Systematizing Data Preparation in Smart Manufacturing via Axiomatic Design: A Toolkit Integrating GUI and Agentic AI
This paper proposes an Axiomatic Design-based, open-source toolkit that integrates a graphical user interface and agentic AI to systematically transform fragmented, manual data preparation in smart manufacturing into a reproducible, scalable scientific discipline, thereby enabling advanced downstream applications like digital twins.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the factories of the future, machines are no longer just tools; they are constant generators of information. Sensors attached to drills, grinders, and lathes capture a relentless stream of numbers, recording the force of a cut, the vibration of a tool, or the texture of a surface. This flood of data is the lifeblood of "smart manufacturing," a vision where machines monitor themselves, predict when they will break, and adjust their own operations to maintain perfection. The promise is a world where production is seamless, efficient, and self-correcting. However, a significant barrier stands between this vision and reality. The raw numbers pouring off the factory floor are rarely ready to be used. They arrive in messy, inconsistent formats, often buried under layers of irrelevant technical notes, split across incompatible file types, or mixed with noise that obscures the actual signal. Before this data can teach a computer how to make a better part, it must be cleaned, organized, and shaped. This process, often dismissed as mere housekeeping, is currently a chaotic, manual struggle that consumes the majority of an engineer's time and effort.
A team of researchers from institutions in Japan and Germany has tackled this bottleneck by treating data preparation not as a series of isolated chores, but as a rigorous scientific discipline. They observed that in many manufacturing experiments, from micro-drilling titanium to grinding glass, the same tedious problems kept appearing: files that were too large for standard software to open, numbers written with commas instead of dots, and signals that needed to be cut into specific pieces to be useful. Rather than writing a new, custom script for every single problem, the researchers applied a design philosophy known as axiomatic design. This approach insists that a system should be built so that changing one part does not accidentally break another. Using this principle, they created a flexible, open-source toolkit that transforms raw, chaotic manufacturing data into clean, ready-to-use information.
The toolkit they built offers two distinct ways to work, catering to different needs and skill levels. The first is a graphical interface, a visual dashboard where a human operator can see the data, click buttons to clean it, and draw lines on a graph to select specific sections. This allows for careful, step-by-step control, ensuring that the human expert can verify exactly what is happening to the numbers. The second approach is more autonomous, powered by an artificial intelligence agent. In this mode, a user simply types a request in plain English, such as "find the highest force in this file" or "remove the first three lines of text." The AI understands the intent, figures out which tools are needed, and executes the complex sequence of steps automatically. Crucially, the AI does not see the raw data itself; it only directs the local software to perform the work, keeping sensitive manufacturing information secure on the user's own computer while still benefiting from the AI's planning capabilities.
To prove their system works, the researchers tested it on a challenging real-world scenario: a micro-drilling experiment involving titanium. In this test, a single tool drilled hundreds of tiny holes, generating a massive file of sensor data for every cycle. The raw file was over 100 megabytes, filled with metadata headers and mixed formatting that would typically freeze standard spreadsheet programs. Using their new toolkit, the researchers first used the visual interface to strip away the unnecessary text, convert the number formats, and isolate the specific force measurements they needed. They then demonstrated how the system could automatically cut the continuous stream of data into forty distinct segments, each representing a single hole, based on the natural pauses in the drilling rhythm. Finally, they used the AI agent to repeat this entire cleaning and cutting process across dozens of files at once, a task that would have taken hours of manual labor. The result was a structured database of peak force values, perfectly organized and ready for analysis.
The study highlights that the future of smart manufacturing does not just depend on building smarter machines, but on building smarter ways to handle the information those machines produce. By replacing ad-hoc, error-prone manual work with a systematic, modular approach, the researchers have shown that the "janitorial" work of data cleaning can be standardized and democratized. Their toolkit is freely available to anyone, from small workshops to large research labs, allowing them to bypass the need for expensive, proprietary software or custom coding. The work suggests that when data preparation is treated with the same scientific rigor as the manufacturing process itself, the path to truly intelligent, self-monitoring factories becomes clear and attainable.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.