
In high-speed food, beverage, CPG, and pharma packaging lines, manual interval sampling leaves critical blind spots. When a continuous inkjet (CIJ) printhead degrades or clogs between checks, hundreds of unreadable or missing codes slip through — leading to quarantine, rework, retailer chargebacks, and traceability risks under regulations like FSMA and MoCRA.
In this episode, we dive into Deep Learning OCR (DL-OCR) and how pre-trained neural network models are replacing traditional template-based vision systems. We explore:
- Why Rule-Based OCR Fails: How specular glare, curved substrates, dot-matrix drift, and tight regions of interest trigger false rejects and cost hours of manual tuning.
- The Deep Learning Difference: How models pre-trained on over 200 font families and millions of images interpret full character strings in context without needing per-job font libraries.
- On-Camera Processing at Line Speed: How hardware with embedded Neural Processing Units (NPUs) — like the Zebra FS42 scanner and NS42 vision sensor — delivers read times as fast as 40 ms at up to 900 parts per minute.
- Engineering Best Practices: Key steps for setting up 100% code verification, pushing expected strings directly from the PLC/MES, and establishing fail-safe reject mechanisms.
Whether you are a controls engineer, plant manager, or quality leader looking to move from reactive sampling to proactive 100% verification, this breakdown covers what you need to specify for your line.
1-888-600-3080
Hosted on Acast. See acast.com/privacy for more information.
Mais episódios de "Manufacturing Innovations Today"



Não percas um episódio de “Manufacturing Innovations Today” e subscrevê-lo na aplicação GetPodcast.








