| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Frontiers of Machine Learning on Materials Discovery
|
| Presentation Title |
Exploring the Limits of Deep Learning for Synthetic Microstructure Generation of Titanium Alloy Microstructures: A Primer to Process-Structure Relationships and Microstructure Fingerprinting |
| Author(s) |
Gowtham Nimmal Haribabu, Michael D. White, Jeyapriya Thimukonda Jegadeesan, Christopher P Race, Philip J Withers, Bikramjit Basu |
| On-Site Speaker (Planned) |
Gowtham Nimmal Haribabu |
| Abstract Scope |
There is great interest in how to control metals microstructure through advanced manufacturing process control and optimisation. Machine learning algorithms can generate computationally expensive simulation results almost instantly, without knowing the governing laws. Deep learning algorithms, especially generative adversarial networks, have demonstrated outstanding performances in synthesizing highly realistic images.
In this work, the StyleGANv2-ADA model is explored as a means of generating synthetic microstructural images of conventionally processed titanium alloy. The effect of diffferent different augmentations is studied. Quantitative metrics and morphometric analysis are used to assess the performance of the generative model and compare microstructural features of real and synthetic images. StyleGAN-like models will be crucial in establishing structure-process relationships based on limited datasets (<600 images) typical of metallurgical process design studies.
In addition, pre-trained Variational Autoencoders are used to generate microstructural fingerprints of synthetic microstructures and compared with fingerprints of original microstructures by dimensionality reduction. |