Background
Component alignment had previously been checked visually. The goal of the project was to standardize that inspection with an automated pass/fail system and create a traceable record for each inspection. The system has two parts: a fixture that positions the camera, and a machine vision workflow that evaluates the images.
Fixture Design
The fixture had to position the camera repeatably across several equipment variants while keeping image geometry and lighting consistent. I designed it in Autodesk Inventor to mount on each equipment geometry and to lock automatically as the camera is rotated onto each inspection position, with adjustable locking sensitivity and play. Controlled lighting keeps shading consistent between images. The camera mount snaps on, the grips carry the cables, and error-proofing features prevent incorrect installation.
Tolerance analysis across the fixture stack verified the fit between moving parts and confirmed that camera position stayed within the required range on every variant.
Manufacturing
The fixture is 3D printed in cleanroom-safe PLA, with heat-set inserts for the fasteners. No available printer could make it in one piece, so I split it into four parts joined by dovetails. Each part was post-processed until the dovetails fit; they hold by friction alone, with no adhesive, and the assembled fixture is rigid with no play.
Vision Model
A large set of images was collected, annotated, and used to train the model. I developed a two-stage machine vision workflow.
Input
Full image
Captured through the fixture.
Stage 1
Object detection
Locates each component in the full image.
Stage 2
Classification
Each detected component is classified as pass or fail based on its alignment.
Separating detection from classification removed unrelated image content from the classification step and made errors easier to diagnose. Borderline cases were evaluated conservatively, because a false pass is the more costly error.
Documentation and Release
I took the fixture from prototype to a released design: engineering drawings, bill of materials, supplier orders and Teamcenter release, so it can be fabricated, procured and maintained.
Result
An automated component alignment inspection system for several equipment variants, combining a repeatable camera fixture with a two-stage machine vision workflow and a digital inspection record.
Before the end of my internship, another process group asked to adapt the fixture design for their own defect detection application.
