Smart Sensors and Predictive Maintenance in Steel Production (SUPREME)

Maintenance is essential to ensure the availability, reliability, and cost-effectiveness of technical systems such as the production facilities of Tata Steel. However, the degradation of systems is a dynamic process, influenced by changes in both the system and its environment. The challenge is to achieve just-in-time maintenance to save on maintenance costs (not replacing too early) and increase system availability (not replacing too late). The research project SUPREME addresses the central research question, 'How can advanced sensor technology and modeling of degradation and failure processes be used to develop a predictive maintenance concept for production systems?'

In early March 2023, we spoke with Prof. Dr. Ir. Tiedo Tinga, who, along with Prof. Dr. Paul Havinga from the University of Twente (UT), supervised this research project. Tiedo is a professor of Dynamics-based Maintenance in the Faculty of Engineering Technology, with a background in Materials Science and Mechanical Engineering. His research focuses on detecting and predicting failures in systems, aiming to develop smart maintenance concepts such as Predictive Maintenance. This is achieved by combining Physics of Failure, a thorough understanding of dynamic system behavior, advanced (condition) monitoring techniques, with data science, and artificial intelligence. Tiedo also works as a professor at the Netherlands Defense Academy. Three junior researchers were active in the project: Ph.D. candidates Melissa Schwarz and Sabari Anbalagan, and postdoc Alireza Alemi.

Data Collection with Sensors

To reduce downtime and associated costs in a production facility, the predictability of failures is crucial. This can be achieved by firstly collecting relevant information about operational conditions and secondly by processing this data correctly to obtain accurate estimates of the remaining time until a failure. In this project, the first challenge was addressed by developing advanced wireless sensor networks, allowing flexible data collection. “By recognizing patterns in the production process that could disrupt the wireless data transfer, we learn here on the job and can further optimize the sensor network to achieve maximum data transfer reliability.”

Utilizing the Data

The second challenge was addressed by developing physics-based models for the failure of the most critical components in the system. By feeding the models with measured variations in operational settings, the time until failures can be predicted, and the appropriate maintenance tasks can be planned. In contrast to the usual data analysis approach, heavily reliant on training with historical data, a physics-based approach has the advantage of predicting even previously unobserved failures. A data-driven algorithm might not recognize such a new failure due to a lack of training data.


In addition to the University of Twente, active partners in the project include Tata Steel, IJssel Technologie, and Samotics. Other contributors include M2i and SKF. “With Tata, we were supposed to start with the HIsarna steel production pilot plant as a case study. However, this plant was (temporarily) shut down during the research, so we worked with Tata to find another case study to focus on. We then investigated the galvanizing line, where we focused on the bearings that need replacement every 2 to 3 weeks on average. When replacing after 2 weeks, it sometimes turns out that a roller could have lasted a bit longer, so it was not optimally utilized. There might also be production missed because this replacement could have been done later. But you can also be too late with replacement, then you have to stop the process and may even discard product that doesn't have the right quality due to the bearing failure. There is a good reason to try to predict when the bearings need to be replaced, and we have developed a suitable model for this in this project.”


“The interaction between science and practical application is always a challenge, especially the translation from the R&D department to the business units. Therefore, our researchers have often been physically present in the project to build a good relationship with the people. We are now reaping the benefits because there is still excellent interaction and exchange of knowledge, information, and data. Even now that the project has ended, the consortium continues to collaborate to exchange knowledge and take further steps.”

What's Next?

Organizing smart maintenance is not only the subject of research in the steel industry but also in several other ongoing studies, Tiedo points out. Models and methods are also being developed at his other workplace - Defense - for dynamically scheduling maintenance for naval ships, helicopters, and vehicles. “Within Tata Steel, this practical application is now in practice. Tata also has other galvanizing lines, so we are exploring how we can apply this model there as well. But in a feasibility study (Take-off call) next half-year, we will also see if we can apply this more broadly in other sectors of the industry where rolls with bearings are used, such as in paper and film. So, it's not just about (scientific) research but also about actually applying the results in practice, which doesn't happen much in our projects. We are quite proud of that.”

Gerelateerde projects

Bekijk hieronder de Projecten gerelateerd aan het thema Smart Sensors and Predictive Maintenance in Steel Production (SUPREME)