QualiZer
Reduce scrap and rework by detecting defects at the line with edge CV and explainable reviews.
QualiZer is in-line defect detection that inspects every unit on the line with edge computer vision, holds or alerts the MES when a defect crosses your threshold, and shows the operator the frame and the reason so a false reject can be overruled in seconds.
Who is this for?
Quality managers and plant engineers on lines where sampling inspection lets escapes through and where a reject needs a reason, not a red light. It runs on industrial GigE and USB3 cameras, often the ones already fitted, and talks to the MES and SCADA you have.
What problem does it solve?
Sampling catches a share of defects and the trend catches the rest a month late in the scrap number. A manual inspector is accurate for the first hour of a shift and human for the rest of it, and the defect that escapes to the customer becomes a supplier claim with no evidence on your side.
The other failure is the opposite one: an automated system that rejects good parts and cannot say why. Operators stop trusting it, the line slows, and the system gets switched off quietly within a quarter.
How does it work?
High-speed cameras capture every unit and an edge device classifies it locally in under 100 milliseconds, with no cloud in the loop, so a network outage does not stop the line. Critical defects can trigger an immediate line stop; minor anomalies log for batch review; the thresholds are configurable by defect type and severity.
The reviewer console shows side-by-side frames with reason codes, so an operator validates or overrules a detection and the system learns from the correction. The false-reject rate is a core KPI, trended by SKU, shift and tool, which is where the real causes tend to show up. Evidence exports feed supplier claims and operator training.
Alerts, hold signals and KPI data flow to the MES and ERP through REST APIs with audit trails, with pre-built connectors for Siemens, Rockwell and Ignition. Where an AOI or 3D sensor already exists, its output can be fused in rather than replaced.
Capabilities
- Real-time defect cues on conveyor/fixture views
- Reviewer console with side-by-side frames & reason codes
- Trend by SKU/shift/tool; alerts to MES/Andon
- Evidence export for supplier claims & training
- Optional AOI/3D fusion where available
Outcomes
- Scrap/rework ↓ 10–25%
- Inspection time ↓ 30–60%
- False rejects ↓ 10–30%
What does it connect to?
- MES and SCADA: Siemens, Rockwell and Ignition connectors; MES and ERP alerts and holds through REST APIs
- Industrial cameras (GigE, USB3), with a site survey to confirm optics and lighting
- Andon and alert routing to operators
- Optional AOI and 3D sensor fusion
How is it deployed, and how long does it take?
Edge-first on NVIDIA Jetson or industrial PCs; the fastest lines use strobe lighting and frame selection so coverage stays complete without overloading the compute. Starting data is sample images of good and bad parts, scoped in the proof of concept.
First-pass yield usually moves within 4 to 6 weeks of the pilot. Full return takes 3 to 6 months as the system spreads across lines, and we would rather say that than promise the first week.
Proof of concept to pilot in 8 to 12 weeks with KPI gates and a reversible rollout. Scrap and rework typically fall 10 to 25 percent, inspection time 30 to 60 percent, false rejects 10 to 30 percent.
Industry context
Explore the related workflows and other solutions for your industry.