Fiber process foreign matter and homogeneity detection

ProductionLine Fiber Cutting
ProductionLine Raw Fiber
ProductionLine-1
Production Line Camera System
Production Line Camera Cabinet
Production Line Monitoring
Production Error Samples
Development Filter Selection Section
UWP Application
Kaizala integration
Application Control Panel PLC Connections And Filter Selection
Application Image Upload and Train Panel
Azure ML Experiment
Microsoft Cognitive Services Panel-1
Microsoft Cognitive Services Panel-2 Performance

Inspiration

The lack of homogeneous distribution of foreign substances and fiber that occurs during automotive ceiling production is at a level that the operators cannot distinguish by eye and it is very difficult for the operators to control because it is a very fast process. Due to these errors, quality and scrap issues in production are at a very high level, so using Microsoft Cognitive and AI services to develop solutions has been our inspiration.

What it does

The developed application detects real-time errors and gives a warning signal to the operators with the integration of the production line automation PLC and the raw material feeding unit is automatically stopped. In addition, with the integration of Microsoft Kazila, a notification is sent to the Production Quality Team. With all these operations, the lines are removed before the production is completed and the quality reaches high levels.

How we built it

We used Microsoft Machine Learning services and Microsoft Cognitive services. We developed the application with UWP C #.

Challenges we ran into

Since the production process continued uninterruptedly, the control had to be made at the time of production. We found the real-time classification process to be quite difficult, and we did a lot of testing about which techniques we should choose from sample image collection and labeling. Our camera selection and filter detection was a challenging process, but Microsoft Cognitive and AI services helped us a lot in overcoming this challenging process.

Accomplishments that we're proud of

In 2019 and 2020, we learned that there is a total scrap cost of 48900 to 52000 € per year due to foreign matter and non-homogeneous dispersion problems. In the first 3 months of 2020, the scrap cost was 14000 €. With the application we developed with Microsoft Cognitive and AI services, we are proud of this benefit that we have achieved a total scrap cost of 200 € in the last 2 months.

What we learned

We learned a lot of new information about Cognitive and AI services, we agreed with the team about how useful and fast they are in processes such as the rapid creation of models and taking them into operation and we decided to do all our projects on these platforms. In addition, we created training teams to expand the use of these tools for the team.

What's next for Fiber process foreign matter and homogeneity detection

We received many project requests as the project is a source of inspiration for the expansion of the project to other production lines and other locations and for our other AI Projects. Therefore, we have created an AI team in our team, thanks to Micrsooft tools, we will quickly activate our projects.