Scripting for Video Inspections

The purpose of this study was to enhance the reliability of deployed components for nuclear weapons by improving the quality and consistency of the final visual inspection before deployment of the weapon. Currently, most of these parts are subjected to manual visual inspection, a process subject to errors related to consistency and accuracy. Furthermore, there is no method for documentation of the inspection. Even state-of-the-art commercial visual inspection systems are inadequate because they require skilled engineers to artfully adjust camera parameters, lens settings, lighting, and processing for each new inspection, part or defect. In this project, we developed technology to automate much of the setup for visual inspection. We implemented graphical simulations and CAD models of the camera, positioning system and parts to be inspected. Using these tools, the system automatically generates camera locations and settings for the camera lens and illumination to produce optimal images for inspection. We refer to this combination of imaging parameters as a “script”. The “script” contains all the instructions necessary to perform the inspection when run on the actual inspection system. In addition, we developed a cataloging system to allow storage and retrieval of the video images gathered during an actual inspection of real parts. We developed several feature extraction algorithms that were able to detect defects in the actual parts but were unable to achieve comparable results using the CAD based parts. A version of this system has been implemented at the neutron generator facility as a key element of implementing the process-based quality manufacturing.