Automated Defect Detection

Challenge

Our client, an expert in the manufacture and sale of smart home appliances, sought to improve the quality of one of its production lines while reducing waste. A key focus was detecting microcracks on gas mixer valves, which, if left unnoticed, could lead to leakages or breakages during final assembly. Detecting these defects manually was too time-consuming, making automation essential for improving production efficiency, reducing rework costs, and enhancing product quality.

Nature of collaboration

To address this challenge, we worked with our client to implement an AI-driven visual inspection system. Through collaboration, we deployed a defect detection model trained to identify microcracks efficiently, ensuring a faster and more reliable inspection process.

What we built

The visual inspection system is operational in one of our client’s factories, integrated into a vision station with robotic arms. The system:

• Uses robotic arms to pick up and present valves to a precision camera.
• Captures detailed images analyzed by an AI algorithm to detect microcracks.
• Automatically sorts valves based on detection results, directing them either for recycling or production.

To develop the AI model, we first deployed a data annotation system, enabling our client’s team to build a labeled dataset. This dataset was used iteratively to train and refine the defect detection model, which now enables real-time inspection with a cycle time of under three seconds.

Schematic view of the Sagacify’s vision system deployed on a rotary
table with robotic arms manipulating parts to take pictures from all angels.

Impact