Düsseldorf, July 24, 2017

Prediction models for continuous casting processes

Steelmaking is an engineering discipline of its own. Enormous forces have to be dealt with, and extreme temperatures and temperature gradients have to be controlled. Each individual process step presents steel producers with the challenge to keep the highly sensitive processes as stable as possible under the influence of numerous adverse conditions. One such process is continuous casting: at more than 1,500 degrees Celsius, the liquid steel flows through a submerged nozzle into a mold where it starts to solidify. Then the partly solidified strand is withdrawn downwards, supported by the strand guide system. As the process continues, the strand is first bent and then straightened. This technology is to provide quality products capable of meeting most exacting requirements such as those specified by the automotive industry. It is actually possible to repair defects in the continuously cast slabs, however, at the expense of yield, time and money.

Therefore a team of SMS group experts is constantly striving to develop new ideas and approaches to further improve the technology of continuous casting. The team has lately taken an entirely new approach to finding enhanced solutions to avoid crack formation during continuous casting. There is already much knowledge available about what may cause such defects and what mechanisms apply. But the SMS group engineers wanted to go deeper into this subject and gather new knowledge that would make it possible to develop even more refined prediction models and detect new causes and mechanisms of crack formation. To this end, they decided to launch the SMS Data Challenge and call on experts in data processing to provide suggestions and new ideas to tackle this problem. They addressed Data Scientists active as specialists in data processing and data evaluation targeted at the development of predictive models.

Exciting tasks for data experts

SMS digital formulated two tasks for the Data Scientists:

  • Developing an algorithm that would be able to predict the initiation of longitudinal cracking of the strand in the mold with greater precision and reliability.
  • A data-based analysis of the mechanisms triggering the formation of edge cracking.

The contest was first announced at meet-ups in Berlin and Cologne and on a microsite. From there the word was spread in the social media. The first Data Challenge launched by SMS digital and SMS group reached more than 100 data experts from all over the world. The participants represented very different disciplines. Those interested in the contest included engineers, statistics experts, biologists and even medical doctors. None of them had had no previous experience with or knowledge about steelmaking.

We are highly pleased with the great response our Data Challenge found among data experts. We met extremely interesting people who on their part found it exciting to work with data from the steel industry. For many of them that was actually the main reason to participate. Maximilian Wagner, CEO of SMS Digital
Participants submitted convincing proposals

For the SMS group engineers, this new approach worked out very well: “The idea was to gather – in a short time – a maximum of new proposals for solutions which we had never thought of before,” explains Dirk Lieftucht, Manager Component Development within the R&D department. “The participants’ task was to reach clearly defined targets, namely to develop a new algorithm that would be able to predict longitudinal cracking better than the existing ones and to provide new knowledge as to the mechanisms that initiate edge cracking,” adds Esra Erdem-Hornauer, engineer working in R&D in the field of process control in continuous casting.

Five participants/participating teams had been able to provide substantial results by the end of the project period of six weeks. After submitting their proposed solutions in writing, they had to present the results to a jury and answer questions asked by the jury members.

The most convincing solution was put forward by Florian Borchert, aged 26, from Berlin. He won the first prize not only for the technical sophistication of the Conditional Random Fields (CRF) algorithm he applied, but also for the practical usefulness of his proposed concept: “The chances that this solution will find its way to practical implementation are very high,” says Markus Reifferscheid, Head of R&D at SMS group. The winner of the second prize was Thanish Batcha, aged 25, from Chennai in India. For both of them, participating in the SMS Data Challenge was a very exciting experience. Borchert found it extremely fascinating to work with real data. “I hope that once my solution has been put into practical use it will lead to a significant increase in the yield of continuous casting plants,” he comments. Key motivation for Batcha to participate in the contest had been the fact that it gave him the chance to work in an entirely different industry. At home in Chennai, he works in a business consulting and IT services agency.