Visual detection of mechanical assembly errors in optical devices

Context

Manufacturer of microscopes and optical measuring instruments. In the assembly of optical devices, component angles decide measurement quality — deviations of a few tenths of a degree are invisible to the eye.

Problem

Subtle mechanical assembly errors such as 89.6° instead of 90° went undetected in inspection and only showed up later in the measurement chain.

What we built

A computer-vision-based system that recognises from the image data how the device is assembled and assesses the overall assembly quality.

Key engineering decision

Integrate the correlation between image data and assembly configuration into the measurement chain. Instead of measuring angles directly, the system learns how the image data changes depending on different assembly configurations — and uses exactly that relationship to detect assembly errors.

Result

• Detection of angle deviations such as 89.6° instead of 90° in optical devices
• Assessment of overall assembly quality directly from the image data