Learn more about the use case:

Tvarit GmbH

Quality assurance


Real-time quality prediction in the production of aluminum coils

What drove the decision to develop an AI-based use case?
The quality tests have so far been carried out at the end of production. Here, a material sample was taken and sent to the laboratory, where about 60 geometric and chemical tests were performed. The tests and their evaluation took about two days. Only after receiving the test results could the coil be classified as "o.k." or "n.o.k.". This interrupted or delayed the supply chain. The percentage of defective parts was about 5%. These had to be reworked with an operational effort of six hours or were distributed with a discount of 60-70%.

What were your expectations or requirements for the AI application?
The client expected real-time monitoring of batch quality and reduction of defective parts.

Which partner did you rely on with which technology for the development?
The implementation of the project including data cleansing, anomaly detection, labeling of data, data harmonization, the creation of the predictive models and the visualization in dashboards for the client was done by our Data Science Department using the modules and algorithms we developed specifically for this use case.

The use of AI solutions must also be accompanied by a change in competences. How have you dealt with this challenge in respect of your employees?
In addition to accurate and reliable classifications and predictions by our AI, the usability and comprehensibility of the outputs is of course of central importance. To ensure that our client can also use our solution effectively, we held two workshops with the decision-makers and employees involved. One workshop at the beginning of the project with the aim of defining the use case in concrete terms and ensuring a clear understanding of it among all those involved. A second workshop was held after the implementation of our solution to ensure that the users of our solution not only understood it, but were also able to use it to its full extent. In this specific case, both workshops involved about 20 employees of the client.

In what timeframe did they implement the AI solution in your company?
The implementation of the project took about six weeks. Due to an insufficient availability of data at the beginning, two runs were necessary to create meaningful and accurate models.

A conclusion to your AI solution:
The implementation of our AI solution for the client significantly reduced the time needed for quality assurance. Instead of waiting two days for the test results, statements about the quality could already be made in real time during production. With a sufficient confidence level of the prediction made by the model, the classification as "o.k." or "n.o.k." could already be made before the final lab results.

In addition, our solution enabled the client to reduce its reject rate by over 75% (5% to 1.2%), resulting in a significant cost reduction. With a model accuracy of 76%, the client is now able to take corrective action in nine out of twelve defective batches (at a cost per coil of around €25,000).

Users of our AI solution also experienced the following benefits:

  • the production department uses our platform to monitor the predicted quality of the batches
  • the development department uses our platform to verify and complement the results of their chemical/physical experiments
  • based on the new insights, management can make informed decisions regarding the supply chain, asset efficiency, and their vision of Six Sigma
  • securing corporate and brand value

One final piece of advice/tip to other entrepreneurs looking to apply AI:
Statistically, about 70% of all Data Science projects fail. To prevent this, we recommend working out and defining the concrete use case and business case in detail together with the partner before the actual kick-off of the project.

More info at: