenging every year. High Entropy Superalloy (HESA) is a novel alloy being developed with the goal of achieving high strength at high temperature and low manufacturing cost. One main characteristic of material is creep resistance which can be described by Stacking Fault Energy (SFE) through deformation mechanism. SFE measurement can be approached in various ways, one of which is machine learning and thermodynamics calculations. This research aims to determine the effect of composition, temperature, magnetic contribution, and mixing entropy towards SFE, determine the best machine learning model, and provide guidance in designing FeNiCrAlCu high entropy superalloy. The simulation is carried out by determining the parameters used for thermodynamic calculation. These parameters are used to calculate the SFE value by utilizing Matlab software. The results of these calculation become the database for machine learning models namely random forest, support vector machines, and neural networks. The database is split into training and test data with ratio of 75:25. These models trained with the training data meanwhile the test data is used to evaluate models’ performance. Models’ performance is measured by several evaluation metrics such as root mean square error (RMSE) and accuracy. Based on the simulation results an increase in temperature increases the SFE value. At the temperature of 300 K: Ni atom addition increases SFE from 56.35 mJ/m2 to 100.53 mJ/m2, Al atom addition increases SFE from 33.11 mJ/m2 to 124.42 mJ/m2, Cr atom addition decreases SFE from 107.89 mJ/m2 to 17.97 mJ/m2, Fe atom addition decreases SFE from 79.93 mJ/m2 to 56.14 mJ/m2, and Cu atom addition decreases SFE from 82.26 mJ/m2 to 63.18 mJ/m2 due to the effect of other elements. Magnetic contribution towards SFE is significant at low temperature, and generally increase of alloy’s mixing entropy can decrease SFE value. The best machine learning model is Neural Network with RMSE of 1.75 and accuracy of 0.98. To design FeNiCrAlCu alloy at temperature of 300 K, the range of each element are Ni 20-25 at%, Cr 15-36 at%, Al 5-20 at%, Cu 9-20 at%, and Fe 20-35 at%.