Machine d''inspection automatisée pour le mirage "human like" de Vials et Seringues

The equipment offer the following functionalities: • Take vials from Akylux or denest vials or syringes from a nest • Make the products pass through no less than 6 color cameras and their lighting to cover all areas while taking a minimum of 24 images by camera • Place back the accepted units in Akylux or nest • Sort and place the rejected units in three separated Akylux or nests • Offer a throughput performance of 10 units/minute The vision controller, based on Occulus.AI, that will be distributed over several systems to reach the required performance will act to: - manage the image acquisition of 10 cameras (6 are used for inspection and 4 for pick/place operations) - process the images, including “traditional” vision operations (filtering, matching…) and deep learning inferences - store the images, with their metadata; it is used as a temporary local storage and the files are transferred to local FileServer (HMI PC) by a background task (low priority) - communicate with the PLC o Input recipe id (per process) o Input product id (per workstation) o Input start/stop/mode o Inputs … TBD o Output state o Output alarms o Output results o Outputs … TBD o Protocol will be based on OPC-UA The FileServer exchange is envisioned as: - Input model files (with a checksum to verify integrity); they are localized on the FileServer to avoid local duplication - Output images: their absolute paths and directories are related to their classification results. All the cameras are triggered by hardware. The vision system includes four types of inspection stations. Each association of a viewing angle, a specific illumination condition and a unit position and orientation will be called a “Pose”. Each pose requires a specific implementation of the inspection algorithm.