SKF's Automated Machine Learning (AutoML) offering SKF Enlight AI applies self-learning algorithms to real-time process data to detect anomalies and predict impending equipment failures. It combines machine process data with information from vibration and condition monitoring. With the results obtained in this way, maintenance teams can be warned in time and provided with all necessary information from the machine. In this way, users avoid machine breakdowns and reduce their costs.
Automated Machine Learning aims to detect irregular data patterns that indicate impending machine failures. This involves processing very large amounts of data and selecting the optimal algorithms to analyse the data streams. This allows for faster modelling and higher accuracy than a traditional AI methodology. This gives maintenance teams the time and information they need for maintenance planning and diagnostics before machine failure occurs, through earlier fault alerts and the transmission of all important data.
"But the exciting thing about this development is that we can combine vibration and process data from equipment," says Eitan Vesely, SKF AI Offering Manager. "When we look at the insights gained and what they mean for practical use, we realise that the result is greater than the sum of its parts."
AutoML promotes new business models
Viewed in isolation, AutoML-based condition-based maintenance is a powerful tool for predicting faults. It provides comprehensive insight into asset health at the sensor, asset and operational levels. But its relevance to offerings such as REP (Rotating equipment performance) is even greater. With its AutoML solution - SKF Enlight AI - SKF can better deliver on this results-based business model. Here, the customer pays a fixed cost rate for a combination offering that can include, for example, monitoring and optimisation of bearings, sensors, lubrication, seals or reconditioning.
Using the data gained with SKF Enlight AI as a basis for decision-making, the customer can work with SKF to improve the performance of their equipment. "With the insights AutoML gives us, we can plan maintenance and optimise spare parts inventories together with the customer," says Vesely. "Being able to avoid unplanned downtime is a key benefit for everyone involved." SKF Enlight AI demonstrates what it can do in a real-world application in the pulp and paper industry.
SKF Enlight AI in action
A major pulp and paper manufacturer in Latin America used SKF Enlight AI in a pilot project with two pumps in the pre-bleaching system. Recurring failures in these components caused production bottlenecks that cost the company hundreds of thousands of dollars in one year. Due to interactions between the pumps, each failure brought the entire pre-bleaching system to a standstill. The paper manufacturer wanted to significantly reduce the number of unplanned shutdowns and lower the associated costs. The company was looking for a user-friendly solution that would provide early warning of malfunctions and output accurate process data from the equipment - like SKF Enlight AI.
Typically, AI systems apply self-learning algorithms to real-time sensor data to detect impending equipment failures. Here, however, the case was different: to test the performance of the SKF solution, the customer wanted to use it to evaluate process data from past failures. On one pump, technicians had detected an oil leak on 26 December. It was classified as non-critical and a planned shutdown for repair was scheduled for the end of January. The second pump unexpectedly failed on 31 December due to bearing damage - two days after vibration analyses had reported the first signs of an impending malfunction. These two shutdowns cost the company $400,000 and disrupted operations for several weeks.
Data analysis with SKF Enlight AI revealed that this situation could have been mitigated by using condition-based maintenance based on ML. Both pumps had been showing anomalies in system behaviour since mid-December. The maintenance team could have scheduled the repair of the first pump earlier and fixed the problems on both pumps during a single scheduled shutdown. This would have saved the company $250,000.
"The evaluation included an analysis of vibration and process data. The result was clear: using both data sources, more faults could be predicted than would have been possible with a single data source," explains Eitan Vesely. "Thanks to the successful pilot project, we now equip three complete plants of the pulp and paper manufacturer with SKF Enlight AI and monitor several hundred different systems with it."