It’s loud and above all it’s hot in the Kempten iron foundry. The metal, up to six tons of iron and scrap parts, bubbles in ceramic crucibles at 1500 degrees until it can be poured from cast iron ladles into the mold boxes, spraying sparks. This is anything but easy, there are many steps in between, many parameters influence whether and how the casting succeeds. With their many years of experience, the workers know what to do. But wouldn’t it also be good to record and evaluate these parameters, i.e. data? The material is expensive, but especially the energy. A melting furnace uses about as much electricity as 3,200 four-person households. Is it possible to save something there?
A lot of energy is consumed here: Production in the Kempten iron foundry.
(Photo: Kemptener Eisengiesserei/Kemptener Eisengiesserei)
Jean-Pierre Hacquin, IT manager at the Allgäu medium-sized company, thought so too. So he set up a partnership with the local university. Florian Huber, a research associate in the field of data science at the university, took a look at some areas of the company together with the process experts. Like the Crucible. They only last for a certain amount of time, then cracks develop and there is a risk that the 1,500-degree hot melt will escape. On the other hand, you don’t want to discard the expensive crucible too early. So when is it time for that? In Kempten they can now calculate this in advance with a program – the ancient cultural technique of iron smelting combined with high-tech. The software was fed with data obtained from repeated measurements on a crucible. With the help of artificial intelligence (AI) algorithms, it reliably indicates when a pan needs to be replaced.
As in Kempten, many companies are currently trying to optimize their processes with the help of AI software. That would have made sense in normal times. But now, with energy prices skyrocketing, there’s even more pressure behind it. From a topic of the future, AI is becoming a tool that provides very specific support in solving problems. However, many companies are still in the early stages, and Peter Herweck knows that this often has to do with data. Herweck is head of the London-based software company Aveva, previously he worked for Siemens and Schneider Electric. The manager is a rather sober person, not someone who throws around buzzwords like artificial intelligence (AI) or machine learning. Only when it comes to all the things that can actually be done with data does he get enthusiastic, at least a little bit. Data is “like a gold mine,” he says, although there is usually enough information available. “But most of it is unused.” He wants to change that.
The art is in making the data usable. The problem: The data that the machines and production systems generate is usually not available in a standardized format. Every manufacturer cooks its own soup, and some systems, especially older ones, have to be retrofitted with sensors in the first place. Herweck’s company has been represented in production plants all over the world for decades, there are currently around 55,000. One focus is the chemical industry, one that is currently the subject of much talk. It consumes large amounts of raw materials and energy and is under pressure from two sides: On the one hand, there are the energy prices, which are rising extremely. On the other hand, the human task of climate change, which makes it necessary to reduce CO₂ emissions and to use limited resources more sparingly.
A single windmill has 1000 sensors
But why don’t you just take a look at the adjustment screws yourself, turn down the controls? If only it were that simple. Many processes, from heating control in a large apartment building to ventilation in a swimming pool to large-scale plants in the chemical industry, are so complex and have such a large number of adjustment screws that it is only possible to optimize them in a beneficial way with the help of AI. The starting point is always the data produced by machines and systems. A common wind turbine alone today contains around 1000 of these sensors. Companies use a wide variety of machines, so the data must first be processed in such a way that they can also be evaluated together. Aveva has developed a kind of data translation program for this purpose.
Because there is enormous potential slumbering in the data. They make it possible to build a digital image of the production plant, a so-called digital twin. The level of detail is so great that, for example, things such as the historical data on the behavior of individual pumps are included, and the power supply must also be reproduced exactly. At some point, Aveva boss Herweck is convinced, you will be able to watch the huge systems down to the last valve in real time from your desk with virtual reality glasses.
It’s not quite that far yet, but in a project with the chemical company Covestro, Aveva tapped into a total of 4,000 sensors in a production plant and, after evaluating the data, discovered some processes that were far too ineffective. According to Aveva, after improvements were made, CO₂ emissions fell by almost 40 percent. Data expert Huber has already experienced something similar: “Even proven process experts are surprised when we ask, based on our findings from the data, whether they have already considered this or that.”
Is the heating working properly?
Artificial intelligence also helps to save energy in the housing industry. The energy service provider Techem, based in Eschborn, connects heating systems to a control center there and thus finds out promptly when a system is working inefficiently. To do this, however, the sensors in the systems must be wirelessly connected to the Internet. The data is then forwarded to the headquarters in Eschborn via secure connections.
So although there are already several examples in which artificial intelligence is integrated directly into the production process – this is not the norm. Many companies use software that can detect when it’s time to replace a tool due to wear and tear, called predictive maintenance, “but there’s still a lot of room for improvement,” says Frieder Heieck, head of the Sonthofen Technology Transfer Center, which belongs to Kempten University . “The model is only gradually establishing itself.” But the interest is great, he notes again and again, driven by buzzwords like AI, even if the real projects are often just statistics. Real AI is needed, for example, when it comes to finding the one from various measurements that is relevant for the application.
The most commonly used AI application in companies is still speech recognition, according to a study by the Leibniz Center for European Economic Research. This is followed by Robotic Process Automation (RPA), i.e. software that relieves employees of routine jobs, for example transferring address data from one form to another. On the other hand, AI is rarely used to control the movement of machines, and even more rarely in marketing or management.
There are many reasons why AI is in its infancy in German companies. Many people in charge of companies are still unsure whether and how AI could actually help them. Specialists who are familiar with it are rare, and they are also expensive. This indecisiveness could take revenge, believes Dirk Pothen, board member at the IT service provider Adesso, one of the largest in Germany with around 5,300 employees. “If you don’t have enough AI professionals, you will find it difficult to benefit from the technology.” While the race for AI is only just beginning, the manufacturing industry should not have a false sense of security. Now decide who can use AI to set themselves apart from the competition in the coming years.