Show simple item record

dc.rights.licenseRestricted to current Rensselaer faculty, staff and students. Access inquiries may be directed to the Rensselaer Libraries.
dc.contributorMishra, Sandipan
dc.contributorSamuel, Johnson
dc.contributorManiatty, Antoinette M.
dc.contributorWang, Meng
dc.contributor.authorGuo, Yijie
dc.date.accessioned2021-11-03T08:56:56Z
dc.date.available2021-11-03T08:56:56Z
dc.date.created2018-02-22T16:05:41Z
dc.date.issued2017-12
dc.identifier.urihttps://hdl.handle.net/20.500.13015/2144
dc.descriptionDecember 2017
dc.descriptionSchool of Engineering
dc.description.abstractThe typical motion control accuracy requirement for ink-jet 3D printing is of the order of microns to guarantee feature resolution. Existing high performance motion stages typically rely on advanced hardware and can be very expensive. On the other hand, designing better feedback and feedforward controllers to realize high precision can reduce the cost of ultra high precision hardware. In ink-jet 3D printing, the motion paths of the stages are typically repetitive paths or can be separated into repeating segments. Thus, iterative learning control (ILC), a class of algorithms that are often used for tasks executed repeatedly to improve tracking performance from iteration to iteration, is well-suited for designing feedforward signals for the motion control system. However, there are two special requirements for the motion control of ink-jet 3D printing. First, the acceleration of motion must be constrained to minimize the deformation caused by acceleration. Second, the peak error in trajectory tracking is of significant importance, in addition to RMS error, so that each deposited droplet is within some small neighborhood of its prescribed location. Typical ILC algorithms are unable to handle these, thus we propose a constrained optimal iterative learning control (CO-ILC) framework that can use different norm cost functions: (1) an infinity-norm cost function, (2) a mixed (2-infinity)-norm cost function and (3) a sequential (2-infinity)-norm cost function to address different types of errors (average error and peak error). The proposed algorithm is validated experimentally on the linear stage in our testbed.
dc.description.abstractFrom a printing process control standpoint, typically the ink-jet 3D printing process is operated in an open-loop manner, i.e., no sensor measurement is used to create corrective actions during the printing process. As a result, the uncertainties during the printing process can result in undesired part geometry, such as edge shrinking, unreliable dimensions and uneven surfaces. One approach to deal with process uncertainties is through closed-loop feedback control. There are two key components to implement closed-loop feedback control, a control-oriented model and a properly designed feedback control algorithm.
dc.description.abstractThe overarching objective of this thesis is to further the state of the art for control of additive manufacturing (AM). AM refers to the general class of manufacturing processes that build up three-dimensional parts layer by layer. In this thesis, we focus on one particular AM process, namely ink-jet 3D printing, where-in parts are built by a print head jetting an ink (say, a photopolymer) layer by layer, with curing between successive layers of deposition. From a control standpoint, there are two key subsystems in an ink-jet 3D printer: the motion subsystem and the printing subsystem. The motion subsystem moves the building plate and the built part along a predetermined path relative to the print head. During this motion, the print head deposits the ink on the building plate at predetermined locations. In order to maximize performance of the 3D printer, both motion control and printing process control algorithms must be designed and hence, in this thesis, we will aim to address both.
dc.description.abstractIn this thesis, first two control-oriented models are proposed, the first is based on droplet geometry superposition, while the other is based on a graph-based characterization of local flow dynamics. The superposition model ignores the flow of the deposited liquid material, while the graph-based dynamic model is able to capture this phenomenon. These two models are compared with an existing flow-based empirical model and validated with experimental results. Then based on the superposition model, a model predictive control (MPC) like feedback control algorithm is proposed, its effectiveness is demonstrated by simulation studies. However, the high resolution of ink-jet 3D printing makes controlling this process a large scale problem, which can make the previous non-scalable algorithm extremely time-consuming. To address this issue, a scalable distributed MPC algorithm is proposed based on the second graph-based dynamic model. In this distributed MPC algorithm, the entire printing region is separated into sub-regions, the optimization problem constructed in MPC is also separated accordingly by reformulating its Lagrangian dual problem. By decomposing the large optimization problem into smaller subproblems, this distributed MPC algorithm reduces the computation time for control significantly and has excellent scalability. This distributed MPC algorithm is also validated though printing experiments.
dc.language.isoENG
dc.publisherRensselaer Polytechnic Institute, Troy, NY
dc.relation.ispartofRensselaer Theses and Dissertations Online Collection
dc.subjectMechanical engineering
dc.titleOptimization-based Control for Additive Manufacturing
dc.typeElectronic thesis
dc.typeThesis
dc.digitool.pid178846
dc.digitool.pid178847
dc.digitool.pid178848
dc.rights.holderThis electronic version is a licensed copy owned by Rensselaer Polytechnic Institute, Troy, NY. Copyright of original work retained by author.
dc.description.degreePhD
dc.relation.departmentDept. of Mechanical, Aerospace, and Nuclear Engineering


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record