The Woodruff School

SUMMARY

Model-based approaches for manufacturing processes play a critical role in enabling first part correct production frameworks. For additive manufacturing processes, many factors determine the final quality of the manufactured components. While physical process simulations play an important role in enabling generalized process design, these are often not well suited for online control and generally are not well suited for component-scale process design due to the inherent computational complexity at that scale. Deep learning (DL) methods offer significant opportunities for online control and offline process design; however, there remains limited understanding regarding the fundamental applicability of such methods. This dissertation seeks to explore the suitability of deep learning-based approaches for online control and offline process design for additive manufacturing, with a particular focus on directed energy deposition and wire arc additive based processes. This work is organized in three complementary studies that explore applicability of deep learning for (1) online process monitoring and control and (2) offline process design. The first study aims to understand the ability of DL-based approaches such as support vector regression (SVR) and convolutional neural networks (CNNs) for regression for detection of anomalies in macro-scale part geometries resultant from deviations in process controllable parameters. The second study aims to understand whether neural networks (NNs) can be leveraged with physics-informed frameworks to improve quality and generalizability of subsequent predictions, as well as whether inverse NNs can be leveraged for prescriptive process design. The final study is designed to explore the generalizability of DL-based approaches to component-scale manufacturing problems, where additional factors such as thermal distortion may complicate predictive modeling. A key factor in these studies that will be considered includes understanding of feasibility to be integrated onto edge devices, with implications for feedback and feed-forward control of machine platforms. Altogether, these studies will inform a comprehensive understanding of the performance, uncertainty, and generalizability of novel data-driven approaches for complex, incremental processing methods such as additive manufacturing.

SUBJECT:	Ph.D. Proposal Presentation

BY:	Zoe Klesmith

TIME:	Monday, December 11, 2023, 1:00 p.m.

PLACE:	Virtual, Virtual

TITLE:	Data Driven Frameworks for Predictive and Prescriptive Control of Incremental Manufacturing Processes

COMMITTEE:	Christopher Saldana, Chair (ME) Thomas Kurfess (ME) Polo Chau (CSE) Peng Chen (CSE) Thomas Feldhausen (ORNL)

SUMMARY Model-based approaches for manufacturing processes play a critical role in enabling first part correct production frameworks. For additive manufacturing processes, many factors determine the final quality of the manufactured components. While physical process simulations play an important role in enabling generalized process design, these are often not well suited for online control and generally are not well suited for component-scale process design due to the inherent computational complexity at that scale. Deep learning (DL) methods offer significant opportunities for online control and offline process design; however, there remains limited understanding regarding the fundamental applicability of such methods. This dissertation seeks to explore the suitability of deep learning-based approaches for online control and offline process design for additive manufacturing, with a particular focus on directed energy deposition and wire arc additive based processes. This work is organized in three complementary studies that explore applicability of deep learning for (1) online process monitoring and control and (2) offline process design. The first study aims to understand the ability of DL-based approaches such as support vector regression (SVR) and convolutional neural networks (CNNs) for regression for detection of anomalies in macro-scale part geometries resultant from deviations in process controllable parameters. The second study aims to understand whether neural networks (NNs) can be leveraged with physics-informed frameworks to improve quality and generalizability of subsequent predictions, as well as whether inverse NNs can be leveraged for prescriptive process design. The final study is designed to explore the generalizability of DL-based approaches to component-scale manufacturing problems, where additional factors such as thermal distortion may complicate predictive modeling. A key factor in these studies that will be considered includes understanding of feasibility to be integrated onto edge devices, with implications for feedback and feed-forward control of machine platforms. Altogether, these studies will inform a comprehensive understanding of the performance, uncertainty, and generalizability of novel data-driven approaches for complex, incremental processing methods such as additive manufacturing.