Collaborative human-machine quality control system: Steps towards automatic machine vision inspectio

This thesis explores ways of increasing the efficiency of an industrial process by resorting to automated machine vision technologies. The research focuses on the quality inspection process in the tire industry.;The general trend found in the literature to improve the efficiency of quality inspection processes is to introduce machine vision systems to replace humans in the visual search and conformity decision tasks. The original contribution of this study is showing that operators should be integrated in the development process and perform continuous validation of each technological sub-component. In such an ambiguous and complex task as quality inspection process of tires, operators' expertise and knowledge needs to be acquired to assure that the technological solutions being proposed sustain the same quality standards. Thus, the machine vision solutions developed during this research project do not aim at replacing the operators, but rather at maximizing the advantages they bring to the inspection system through a Computer Assisted Inspection (CAI). This will continue up to a moment in which technology reliability is demonstrated as adequate and the automated solutions can be deployed as a stand-alone inspection method.;The thesis is based on qualitative and quantitative research undertaken over three years in collaboration with Continental Mabor SA. Initial chapters explore the current inspection methods used by specialized operators. Later chapters describe the underlying concepts and the re-design process of the inspection system. This proposed process follows a framework that considers scenarios of different levels of automation. A prototype suitable for industrial environment was developed and made possible proving the validity of the proposed solution. Each sub-component of the system was tested and validated through systematic experimentation. Special focus was given to the image-acquisition station, as the appropriateness of the images influences both human-based and automatic subsequent quality assessments.;In the chapters focused on the results it is shown that combining operators' knowledge, machine vision technologies and automatic detection algorithms contribute to an increase in process efficiency (higher throughput) and effectiveness (increase the number of correct decisions). The baseline strategy for automatic imperfection detection based on a selfadaptive and deformable template match (SAD-TM) technique is proposed in this dissertation and validated for a number of cases. Future work should focus on the continuous development of automatic detection algorithms, enlarging number of imperfections tested and refining its detection capabilities.;The main outcome of this thesis is the development on the understanding of the potential benefits of introducing machine vision technologies in the quality inspection process of tires. The proposed strategy of complementing human and automation towards the development of more efficient processes is expected to be applicable in other environments besides the tire industry.;Regarding the outcomes that are relevant to the industrial partner, the performed research suggests that the industrial implementation of the proposed system is viable and should occur iteratively, attempting to a continuous increase of level of automation.