Date of Award

2024-05-01

Degree Name

Master of Science

Department

Computational Science

Advisor(s)

Zhengtao Gan

Abstract

Laser Powder Bed Fusion (L-PBF) is a renowned additive manufacturing technique, celebrated for its capability to construct intricate metal components with remarkable precision. However, one of the main challenges with L-PBF is the formation of complex microstructures, which can significantly impact the final material properties. To address this issue, our study proposes a physics-guided and machine-learning-aided approach to optimize scan paths for achieving desired microstructure outcomes, such as the generation of equiaxed grains that enhance material properties. By using phase-field modeling, a physics-based computational method, we gain insights into microstructure evolution. To reduce computational costs, we train a surrogate machine-learning model using a 3D U-Net convolutional neural network and single-track phase-field simulations as dataset. This enables the machine learning model to predict crystalline grain orientations accurately based on the initial microstructure and thermal history. As a preliminary approach, we investigate three primary scanning strategies; vertical serpentine, spiral serpentine and diagonal scanning at various hatch spacings to identify the most effective paths for achieving the desired microstructure. This lays the foundation for a comprehensive examination of how different scan paths and parameters affect the resulting microstructure. By combining this strategic analysis with our advanced modeling techniques, we provide insights into how scan path influences the attainment of optimal crystalline grain structure in L-PBF processes. This approach not only enhances our ability to predict microstructural outcomes but also advances the precision manufacturing capabilities of L-PBF, merging theoretical knowledge with practical application to guide future advancements in additive manufacturing. Importantly, our methodology achieves a computational time reduction by approximately three orders of magnitude, underscoring the efficiency of our ML approach in accelerating the design process.

Language

en

Provenance

Received from ProQuest

File Size

57 p.

File Format

application/pdf

Rights Holder

Augustine Twumasi

Download

Share

COinS