Deep Learning Pipeline for Defect Detection

Deep Learning is successfully applied for detecting visual quality defects but comes with several challenges, e.g., annotating the data. Various tools, technologies, and frameworks have been introduced to simplify the implementation of DL. However, manufacturing companies often lack a general understanding of structuring a DL project and making relevant design decisions. In this paper, we demonstrate a methodological approach to integrate DL for defect detection in production. As a result, challenges and decisions are outlined in a pipeline that guides users through the DL project development process. The pipeline is validated based on surface defect detection for the lithium-ion battery coating process.