Industrial Manufacturing 2025-02

Computer Vision for Quality Inspection

Implementation of an edge computing-based computer vision system to detect millimeter-scale defects in assembly parts in real time.

99.8% Detection Accuracy
< 0.5% False Positives
120 parts/min Scanning Speed
15% annually Material Savings

The Challenge

An automotive component production line suffered high rejection rates from end customers due to defects that were imperceptible to the naked eye during manual quality control. The speed of the line made exhaustive inspection by operators impossible.

The Solution

We deployed an Edge Computing system with high-speed industrial cameras:

  1. Real-time Capture: Images synchronized with the assembly line.
  2. Local Inference: A CNN (Convolutional Neural Network) model trained specifically with thousands of defect images and executed on Edge TPUs to guarantee < 20ms latency.
  3. PLC Integration: Direct communication via PLC to automatically eject anomalous parts from the line.

The Result

The system reduced customer complaints to zero within the first three months and allowed the factory to increase production speed without sacrificing quality, achieving significant savings in material waste.

Tech Stack

Python OpenCV TensorFlow Edge TPU