In recent decades, metal additive manufacturing has seen rapid advancements, offering promising applications across various industries. However, addressing existing challenges in metal AM, such as process stability, defect avoidance, and quality control, is essential for fully exploiting its potential in fabricating parts with a desired geometry, as well as tailored microstructures and material properties. As a result, there is a pressing need for process optimization and control in additive manufacturing.Digital twins, as virtual counterparts of physical systems, provide valuable insights into AM processes and facilitate high-performance, real-time control. This thesis concentrates on developing physics-based and data-driven models to construct an integrated offline-online digital twin framework for metal additive manufacturing. In the offline component, physics-based computational models are employed to optimize process parameters under ideal conditions. In the online component, physics-informed data-driven models reveal actual process conditions and predict final part qualities using partially observed signals, facilitating feedback for online control applications. The major contributions of this thesis are: 1) Development of part-scale GPU-accelerated thermal and thermomechanical simulations for temperature and residual stress prediction in metal AM processes, and enhancing the efficiency of physics-based simulations in offline design applications. Experimental validation confirms the effectiveness of these models. 2) Creation of a data-driven model to reconstruct 3D melt pool geometry from 2D coaxial melt pool images, and a physics-informed machine learning model for predicting full-field temperature from partial temperature measurements. These models demonstrate the feasibility of revealing and predicting hidden process conditions from partially observed process signatures in AM processes, enabling online monitoring and control of otherwise unmeasurable process variables, such as 3D melt pool geometry and material cooling rates. 3) Establishment of a simulation-guided process design and control framework utilizing physics-based models. In the offline stage, time-series laser power is optimized based on simulation models to control process variables, including melt pool depth and temperature. In the actual process, the optimized time-series laser power serves as the feedforward input. Subsequently, a feedforward-feedback temperature controller is developed, integrating offline-designed laser power with online temperature measurements to achieve high-performance melt pool temperature control. This thesis showcases the effectiveness of the developed models in the process optimization and control of additive manufacturing processes. The physics-based and data-driven models presented herein will significantly contribute to advancements in metal AM.

Creator

• Liao, Shuheng

DOI

• https://doi.org/10.21985/n2-6g90-7048

Subject

• Mechanical engineering

Language

• en

Alternate Identifier

• http://dissertations.umi.com/northwestern:16572
• etdadmin_upload_985942

Keyword

• Physics-based Model
• Digital Twin
• Data-driven Model
• Machine Learning
• Additive Manufacturing

Date created

• 2023-01-01

Resource type

• Dissertation

Rights statement

• In Copyright