The basic machine For grinding a steel ball bearing, about 1900 has been the same, but manufacturers are steadily automating everything around it. Nowadays the process is powered by a conveyor belt and is for the most part automatic. The most urgent task for people is to find out when things go wrong – and even that can soon be transferred to AI.
The Schaeffler factory in Hamburg starts with steel wire that is cut and pressed into rough balls. Those balls are hardened in a series of ovens and then go through three increasingly precise sharpening machines until they are spherical within a tenth of a micron. The result is one of the most versatile components in modern industry, making joints with low friction possible, from lathes to car engines.
That level of precision requires constant testing – but when defects pop up, it can be tracing a puzzle. Testing can be a defect at a certain moment on the assembly line, but the cause may not be clear. Perhaps the torque is switched off on a screw tool, or a new replacement grinding wheel has influenced the quality. Tracing the problem means comparing data over multiple pieces of industrial equipment, which has not designed any of this in mind.
This can also be a task for machines soon. Last year, Schaeffler became one of the first users of Microsoft's factory agent, a new product powered by large language models and specially designed for manufacturers. The Chatbot style tool can help detect the causes of defects, downtime or surplus energy consumption. The result is something like chatgpt for factories, where the models of OpenAI are used on the backend thanks to the company's collaboration with Microsoft's Azure.
Kathleen Mitford, the vice -president of Microsoft for global industrial marketing, describes the project as “a reasoning agent who operates on top of production data.” As a result, Mitford says: “The agent can understand questions and translate them with precision and accuracy against standardized data models.” So a factory worker can ask a question if “what causes a higher than normal level of defects?” And the model could answer with data from the entire production process.
The agent is deeply integrated into the existing Enterprise products from Microsoft, in particular Microsoft Fabric, the data analysis system. This means that Schaeffler, which runs hundreds of plants on the Microsoft system, can train his agent on data from all over the world.
Stefan Soutschek, the vice -president of Schaeffler who is in charge, says that the scope of data analysis is the real power of the system. “The big advantage is not the chatbot itself, although it helps,” he says. “It's the combination of this OT [operational technology] DATAplatform in the backend and the chatbot that relies on that data. “
Despite the name this is not an agentic AI: it has no goals and the powers are limited to answering all questions that the user asks. You can set the agent to carry out basic assignments via the Microsoft Copilot Studio, but the goal is not to have the agent make his own decisions. This is mainly AI as a tool for data access.
