Case Study Automitive - Inspection of turned parts

"By working together with deevio, we were able to automate a previously manual process from start to finish."

CEO, Parts Control

The Project

Deevio is working together with the company Parts Control from Bavaria. Before working together, the turned parts visible in the pictures below have been inspected manually. The challenge to this project was that the turned parts exist in different variations, are inspected from all sides, and even the smallest defects on the approx. 15 mm long turned parts must be found.

Three types of turned parts

In addition to measuring the turned parts, the following surface defects, among others, must be identified as defects:

Pressure marks
Chamfer too large or too small
Dirt in the bore
Chips in the bore
Material defect

The goal in this project was not only to develop a suitable AI model for defect detection, but also to design a complete and customised vision system and integrate it into the inspection process.

deevio's approach to the inspection system

Deevio has developed an inspection system that can inspect the turned parts from all sides and also detect surface defects. For this purpose, the turned parts are first measured by a camera on a conveyor belt and then transported to the inspection cell.

Here, images are taken from all visible sides (from the side and above). If a defect is detected in this step, the part is immediately mechanically rejected. Then the part is mechanically rotated and an image is taken of the side not yet taken and, if defective, then sorted out. The design and functionality of the inspection system made it possible to collect images of the right quality to develop an AI model.

Inspektionssytem von Innen

Drehteil von oben

deevio's approach to the AI model

Compared to traditional machine vision, where rigid rules are defined on the basis of images and partial areas of the images are checked, the development of the software with AI is different. AI uses sample images of real products and learns from them independently what constitutes a good part and what constitutes a defective part. In the process, AI also learns to deal with the variability of products, guaranteeing high accuracy in detection while keeping pseudo-rejects low. This makes it particularly suitable for complex inspections and thus expands the spectrum of machine vision.

For the training process, 100 sample parts per defect type - i.e. 100 with pressure marks, parts with chips in the bore, etc. - were collected.

Pictures are taken of each sample of each defect category, then defects are marked on the picture and thus the AI is shown which defect type is present. Based on this data, an AI model is developed. After that, the AI model is ready and can be tested with new images not seen before. The process is much faster than with traditional machine vision, which also means that it is possible to react to changes more quickly, since no complex programming is required.

In addition, the AI can also determine where exactly the defect is on the part. This enables the customer to keep statistics on the defects that have occurred and improve the production process.

Erkennung einer Druckstelle

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.