Fraunhofer Institute for Production Systems and Design Technology
Fraunhofer Institute for Production Systems and Design Technology

When Data Becomes a Burden

Many machining companies face the same dilemma. They collect process data during production, yet expertise for contextualizing and evaluating it is often missing. Triggered sensors can help.

Optimizing manufacturing processes hinges on two key elements: effective data collection and rigorous evaluation. Let’s consider a medium-sized company. The machine tools used in production are capable of providing and storing process data, but evaluating this data would require time that employees do not have.

This is a common issue for many companies, especially small and medium-sized enterprises (SME). The data is often not even stored, but instead only monitored sporadically during machine operation. Some companies have their machines continuously record everything from drive currents and spindle temperatures to vibrations. However, without a clear structure, these data volumes often remain unused on servers. SME in particular simply do not have the capacity to sort and evaluate data retrospectively. Thus, it is not a good option for our example company.

Other companies only collect data manually for individual machining processes that need to be examined more closely, for example, due to a tool failure. A worker at the machine for instance is instructed to »press the button when the milling process starts.« This approach is riddled with errors and makes it almost impossible to reliably assign data to components, tools, and process steps. This is especially true in contract manufacturing, where certain parts are produced very rarely and at irregular intervals. This results in the loss of valuable knowledge about the necessary procedures and processes. If the order comes up again next year, employees may no longer be with the company, or experience may not have been documented.

In our example company, variant diversity and constantly changing component geometries, batch sizes, and machining strategies are the norm. High effort, uncertain data quality, and little added value for data-driven optimizations? Under these conditions, manually planned or random data collection is not a good alternative for our company.

© Fraunhofer IPK / Larissa Klassen
© Fraunhofer IPK / Larissa Klassen
Unstable processes can be detected by inspection under a microscope.
© IWF TU Berlin / Dominik Hasselder
This is what a milling cutter should look like in a stable process: no breakouts on the cutting edges.

Targeted measurement

The solution is triggered data acquisition. The sensors are activated only when necessary, ensuring seamless and reliable recording. At the start of a defined milling process or when the machining cycle of a component-critical geometry begins, it is not the worker who switches on a sensor – it is a signal directly from the process.

This results in smaller, more relevant data sets that can be clearly assigned to individual components and process steps. It is clear that storage space requirements, data processing costs, and personnel deployment are drastically reduced. At the same time, the database’s consistency and informative value increase, making it the foundation for reliable analyses and optimizations.

From raw info to insight

Today’s machine tools already emit a wide range of usable signals. Current sensors provide information about drive power, acceleration sensors help to identify unstable processes, and temperature sensors monitor spindles. Record this data in a targeted manner in the relevant process sections and combine it with contextual information such as order number or tool ID, and this will result in a high-quality data set with concrete practical relevance.

At the IWF at the Technical University of Berlin, researchers have been working for years to use such structured data to implement comprehensive process monitoring. This monitoring can detect cutting breakouts and tolerance deviations and also identify other process anomalies at an early stage. One key benefit is that imaging sensors, such as microscopes used to monitor tool cutting edges, can be triggered automatically. This provides images and process data that are precisely correlated in time.

Our example company now has a comprehensive overview of a critical milling process for the first time: a shared, highly compressed data set consisting of the machining program, acceleration sensor data from the spindle, and high-resolution images of the tool cutting edges. Algorithms can be used to automatically extract breakouts on individual cutting edges from the acceleration data and identify their causes in the long term. This is a task that would be almost impossible to perform manually.

The advantages of triggered data acquisition are clear: simplified data evaluation is just the beginning. The next step is to use temporal correlation to automate data labeling, making it usable for AI approaches and self-learning optimization strategies. This will save our medium-sized example company even more time in its busy production schedule.

Integration into mixed machine parks

However, one obstacle to implementation remains: not every SME has a modern, networked machine landscape, unlike our example company. Surveys conducted as part of the nationwide ProKI network have clearly demonstrated the wide range of initial situations in SME, from fully digitized machine parks to manufacturing environments without an internet connection. It is clear that the different starting points must be taken into account when integrating the new approaches.

The TU Berlin’s IWF has developed various integration paths to make the advantages of triggered data acquisition available to everyone.

  • Simple digital signals directly from the G-code automatically switch sensors on and off via the machine control system’s input and output terminals. Modern sensors generally already have the appropriate trigger inputs, providing a quick solution without any major intervention in the machine control system.
  • Retrofitting approaches with edge devices, including modular digital input and output terminals or OPC UA interfaces, connect older machines to modern data acquisition and visualization platforms.

This allows existing machines to become data-capable resources with minimal effort. SME without their own development department should take advantage of these benefits.

Context is king

The greatest added value is created when the triggered sensor data is systematically stored in a central database and enriched with contextual information. This clearly identifies and links processing steps to quality 
assurance results.

This allows companies to reliably monitor tool life, detect deviations earlier, and plan setup and maintenance intervals more accurately. Downtime is reduced, processes are accelerated, and a steadily growing body of digital process knowledge is created. This knowledge supports informed decisions and helps to respond quickly to changes.

SME in the machining industry, like our example company, should collect triggered data. This is a practical and cost-effective way to enter data-driven manufacturing. Rather than relying on data ballast or manual work rife with errors, they can create a reliable, streamlined database – essential for modern optimization algorithms, AI applications, and sustainable process improvements.

Funding notice

The Demonstration and Transfer Network AI in Production (ProKI-Network) is funded by the German Federal Ministry of Education and Research in the 
program »Future of Value Creation – Research on Production, Services and Work« (FKZ 02P22A000 to 02P22A070) and supervised by the Project Management Agency Karlsruhe (PTKA).

You might also be interested in: