15_Oben.jpg

Case Study Precision Tooling - Inspection of milling part

"With deevio's AI software, we have detected even the smallest defects that we could not detect with conventional image processing in this way"


Quality Manager FRANKEN

The Project

 

Emuge Franken produces milled parts for various industries and inspects the quality of these milled parts with specially trained personnel, since even the smallest defects prevent the products from being delivered to customers.

However, due to the complexity of the task and the variability of the products, pure measurement with conventional image processing is not sufficient here to correctly classify the 12 to different defect classes. 


 

The Implementation

 

Instead of replacing the existing inspection system, deevio was able to build on the existing image data, create an AI model, test and validate it, and integrate it into the production process together with FRANKEN S7 Siemens machine integration.

For this purpose, FRANKEN together with deevio sorted the images from the inspection system into the 13 different categories in the first step, with about 50 pellets per category.

 

frankenbohrer2.png
frankenbohrer.png
franken 1.png

Image acquisition and the creation of the AI software

franken2.png
franken3.png

No complex test code is required for the creation of the AI software; sample parts with the defects to be tested are sufficient here. deevio takes pictures of these, marks the defect on the picture together with the automotive supplier and can then carry out the first tests after just a few hours.
Providing and evaluating sample parts is the only thing deevio's customer had to do here.

In doing so, deevio not only shows that a throttle valve is defective, but also where in the image the defect is.
This allows the customer to keep statistics on the defects that occur and improve the production process.

Based on the data labeled in this way, deevio created a multi-classification model that can classify the beads into the various categories. After a short time, the accuracy here was already over 98%. 

Another challenge was the evaluation time: up to 40,000 images had to be evaluated in less than 3 minutes (~0.004s per image). For this, deevio used its high-speed AI box, which was also integrated later. Due to the corona pandemic, the complete project was done remotely by BASF and deevio, including the acceptance into live production.

deevio's AI software is now reliably running live in FRANKEN's production since 2020.

Continues development of the AI model

FRANKEN's goal for the AI project with deevio was also to be able to continuously improve the models and add new products independently. 

To do this, FRANKEN uses deevio's Self-Training feature, which enables them to enhance the AI model with new images from production of new defect classes or product types and increase accuracy without having to create codes themselves.

Before and after

 

Before working with deevio, FRANKEN was only able to manually assign beads, which made it harder to draw conclusions about the causes of defects in production.

With deevio, FRANKEN is now able to access automated categorization with high accuracy, have a detailed basis for drawing conclusions about production and troubleshooting, and FRANKEN can independently add new inspection criteria and products.

How can you benefit?

 

If you perform manual quality inspections or have a camera system you are not satisfied with, feel free to contact us. We will find an individual solution for your application and deliver the complete system fitted to your needs.

Get in touch with us here.