From 20 % to 99 %: How We Improved Car Parts QR Code Recognition Rates

The article presents a real-life case study of how to boost the QR code recognition rate of automotive parts from 20 % to 99 %, covering three areas--hardware upgrades, algorithm optimization, and process adjustments--to solve the common problems of QR code dirt, glare, and blurring. It is suitable for reference by production management and technical teams.

Why can't the QR codes on the car parts be scanned?

We've encountered this kind of problem on the production line: The QR codes on the parts are either reflective or covered in oil, so the scanner can't read them no matter how hard it tries.At first the recognition rate was only 20 %, which meant that we wasted a lot of time going back over the work.Later, the company discovered that the problem wasn't just in the equipment. Lighting conditions, the angle of the parts, and even the habits of the employees could all cause the scanning to fail.

Hardware upgrade: First, give QR codes a "high-definition stage.

The first step in improving recognition rates is to make the QR code more clearly readable.

Select the right industrial camera and light source.

The flash of a camera phone is not bright enough to illuminate the metal parts of a watch.We switched to an industrial camera with a polarizing filter and a ring of cold lights, and even if the surface of the part is covered in a film of oil, we can still photograph the QR code clearly, as if it had just been printed.

The fixed supports are more reliable than human hands.

Testing showed that if there was a 2-centimeter deviation in the position of the barcode, the accuracy of the identification dropped by 30 %.Later, we designed a magnetic base with an adjustable stand to fix the scanning distance and angle, doubling the stability.

Optimizing the algorithm: Teaching machines how to "guess" QR codes.

Once the hardware is in place, the software has to keep up.

Multiple algorithms solve the problem of dirt.

With traditional algorithms, when a QR code is damaged or dirty, it's a write-off.We combined image restoration, edge enhancement, and deep learning to make the system able to read data even if 15 % of the pattern is missing.

The dynamic threshold is used to address the problem of reflections.

The reflected intensity of light can vary by as much as three times over a distance of 10 centimeters along a conveyor belt.We added a real-time brightness detection function, like giving the camera a set of photochromatic sunglasses, so that no matter how strong the light, it can be suppressed instantly.

Streamlining the process: Making scanning a habit

This last punch raised the accuracy rate to 99 %.

Finding the best angle for the part.

After 200 tests, it was found that QR codes placed in a recess on the side of a part would suffer 80 % less interference from ambient light than if placed on a flat surface.Today, before a new mold is put into production, it must first undergo a "QR code location review.

Implement a system of daily reports on the quality of the scanning.

Every day, they would tally up the number of failures at each workstation and print out blurry QR code images to post in the workshop to help everyone look for patterns.Last month, an old technician discovered that a certain angle of tilt could prevent the residue of cutting fluid from getting into the machine, which helped us increase the recognition rate by another 1.2 %.

After the plan was implemented for half a year, the return rate dropped 70 %. Even suppliers came to take notes.In fact, there's no secret technology to improving recognition rates. The key is to tightly mesh the three gears of personnel, equipment, and procedures.If you ever run into a similar problem, why not take 20 blurry pictures of the QR code with your mobile phone and send them to the manufacturer?