|About this Abstract
||NUMISHEET 2022: The 12th International Conference on Numerical Simulation of 3D Sheet Metal Forming Processes
||Web Scraping and Data Mining of Microstructure and Formability Data of Aluminum Alloy Sheets for Machine Learning
||Lalit Kaushik, Ki-Seong Park, Shi Hoon Choi
|On-Site Speaker (Planned)
||Shi Hoon Choi
The prediction capabilities of a machine learning (ML) model heavily depends on both the quality and quantity of data used for the training. Dataset creation remains the biggest challenge as it is quite labor intensive and time-consuming task. To this end, web scraping techniques becomes very handy to programmatically collect data available on scientific platforms such as ScienceDirect. This technique was applied with python libraries to automatically extract the published scientific data such as chemical composition, manufacturing history, microstructural and textural attributes which affect the formability of the aluminum alloys. Furthermore, the data mining based on the image processing and optimization algorithms was deployed on the extracted data (text and images) to augment the data features. Moreover, mean field visco-plastic self-consistent (VPSC) simulations were performed to obtain formability indicators such as plastic strain ratios (R-values) and forming limit diagrams (FLDs) which are target values for supervised learning. Additionally, visco-plastic fast Fourier transformation (VPFFT) simulations were performed on the optimized microstructure to analyze the strain localization and heterogeneities of deformation behavior.
||Definite: At-meeting proceedings