Publication Date
2025
Document Type
Thesis
Committee Members
Hamed Attariani, Ph.D. (Advisor); Henry D. Young, Ph.D. (Committee Member); Ahsan Mian, Ph.D. (Committee Member)
Degree Name
Master of Science in Mechanical Engineering (MSME)
Abstract
Additive Manufacturing (AM), particularly Laser Powder Bed Fusion (L-PBF), has gained significant traction in fabricating complex, high-performance metallic components. However, the inherent complexity and computational cost of high-fidelity simulations pose challenges for real-time monitoring and optimization of multi-laser powder bed fusion processes. This study proposes a data-driven surrogate modeling approach using a deep learning architecture to efficiently and accurately predict three-dimensional temperature distributions during ML-PBF. A 3D convolutional neural network (CNN) model, named Decoder-CNN, is developed and trained on a dataset of simulated thermal fields corresponding to various process configurations, including different laser power, scanning speed, and beam arrangements. The model architecture integrates a multi-layer perceptron (MLP) for process parameter encoding and a decoder network to reconstruct the thermal field in full 3D resolution. To improve model generalizability and spatial precision, data augmentation techniques were applied. The proposed surrogate model achieves strong predictive performance, with an average relative root mean squared error (RMSE) of approximately 5% in the solidifying regions, an Intersection over Union (IoU) exceeding 85%, and an R2 score around 90% when compared with Finite Element Method (FEM) ground truth. Visualization of qualitative results confirms that the model can reliably capture the geometry and boundaries of the melt pool, despite minor deviations at sharp interfaces due to MSE smoothing effects. This work demonstrates the viability of using deep learning as a fast and reliable alternative to computationally expensive simulations, paving the way for real-time process monitoring, in-situ control, and design optimization in advanced metal AM workflows.
Page Count
97
Department or Program
Department of Mechanical and Materials Engineering
Year Degree Awarded
2025
Copyright
Copyright 2025, all rights reserved. My ETD will be available under the "Fair Use" terms of copyright law.
ORCID ID
0000-0002-2453-2787
