Show simple item record

dc.contributor.advisorAlshraideh, Hussam
dc.contributor.authorZeinab, Zeinab Jihad
dc.date.accessioned2024-02-29T10:24:09Z
dc.date.available2024-02-29T10:24:09Z
dc.date.issued2023-12
dc.identifier.other35.232-2023.70
dc.identifier.urihttp://hdl.handle.net/11073/25480
dc.descriptionA Master of Science thesis in Engineering Systems Management by Zeinab Jihad Zeinab entitled, “Image-CNN Based Process Control of Profile”, submitted in December 2023. Thesis advisor is Dr. Hussam Alshraideh. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).en_US
dc.description.abstractFor quality inspection purpose, control charts have been widely adopted successfully in manufacturing industry throughout the years. Smart Manufacturing (SM) has emerged as a key concept for articulating the ultimate goal of manufacturing digitization as a result of the advancement of technologies like Artificial Intelligence (AI). For SM, an automatic process that can handle massive amounts of data from ongoing, concurrent processes is needed. In comparison, recognizing patterns in data and defect classification present challenges for typical control charts. To resolve these problems, Deep Learning (DL) algorithms proved to be an effective analytical tool that can aid in fault detection. The early classification of flaws and defects in machinery or manufacturing processes can be easily achieved by a detection monitoring system capability. In this thesis, a DL-based framework for monitoring profile generating processes is presented. The framework relies on the presentation of profile time series data as two-dimensional images, for which four transformation algorithms were explored including Gramian Angular Field (GAF), Markov Transition Field (MTF), and Recurrence Plots (RP). Proposed framework was evaluated through two case studies. In the first one, a tapping process is considered while a 3D printing process is considered in the second case. Proposed model achieved an accuracy level of 91.6% for the tapping dataset outperforming previous model performance reported in the literature of 84.04%. Similarly, the model showed an improved performance level over existing literature for the 3D printing process data with accuracy levels of 96.6% and 92.6% for the small and large versions of the data, respectively. Our proposed framework provides an automatic feature extraction step as it relies on DL technology providing a major advantage over existing models in the literature that assume a preexisting set of features to be used.en_US
dc.description.sponsorshipCollege of Engineeringen_US
dc.description.sponsorshipDepartment of Industrial Engineeringen_US
dc.language.isoen_USen_US
dc.relation.ispartofseriesMaster of Science in Engineering Systems Management (MSESM)en_US
dc.subjectDeep Learningen_US
dc.subjectQuality Process Controlen_US
dc.subjectManufacturingen_US
dc.subjectTime series classificationen_US
dc.subjectGramian Angular Fielden_US
dc.subjectMarkov Transition Fielden_US
dc.subjectRecurrence Plotsen_US
dc.titleImage-CNN Based Process Control of Profileen_US
dc.typeThesisen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record