Severstal introduces digital predictive model to prevent failures at mill-2000

Cherepovets Steel Mill, one of the world’s largest integrated steel plants (part of Severstal’s Russian Steel division), has launched the commercial operation of a digital predictive model to prevent failures at the hot rolling mill 2000 and thereby reduce its downtime. This model calculates the probability and risk of the pinion stand bearings overheating, which is one of the most frequent and costly causes of unit shutdown. This is the first predictive maintenance model introduced at the CherMK plant as part of Severstal’s digital transformation strategy.

The model’s forecasting is based on the data stream collected from onsite temperature sensors located at the mill. The digital model developed by the Company’s in-house experts analyses this data and produces a forecast of the temperature regime for the next period of time. An operator receives an immediate notification in any instance where indicators deviate from the set norm. The timeframe of failure forecasting is sufficient enough to enable the operator to take the measures necessary to prevent an unplanned shutdown of the mill.

These models are viewed as a separate unit on the controller’s screen, with an additional option to access it via the browser on a personal computer. In the future, a mobile app will be developed as well as a text message notification service to alert to any abnormalities.

Sergei Dobrodey, Director of repairs of the Severstal Russian Steel division, said: "Previously, the sensors only detected a unit malfunction after it had occurred. We required a forecasting horizon of potential malfunctions in order to avoid breakdowns and stoppages of the most critical equipment. We anticipate that this predictive model will reduce the downtime resulting from pinion stand bearings overheating by 80%. We plan to introduce similar models for other types of failures at mill-2000, as well as in other units.”

Igor Bardintsev, Chief Digital Officer at Severstal, commented: "Over one hundred thousand parameters are controlled by sensors at mill-2000 alone, which gives us considerable opportunity to implement predictive analytics projects. This is the first model we have introduced for repair management as part of a large-scale predictive maintenance programme. This particular model is based on a machine learning algorithm, therefore the more data parameters are processed and compared, the more precise the model becomes. Severstal recently created its own Data Lake which allows us to collect and process more information and implement projects using artificial intelligence."