Implementing AI vision for Quality Inspection within a Manufacturing Environment

Abstract

All manufacturing companies strive to ensure high quality for their products, not least concerning manual assembly, in order to remain competitive. Despite all preventive strategies available to eliminate the occurrence of deviations, additional reactive quality inspections are sometimes required. These are primarily performed by humans since they are flexible and can easily be placed wherever needed and learn new tasks. However, there are also drawbacks with using manual inspections, as these tasks are both demanding as well as costly to implement. Therefore, companies now seek to take advantage of the constant technological development to explore new methods that are both cost-efficient and easy to install quickly, wherever needed. This thesis presents an explorative study about implementing a suggested cost-efficient AI vision system to detect product deviations in manual assembly. The study emphasises both drawbacks and advantages of the technology in its current state. Through exploratory testing, several impactful factors were identified that influence the results of AI vision. A robustness test was thereafter conducted to evaluate the system's sensitivity towards changes. This was done to establish requirements and limits for implementation. Lastly, a suggested setup for implementation was tested in a manufacturing environment to validate the findings from earlier testing.