Autonomous Quality Control for an Industrial Manufacturer

The Challenge

A family-owned manufacturer of automotive components, based in the province of Valencia with over two decades of history, was facing a critical operational challenge: its three production lines generated a volume of parts impossible to inspect reliably through manual methods.

The undetected defect rate had reached 4.2%, translating into client complaints and returns that were eroding both margin and reputation. The company evaluated cloud-based machine vision solutions, but the projected budget — EUR 2,800 per month in API costs — exceeded what the business could sustain as an indefinite recurring expense.

The operations director put it plainly: they needed a solution that would outlast any subscription contract.

The Solution

We designed an edge-first architecture with three local inference nodes — one per production line — each running vision models optimised for the specific part type being inspected.

Deployment components included:

• Hardware: Three compact inference units installed beside the production lines, connected to the existing high-resolution industrial cameras
• Vision models: Trained on 8,000 reference images from the client’s own catalogue, including positive and negative examples
• Predictive maintenance: The same nodes monitor vibration, temperature, and current draw from the main motors to anticipate failures before they occur
• Operational dashboard: Real-time panel visible from the shop floor and management office, with configurable alerts via WhatsApp and email

The solution runs without internet connectivity. A network outage does not stop inspection or compromise production data.

The Results

After six months of continuous operation:

• The undetected defect rate fell from 4.2% to 0.25%
• Customer complaints related to quality dropped by 89%
• The system has processed over 2.3 million inspections without human intervention
• The predictive maintenance module prevented two unplanned stoppages, with an estimated saving of EUR 34,000 in downtime costs
• The total project cost was recovered in 14 months, compared with 38 months for the cloud alternative

The company has initiated conversations to replicate the deployment at its second plant in Castellón.

Technology Used

• Hardware: 3x NVIDIA Jetson Orin Nano + existing industrial cameras
• Models: Custom-trained computer vision model using 8,000 catalogue images
• Dashboard: Real-time web panel with WhatsApp and email alerts
• Timeline: 6 weeks development, 2 weeks calibration per line

VORLUX AI Perspective

Spanish manufacturing has a unique opportunity: powerful and affordable inference hardware combined with mature vision models puts quality control automation within reach of SMEs that until now could only aspire to it on strategic planning spreadsheets.

The key is not the technology itself — it is deploying it in a way that makes costs predictable, keeps data on the factory floor, and ensures the system works when the cloud provider is unavailable.

Request a free assessment →