Camera-Based Quality Inspection: Liquid Fill Level Control
How we check fill level in bottles and jars at line speed with backlighting and edge detection. An application combining classical methods with deep learning.
On filling lines, under- or over-filled products are both a cost and a compliance problem. Mechanical and capacitive sensors need readjustment whenever the product geometry changes; camera-based inspection manages recipe changes in software.
Lighting is everything
In transparent and translucent liquids the level line has low contrast. With backlighting the liquid column stands out as a silhouette; an edge-detection algorithm measures the level with millimeter precision. For colored liquids narrow-band filters are used.
Classical method + deep learning
Level measurement is done classically, while exceptional situations such as foam, drops and label shift are left to a deep-learning classifier. PIROBOTIK Quality Check runs both engines in the same flow; the result is sent to the PLC as a digital output and the faulty product leaves the line.
On a typical line we achieve 120 products per minute, 99.6% correct decisions and traceability down to the serial number.