Intelligent Additive Manufacturing: a Holistic Approach to the Optimization of Data-Poor Complex Processes
Open access
Author
Date
2024Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
Additive manufacturing has introduced unprecedented novelty to the way we conceive, design, and produce objects. Additive processes offer several advan- tages over traditional manufacturing methods, such as design freedom, material and resource efficiency, and shortened supply chains. However, a large gap ex- ists between the theoretical capabilities of additive manufacturing and its current implementations.
Numerous additive manufacturing techniques currently operate below the efficiency and performance levels that could be achieved by implementing ad- vanced sensing, control, and optimization strategies. Fused filament fabrication and atmospheric plasma spraying certainly rank among these processes. Both methods are poorly captured by physics-based or machine learning models, use no real-time feedback from the process, and are generally operated with open-loop and suboptimal strategies. These characteristics ultimately increase the economic and environmental cost of manufacturing. They also limit the applicability of these versatile techniques to simple use cases.
In this thesis, we aim at unlocking the full potential of additive manufacturing in general, and of fused filament fabrication and atmospheric plasma spraying in particular. To do so, we propose a comprehensive suite of methods that includes optimization, sensing, modeling, and control at all timescales. Our holistic ap- proach is directed by a systems theoretic perspective. We address the general process optimization task in a hierarchical fashion to efficiently optimize all the interconnected subsystems that constitute an additive manufacturing process.
We begin by presenting two optimal approaches to trajectory generation, for processes in which the properties of manufactured parts are affected by the deposition direction. Both methods are applied to fused filament fabrication. The first method produces stress-aligned non-planar trajectories, and was tested experimentally on load-bearing brackets, showing a 44× improvement in failure strength and a 6× improvement in stiffness over conventional printing. The sec- ond method focuses on computational efficiency. Compared to the first method, it was shown to produce trajectories that increase the stiffness of a planar specimen by a further 9 % while reducing the trajectories computation time by 99 %.
Next, we introduce two methods for the configuration and continuous adap- tation of manufacturing process parameters. We propose a novel Bayesian op- timization acquisition function that efficiently guides the search for optimal parameters, and validate it in simulation and experiments. We then extend this probabilistic technique by fusing it with a filtering algorithm that predicts the equipment status at the next manufacturing run. We demonstrate that these methods enable practitioners to quickly discover suitable atmospheric plasma spraying parameters, and to continuously update them to counteract process drift.
Finally, we present two novel sensing techniques developed to collect data from a fused filament fabrication process at different time scales. The measure- ments from each sensor are used to perform feedback control of the printing process at the respective sampling rate. With a laser sensor, we evaluate the quality of deposited layers and use these data to reduce manufacturing time by 66 % with an automated parameter adaptation algorithm. We then introduce Force Controlled Printing, a novel approach to the closed-loop control of fused filament fabrication. Using feedback from a real-time extrusion sensor, we are able possible to deposit lines of desired width, to reject large disturbances in the process, and to produce high-quality parts, significantly outperforming the ubiquitous open-loop approach. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000685453Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichOrganisational unit
02650 - Institut für Automatik / Automatic Control Laboratory03751 - Lygeros, John / Lygeros, John
Funding
180545 - NCCR Automation (phase I) (SNF)
More
Show all metadata
ETH Bibliography
yes
Altmetrics