This presentation will showcase as to how deep learning models will identify vials that have been falsely rejected due to a defect that current systems cannot differentiate. The work will begin with the inspiration for this project, and an overview of the technology involved in the neural networks utilized in this project. The solution focuses on how images were collected from several systems, then used to build a deep learning model capable of differentiating between true and false rejects for multiple different automated inspection machines. The next portion will be on how user interfaces were built around the models to convey the analysis done by the model in order to make it viable from a business perspective. The next portion will be on the case study where the model was tested with production batches specifically on identifying false rejects due to bubble generation in the upstream process. Finally, the results of the study will be highlighted as well as future opportunities, such as live machine learning for vision inspection, for tools like the one created here in vision inspection, and throughout the downstream processes affecting automated vision inspection.
Learning Objectives:
Learn about deep learning tool application in manufacturing
Learn how AI/ML/deep learning can be applied to reduce rejection rate in product manufacturing application works in product rejection in manufacturing
Learn how image analysis application works in product rejection in manufacturing