Untitled design (5)

What if a furnace was intelligent enough to improve its own performance?

When Mechatherm International Limited was asked about the future of Aluminium furnaces, there were many potential candidates that could have been chosen, such is the growth and innovation within the Aluminium industry at present.  After much internal discussion, it was decided sharing the results of an internal project that looked at the application of Artificial Intelligence to the problem of reducing energy usage, was the best move.

This project was born out of Mechatherm’s Service Division.  Following many successful visits to existing clients to enhance the performance of equipment that had been installed, they believed that more clients should be able to take advantage of the service being provided.  Having previously demonstrated this tuning process could improve a furnaces performance by as much as ten percent but with good engineers being a limited commodity, this service was difficult to scale.

What if the process our engineers were performing could be automated?

Research began into methods that would achieve the goal of automating the service provided as Mechatherm attended the Future Aluminium Forums in both Milan and Warsaw. Delegates discussed large-scale data processing using Artificial Intelligence and Big Data to solve real-world problems.  The concept was becoming a reality, with key players in the field of Industry 4.0 coming together to share ideas.

Mechatherm have installed more than 300 furnaces in 30 countries globally, but to begin developing their idea they would need just one client willing to commit to trial, up until this point, unproven technology.  A UK based client offered a live production machine, agreed to allow access, and give Mechatherm a chance to test their theory.

With an idea and a location, it was time to begin building a team.  Firstly, a specialist software engineer experienced in managing large databases, A.I and Big Data was appointed to further develop the idea, utilising cloud technology to effectively negotiate millions of furnace records.  To support the internal development of the project several consultants were recruited to provide expert knowledge in areas that an experienced hand was needed.

Given time, Commissioning Engineers learn the intricate performance peculiarities and can best advise operators how to maximise a machine’s efficiency.  Combining several observations, operators can be made more aware of how to get the best from a piece equipment, specifically when a furnace is ready to receive or transfer material, effectively reducing the delays associated with normal operation.  In technology terms this would equate to using an offline brain watching and learning how the furnace operates, gauging energy use per cycle and suggesting when the optimum parameters have been met.

A trial system was built and long before it could be implemented on a working furnace it was presented to their partner for approval, it was at this point the first obstacle was encountered.  It was around this time that some high-profile industrial IT sabotage cases were in the news, and as the furnace was to be connected directly to the internet using cloud servers, subsequently support was withdrawn.  The project had to be taken back to the drawing board and an alternative solution built, if an online cloud system could not be used then an offline server would be the only option.  Mechatherm began work on their own A.I engine server at a local level and found that the speed of feedback from querying millions of records to be on par with the cloud services.  The project was back on track and with support reinstated an implementation date was set.

Installed, the system required a few minor adjustments to the furnace control software adding the ‘Big Brother’ concept of the server and advising what action should be taken.  A simple traffic light system was shown on a new screen giving operators an easy visual to make the choice to attend the load.

With the server stored in an air-conditioned room away from the furnace, a monitoring station was set up allowing the Mechatherm team to observe the performance with no interruption to the furnace’s operation – all that was needed now was to press go!

Initial signs from the first couple of weeks indicated the system had a long way to go before it could reliably advise operators when action should be taken, this was the second major setback.  With constant feedback from Mechatherm’s global team of consultants it steadily became clear that with any new system like this, it needed to learn, it needed data, as it simply didn’t know enough to perform the role it was given.  A KPI system was created to measure accuracy and record how intuitive the predictions of an early finish of a charge were based on time as a percentage.  This began at around 50% accuracy.  The team was confident that with time this would improve.

Steadily, over many weeks, the system data grew to tens of millions of records and with the increased volume, predictions held more certainty.  The accuracy edged towards 60% and continued to improve as more and more of the decisions showed a true representation of the furnaces state.  This system was based purely on recorded data, devoid of any need to share details about what had been loaded into the furnace, ruling out any operator inaccuracies and exclusively responding to changes in temperature and output of the combustion system.

It was agreed with Mechatherm’s partner that at 85% accuracy the system recommendations would be followed for the charge and transfer requests, this would allow the calculation of fuel savings compared to standard operator usage.  Finally at 88% the system took over and the last recorded fuel savings were at 7%.  The system was working.

Just as the project was beginning to gain momentum and show positive results the final obstacle was encountered.  The Global pandemic hit and investment and support for industry 4.0 was understandably withdrawn.  The project was temporarily shelved.  As the world begins to re-emerge and fuel savings of 7% on initial trials already proven, Mechatherm continue to develop this technology with the aim to standardise this technology for all of their M and X-Series furnaces from 2023.