Optimal kiln utilization with AI

It is cylindrical, about four meters high and one meter wide, and it is located at the Heraeus Comvance production site in Bitterfeld, Germany: the special electric kiln for the production of quartz glass. Its four levels must be loaded with cylindrical molds of various sizes filled with quartz granules. At around 2,000 degrees Celsius, this granulate melts and quartz glass is produced. Sounds simple enough. But the key question with this oven is how to fill the shelves in the best possible way to process orders optimally. "When we first thought about using AI in 2023, the order intake was already well above the maximum capacity of the furnace. So the managers were happy about every extra part that could fit in the oven. By using AI, we were able to optimize the loading in such a way that the capacity limit was increased immensely. We are now able to process 50% more material and are able to handle the order situation much more efficiently. We have exceeded all our initial expectations," reports Christian Piazzi, Program Manager AI.  

Previously, the filling process was entirely manual: highly experienced employees packed the oven four times a week, or about 20 times a month, based on the order quantities received and confirmed delivery dates. These experts were always under pressure due to the high demand. Because of the high demand, there was no time to research the best way to fill the oven; instead, they had to rely entirely on their years of experience.

At first glance, "packing small things into a big thing" did not appear to be a major challenge for AI programming. "However, our obligatory literature search quickly revealed that there was hardly any literature or algorithms for small cylinders in big cylinders," explains Piazzi. "We had to start from scratch." A data scientist from the Digital Hub dug deep into the subject and tested many different options. It paid off: the result is a more than 50% improvement in volume utilization, which translates into about 1.3 million euros in higher annual product output and fewer oven runs. This in turn translates into 120 MWh of energy savings and 114 tons less CO2 emissions per year. 

Even AI expert Piazzi did not expect this: "Even though the figures are based on a simulation verified by experts, they surprised us in a very positive way, especially the increase in output of 1.3 million euros." Now it is time to act, because the figures show that the necessary oven conversion is more than worthwhile. For this purpose, a skeleton is integrated into the oven, with the help of which the four levels rest on the edge and can no longer tilt due to the placement of smaller and larger cylinders. The AI, which has been trained on eight different cylinder geometries, is then ready to go. The human simply enters how many cylinders of each size are to be produced this month, and the AI visualizes the 20 optimal fillings on the computer. The filling itself, however, is still done manually. At the end of the year, Heraeus Comvance will commission a second kiln, which, thanks to AI, will also achieve higher output.  

Using AI as a packing master is an excellent idea. This was also demonstrated at the internal Heraeus Awards 2023. The project won the "Focus Award" category, which recognizes rapid adaptation to market challenges.  Heraeus is also using the new tool for departments in other operating companies. According to Piazzi, it doesn't always have to be the cylinder-in-cylinder challenge: "We have several other potential applications within our group for which we are currently investigating the use of our packaging AI."