The manufacturing sector has been undergoing significant automation changes, but the fundamental machinery, like the one used for grinding steel ball bearings, remains largely the same as it was over a century ago. While processes have been automated, one crucial area still in human hands is identifying and resolving errors during production. This task may soon be handed over to AI technology.
In the Schaeffler factory in Hamburg, the production begins with steel wire, which is transformed into rough balls that undergo processing in a series of furnaces and precision grinders. This meticulous process results in highly precise components essential to a range of industries, from automotive to machinery.
Maintaining such precision requires rigorous testing, yet identifying defects can be complex. Problems may arise anywhere in the assembly line, making it challenging to determine the actual cause—like an improperly calibrated tool or a faulty grinding wheel. Traditionally, this requires humans to sift through data from various machines and systems that were never designed for seamless analysis.
To enhance this process, Schaeffler has started using Microsoft’s Factory Operations Agent, an AI-powered tool developed to assist manufacturers. This chatbot-like system employs large language models to find potential causes of defects, equipment downtime, or high energy consumption, functioning somewhat like ChatGPT but tailored for industrial needs.
According to Kathleen Mitford of Microsoft, this agent acts as a reasoning tool that can interpret user inquiries and provide precise answers based on standardized data models. For example, if a worker queries the system about an unusual spike in defects, the AI can correlate information across the manufacturing processes to provide actionable insights.
The integration of this AI tool with Microsoft’s existing enterprise systems, particularly the Microsoft Fabric data analytics platform, amplifies its value. Schaeffler operates numerous plants worldwide, allowing the Factory Operations Agent to be trained on a vast array of data.
Schaeffler’s IT vice president, Stefan Soutschek, noted that the real advantage lies not just in the AI bot but in how it combines operational technology with backend data analysis. Although it currently functions mainly as a data access tool, users can execute basic commands, but the AI doesn’t possess the autonomy to make independent decisions.
This integration is particularly crucial in bridging the existing IT/OT (Information Technology/Operational Technology) gap within manufacturing, enabling staff to pose troubleshooting questions conversationally.
Expected to exit its preview phase later this year, the Factory Operations Agent stands in a competitive landscape as various tech companies aim to capitalize on the advancements in large language models. Companies like Google are also developing tools to unlock data from industrial devices, positioning themselves to cater to manufacturers seeking to modernize their operations.
While increased usage of AI in manufacturing brings opportunities, it also raises safety concerns. Systems that analyze data, as is the case with the Factory Operations Agent, pose risks if users do not recognize when the AI may be failing or troubleshooting inadequately. Duncan Eddy, director at the Stanford Center for AI Safety, emphasizes the unpredictable nature of these AI systems, underlining the importance of careful oversight and intervention capabilities.
