Inspecting Multiple Soft-seated Connectors in a Single Recipe

Background & Challenge

One of the world’s largest electronics manufacturers needed an automated solution to verify the proper assembly of complex connectors, including connectors, cables, and security latches. The inspection was particularly challenging because the defects were subtle and difficult to distinguish, while reflections and variable lighting further increased inspection complexity.

Solution

To address these challenges, Advantech partnered with Overview AI to implement an AI classifier trained with only 10 images, including five good and five defective samples. The solution successfully identified multiple defect types within a single inspection recipe and generated a confidence score for each inspection result. This approach enabled the customer to establish clear pass/fail criteria, improve inspection reliability, and automate a complex quality control process.

System Diagram

Project Implementation

• ICAM-540: Advantech's industrial AI camera powered by NVIDIA Jetson Orin.
• OV GenAI Tools: Overview AI's Advanced GenAI Tools simplify inspection workflows and generate training data directly on the AI camera.

Benefits

• 100% Inspection Accuracy: Detects subtle defects despite reflections and variable lighting, achieving 100% accuracy across 33 test samples. 
• Minimal Training Data: Trains AI models with just 10 images to detect multiple defect types in a single recipe, reducing data preparation efforts and simplifying AI deployment.
• Reliable Defect Detection: Applies a 98% confidence threshold to identify defective parts, helping manufacturers improve inspection consistency and maintain quality standards.
• Rapid AI Deployment: Completes setup and model training in just 1.5 hours, accelerating implementation and reducing the time required to bring AI-powered inspection into production.

AI Vision Partner - Overview AI