Content area
Full text
Data is said to be the 'gold of the 21st century'[1]. The reasons for this comparison are the high values and profits derived from data in other branches like the telecommunications and consumer-industry. In the metals branch the usage of production data by disciplines like machine learning or predictive modelling is still in its infancy. One major difference between the metals-sector and the consumer-industry is the availability of data in a cloud-based environment and the openness to cloud-based technologies. By Roger Feist·
In recent years, many applications and impressive studies about artificial intelligence (AI) and machine learning (ML) were presented to the public. Maybe the most impressive one was Google's AlphaGo which was able to win games against human champions[2]. Other AI-applications like speech- or text-recognition are already in use on every smartphone. But also more business related applications are published for the financial sector or in the production industry.
All these applications have something in common; their success is based on big amounts of data. AlphaGo's success is based on neural networks generating millions of different games and gaming-situations for its own training. It was only possible to realise it inside a cloud-infrastructure, where big amounts of memory and computational power from over 1200 CPU could be reserved for some days. The huge computational infrastructure of the web built by companies like Amazon, Google and Microsoft offers flexible usage to private and business customers and relieve their users from IT-administration-tasks. On cloud-platforms powerful tools for data access and analysis are available and can be used under economic conditions.
Obviously many of the successful approaches from other industries and many of the used algorithms are also applicable to tasks in the metals-production but the number of real applications is still small. In spite of the high performance and usability of the cloud based tools they are almost not used in the aluminium industry. One reason may be the strength of this industry in 'classic' ways of data analysis.
Most producers have invested in powerful systems for data collection and analysis on premise. If events occur like strip-breaks or breakdowns the data is used by the maintenance staff to find the root course of the event. Additionally some applications compute 'aggregates' like 'maximum speed' or...