Particulate Matter Inspection
Detect and classify particles and impurities in vials, ampoules, and filled syringes
There are a number of contaminating particles that can end up inside of a vial or ampoule, from contaminants during the manufacturing process, to fragments of the primary packing itself, including glass. Glass particulates can result from the manufacturing process, particularly the blowing process of ampoules, or from handling anywhere along the production line. Presence of particulates are among the top ten reasons for recalls of liquid pharmaceuticals.
Special lighting is required to not only penetrate the container but also ensure sufficient contrast for the smallest particles of interest. Unfortunately, this lighting can also result in confusing reflections and halation.
Particles can be difficult to identify in the liquid since they can range in color from dark to light, move differently in the fluid, are different sizes and shapes, can be confused with bubbles and are in different container materials and shapes that have varying light transmission qualities.
All of these factors make this application challenging to solve with conventional machine vision.
Cognex Deep Learning combined with High Dynamic Range (HDR+) technology is an ideal solution for particulate matter inspection. Cognex Deep Learning is trained on all types of particles appearing in vials and ampoules—the various shades and sizes, whether or not mixed with bubbles, and their appearance through the range of reflections and refractions of vial and ampoule glass.
Cognex High Dynamic Range Plus (HDR+) technology produces uniformly illuminated images, minimizing halation and reflectivity, in a single capture without compromising line speed. HDR+ differs from standard HDR as it can be done with a single acquisition at high-speed on moving parts, whereas standard HDR would need to be stationary and capture multiple images to obtain the same results.
To distinguish particles floating in the liquid from surface features, the vial or ampoule cylinder is spun and then stopped. A series of images taken as the particles slow and settle allows deep learning to clearly distinguish between spots in the fluid and in the container.