Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Electrical and Computer Engineering (Holcomb Dept. of)

Committee Chair/Advisor

Hai Xiao

Committee Member

Fei Peng

Committee Member

Jianhua Tong

Committee Member

Yongkai Wu

Committee Member

Adam Hoover


Material microstructure prediction based on processing conditions is very useful in advanced manufacturing. Trial-and-error experiments are very time-consuming to exhaust numerous combinations of processing parameters and characterize the resulting microstructures. To accelerate process development and optimization, researchers have explored microstructure prediction methods, including physical-based modeling and feature-based machine learning. Nevertheless, they both have limitations. Physical-based modeling consumes too much computational power. And in feature-based machine learning, low-dimensional microstructural features are manually extracted to represent high-dimensional microstructures, which leads to information loss.

In this dissertation, a deep learning-guided microstructure prediction framework is established. It uses a conditional generative adversarial network (CGAN) to regress microstructures against numerical processing parameters. After training, the algorithm grasps the mapping between microstructures and processing parameters and can infer the microstructure according to an unseen processing parameter value. This CGAN-enabled approach consumes low computational power for prediction and does not require manual feature extraction.

A regression-based conditional Wasserstein generative adversarial network (RCWGAN) is developed, and its microstructure prediction capability is demonstrated on a synthetic micrograph dataset. Several important hyperparameters, including loss function, model depth, number of training epochs, and size of the training set, are systematically studied and optimized. After optimization, prediction accuracy in various microstructural features is over 92%.

Then the RCWGAN is validated on a scanning electron microscopy (SEM) micrograph dataset obtained from laser-sintered alumina. Data augmentation is applied to ensure an adequate number of training samples. Different regularization technologies are studied. It is found that gradient penalty can preserve the most details in the generated microstructure. After training, the RCWGAN is able to predict the microstructure as a function of laser power.

In-situ microstructure monitoring using the RCWGAN is proposed and demonstrated. Obtaining microstructure information during fabrication could enable accurate microstructure control. It opens the possibility of fabricating a new kind of materials with novel functionalities. The RCWGAN is integrated into a laser sintering system equipped with a camera to demonstrate this novel application. Surface-emission brightness is captured by the camera during the laser sintering process and fed to the RCWGAN for online microstructure prediction. After training, the RCWGAN learns the mapping between surface-emission brightness and microstructures and can make prediction in seconds. The prediction accuracy is over 95% in terms of average grain size.



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.