Camera system for airbag component assembly inspection, for the automotive parts industry

How the brief came about

The brief for this work was to have a camera confirm that the plastic components of the airbag assembly were complete and correctly positioned before going on to the next step. The parts tested covered 4 models, which the documents call Model A to Model D, and each model has a left-hand part and a right-hand part. Plastic pins must be fitted on the workpiece, and some models must also have a slide plate, black tape or an adhesive pad.

The airbag is a safety component. A missing pin or a slightly misaligned adhesive pad is hard to see by eye, especially when there are several models and the left-hand and right-hand parts look similar. The customer therefore wanted a Machine Vision system (a system that uses cameras and software to inspect workpieces in place of the human eye) that inspects every part against the same criteria.

Requirements and constraints

We set out the prototype's requirements in the job's requirements document and the per-workstation requirements. The system must inspect 5 items on workpieces in 4 models, and must send inspection results as electrical signals to the equipment at the workstation.

• Check that plastic pins are fitted at every point, on the workpiece plates of all models
• Check the position of the slide plate on Model A, the black tape on Model C and the adhesive pad on Model D
• Sort good parts from defective parts automatically
• Support both left-hand (LH) and right-hand (RH) parts of every model
• The per-station prototype must receive signals from 2 switches simultaneously and output 24 V on 3 channels: pass, fail and status
• On the black plastic parts (PA66-GF30 and PP-GF30 are grades of glass-fibre-reinforced plastic), round nuts must be distinguished from square nuts, to catch cases where they are fitted in each other's positions

Why an overhead camera, an Area Light and image processing

We placed the camera above the workpiece and used an Area Light because this work needs the edges of the pins, tape and adhesive pads to be clear enough for the program to separate them from the workpiece background. The test rig uses a Cube camera at a distance of 20 cm with an 8 mm lens. Images from the camera go to a Medusa processing unit, which can connect 2 cameras, and an image processing program then locates the points to inspect on each model.

We started with a test rig because we first had to confirm that the camera images showed every inspection point on all models, and how many cameras were needed. The test rig results could therefore tell us which inspection points can share a single camera.

What we built and tested

This prototype has 2 parts: the hardware test rig, and the inspection program that we wrote and tested on that rig.

• Hardware test rig: Cube camera, 8 mm lens, Area Light and Medusa processing unit
• Assembly inspection program covering 5 items for workpieces in 4 models
• Per-workstation prototype program that receives signals from switches and sends inspection results to the workstation as electrical signals
• Inspection program for black plastic parts that distinguishes round nuts from square nuts

Test sequence

We tested in 3 stages: first confirming that every item could be inspected, then testing per station against the signal requirements, and adding a program for the black plastic parts.

• Test inspection of the 5 items on the test rig and summarise the results in a test report
• Test the per-workstation prototype program with good and defective workpieces
• Add an inspection program for PA66-GF30 and PP-GF30 black plastic parts that distinguishes round nuts from square nuts

Details of test results on real workpieces

The test report concludes that the prototype could inspect the assembly of plastic airbag components in 4 models on all 5 items. Pin inspection on Model B and adhesive pad position inspection on Model D use 1 camera.

In per-station prototype testing on the test rig, the system judged good workpieces 100 times and defective workpieces 100 times, correctly every time. These results were measured on the test rig, where camera distance and lighting can be held constant. Use on a production line requires further trials under on-site lighting conditions and work timing.

What we learned

This work offers 3 principles for camera-based assembly inspection.

• Test both good and defective parts, and report the results separately. Correct results on good parts alone do not show that the system can catch defective parts, so this work tested good and defective workpieces 100 times each.
• The test sample set must cover everything that makes images differ, such as workpiece model, left-hand versus right-hand part, and pin colour. In this work, white and blue pins had to be tested separately.
• Define the input and output signals with the workstation from the prototype stage. Inspection results from the camera become usable when the equipment at the workstation can receive the pass, fail and status signals as agreed.

How to apply this approach to your assembly line

If your assembly line must confirm that pins, clips, tape or adhesive pads are all present and correctly positioned, you can use the same approach as this prototype: an overhead camera, lighting that gives even illumination, a point-by-point inspection program, and pass / fail signals to the workstation. What must be adjusted is the camera distance, lens, lighting and number of cameras, according to the size and colour of your workpieces.

• Prepare sample workpieces, both good and defective, for every model, both left-hand and right-hand parts
• Specify clearly which points must be inspected, and whether each point only needs a present-or-absent check or a position check as well
• Agree the input and output signal format with the equipment at the workstation from the start
• Plan a trial on the production line before full-scale use