Adaptable automated visual inspection: scheme and implementations

Simple item page
dc.contributor.advisorSun, Qiao
dc.contributor.authorSun, Jun
dc.date.accessioned2017-12-18T22:30:29Z
dc.date.available2017-12-18T22:30:29Z
dc.date.issued2012
dc.descriptionBibliography: p. 186-197en
dc.descriptionSome pages are in colour.en
dc.description.abstractIn the manufacturing industry, there is a growing need for an automated visual inspection (A VI) system that can perform inspection tasks, adapting to job changes without excessive retuning efforts. This thesis presents an adaptable A VI scheme (AA VIS), with which an A VI system can be trained or updated on-line to accommodate new patterns. As such, the system can be quickly deployed to undertake new inspection tasks or adapt to new operation conditions. The key element in the design of the proposed AA VIS is an adaptable inspection model that integrates subject localization, feature extraction, and classification functions as sub-models. When implemented using machine learning techniques, the inspection model can be trained or updated on the go. Efficient online learning strategies are proposed for the implementation of the AA VIS, including training sample selection, minimizing manual verification efforts, and training sufficiency estimation. For subject localization, a novel template matching method is developed to allow tolerance against image distortions such as rotation and changes in size and position. The edge-based distance transform (DT) technique is employed in this method with a coarse­to-fine matching strategy. An efficient heuristic searching algorithm is proposed to achieve searching efficiency and matching accuracy. Upon identifying the Region of Interest (ROI), image features are extracted through the principal component analysis (PCA). An efficient online training algorithm is proposed to construct the feature model. A temporary classification model is used to assist the online training of feature model. In addition, a more sophisticated classification method such as support vector machine (SVM) technique is employed to construct a reliable classification model. An efficient online SVM algorithm is proposed for training the classification model, which emphasizes on the selection of model parameters, online updating, misclassification­penalty adjustment, and training sufficiency estimation. Throughout the development of the AA VIS, data collected from an automobile­parts assembly inspection cell are used to assist system implementation and verify the fulfilment of design goals. The unique features including reliability, efficiency and adaptability provided by the proposed AA VIS will help increase the benefit and functionality of an A VI technique to the improvement of manufacturing quality and productivity, as well as cost reduction.
dc.format.extentxix, 206 leaves : ill. ; 30 cm.en
dc.identifier.citationSun, J. (2012). Adaptable automated visual inspection: scheme and implementations (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/4706en_US
dc.identifier.doihttp://dx.doi.org/10.11575/PRISM/4706
dc.identifier.urihttp://hdl.handle.net/1880/105707
dc.language.isoeng
dc.publisher.institutionUniversity of Calgaryen
dc.publisher.placeCalgaryen
dc.rightsUniversity of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.
dc.titleAdaptable automated visual inspection: scheme and implementations
dc.typedoctoral thesis
thesis.degree.disciplineMechanical and Manufacturing Engineering
thesis.degree.grantorUniversity of Calgary
thesis.degree.nameDoctor of Philosophy (PhD)
ucalgary.item.requestcopytrue
ucalgary.thesis.accessionTheses Collection 58.002:Box 2095 627942967
ucalgary.thesis.notesUARCen
ucalgary.thesis.uarcreleaseyen
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
thesis_Sun_2012.pdf
Size:
89.6 MB
Format:
Adobe Portable Document Format
Description:
Thesis
Download
Collections
Legacy Theses