Case Study: Detecting minute bent pins and dust particles with two images and a telecentric lens

Problem:

Manual inspection of the header connectors was slow, costly, and unreliable, leading to a consistent 2% miss rate for subtle pin bends. This was hard due to:

• Perspective Distortion: With a standard lens, pins on the connector edge appear to be at a different angle than the center causing overkill.
• 3D Variation: Pins can be bent slightly in any direction (X, Y, or Z axis), which is challenging for 2D rule-based systems to handle reliably.
• Multiple defects: In addition to bent pins, dust particles could settle on connectors. These were individually, manually cleaned prior to packing.

Solution:

We implemented a classifier algorithm on the OV80i using a telecentric lens with 0.3x magnification to reduce perspective error. This setup ultimately allowed us to reliably capture all 7 defect types in 1 system.

Setup:

Step 1: Hardware setup

• Camera: OV80i at a working distance of 110 mm
• Lens: 0.3x magnification telecentric lens
• Light: Standard ring light

Step 2: Alignment and inspection setup

Set up alignment so each component can be reliably inspected, independent of how they are placed by machinery. Left and bottom edges are used for alignment

Create two types of inspections:

• ROI covering top and bottom corners for reference measurements
• ROI covering pin tops

Step 4: Training Classification

We trained on 2 samples - 1 good connector where all pins were correct, 1 bad connector where all pins were bent. Total of 28 ROIs (14 pins on each)

Step 5: Testing

Tested on 15 good samples and 10 defective samples, achieving 100% accuracy with zero misses and zero overkills. Examples of a pass, fail due to a bent pin, and fail due to debris and a bent pin are shown below