Face and voice recognition, cancer detection algorithms, autonomous cars and many other technological innovations have made artificial intelligence (AI) a hot topic. Erik Hallin, Lead Data Scientist at Uddeholm AB, visited Karlstad University on 23 May to present how he uses machine learning, one of the key components of AI, to ensure predictable maintenance in the company’s steel production line.
Machine learning involves building algorithm-based models that learn from data without receiving explicit instructions. In his lecture, Erik Hallin described the logic of some fundamental algorithms and the advantages of machine learning, something which Uddeholm has used since 2017.
“Machine learning is a technique we really believe in. It helps us to keep costs down because we can quickly predict when something is about to break in the production line. This gives us a head start in addressing the problem before it causes a stop,” says Erik Hallin.
Production quality in focus
Uddeholm includes data from all production stages. The most modern ESR furnace has 75 000 sensors. Older machines and furnaces also have sensors, but much fewer, because these must be monitored manually. The sensors generate a large amount of data which Erik and his colleagues analyse to monitor production quality.
There are several reasons why AI is currently such a hot topic. Open Source platforms, which provide good tools that can be downloaded free of charge, and affordable sensors and the good storage capacity many of them have are contributing factors, according to Erik. So what is the greatest challenge of using machine learning in industry?
“Technically, it is creating a sustainable system that can still be managed easily. We have many machines that have been used for many years. They still work well, but are not equipped with the same number of modern sensors that new machines have. We would ideally like to have a data collection system based on the same type of software, but this will take time because the machines are only replaced when they cannot keep up production. Another challenge is finding staff with the right skills. Working with machine learning is still new in Sweden, and many do not yet have the expertise.”
