| About this Abstract |
| Meeting |
MS&T23: Materials Science & Technology
|
| Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
| Presentation Title |
High-throughput, Ultra-fast Laser Fabrication of Alumina sample arrays and Deep-learning Based Prediction of Realistic Alumina Microstructure from Hardness |
| Author(s) |
Xiao Geng, Jianan Tang, Siddhartha Sarkar, Tianyi Zhou, Jianhua Tong, Rajendra K. Bordia, Hai Xiao, Dongsheng Li, Fei Peng |
| On-Site Speaker (Planned) |
Xiao Geng |
| Abstract Scope |
To rapidly explore the microstructure-property relationship in laser-sintered alumina, we demonstrated an ultra-fast fabrication of alumina sample arrays and the high-throughput hardness characterization of these sample units. These experimental data were used for deep-learning (DL) models training. We demonstrated an inverse, high-fidelity microstructure prediction method based on a modified generative adversarial networks (GAN), which we named as ‘Microstructure-GAN’. Compared to our previous DL-based microstructure prediction, the resolution of predicted images was greatly improved. The microstructure details (e.g., small pores and grain boundaries) can be clearly observed. The Features at the nanometer scale (~50 nm) were recognizable in the predicted 1000x micrographs. The accuracy of Microstructure-GAN prediction was validated by a pre-trained convolutional neural network (CNN). The relative root means square error (RRMSE) of the predicted micrographs was shown within 4.8% - 8.0% from the target hardness. It indicated our Microstructure-GAN had high accuracy and good robustness in prediction. |