Why industry is betting on AI in machine tools

Short-term effects are overestimated, long-term changes underestimated: This is how experts describe the perception of disruptions such as artificial intelligence (AI). Where AI is already being practically used in machine tools, what limits science sees, and what alliances are all about.

High-tech company Trumpf sees AI as an opportunity to differentiate itself from global competition in the manufacturing sector. “Europe has a unique opportunity to combine its deeply rooted industrial expertise and strong sense of responsibility with AI technology,” emphasizes Sarah Engel, head of AI at Trumpf. “This is where we fundamentally differ from nations such as the United States and China.”

Sarah Engel backs this up with concrete examples. For automated sorting of a wide variety of parts in laser cutting machines, SortMaster Vision is used. To this end, Trumpf has integrated adaptive robotics functions, including automated perception and the planning of robot paths.

This allows a robot to autonomously identify the parts using AI-based image recognition. The software independently calculates the robot’s motion plan, including the gripping points. In doing so, SortMaster Vision automatically retrieves all part information from the cutting program. This eliminates any need to program the sorting process. In addition, the process creates maximum flexibility in terms of part variety and batch sizes down to a quantity of 1.

Improved parameters for the process

Trumpf is already using AI for the core process itself, laser cutting. The “Cutting Assistant” improves the quality of cut edges, saves time, and enables even “beginners” to achieve high component quality. The machine operator uses a handheld scanner to capture an image of the cut edge of the component. The assistant evaluates the edge quality on the basis of objective criteria such as burr formation. Using this information, the algorithm proposes improved parameters for the cutting process. The machine then cuts the sheet again. No prior experience with laser cutting is required. The Cutting Assistant is integrated into the machine’s software. As a result, the optimized parameters can be transferred seamlessly into the software without any programming.

Protecting user expertise

The data from applications in the field continuously flows into the solution. This approach enables even faster and more reliable results because the self-learning system continuously improves. Trumpf ensures that the algorithm does not spread the user’s expertise. “When developing AI, we always first consider what data will be used for which purpose, what the risks are, and how we can keep them in check,” says Sarah Engel, referring to the close collaboration with the in-house legal, IT, and cybersecurity departments.

Another AI function from Trumpf, already in use in the automotive industry, inspects the quality of complex components, such as batteries for electric vehicles. A camera photographs the weld seam in the cell, and the AI analyzes whether it meets the specified criteria. In doing so, the AI works with traceable measurement values. This is important for transparent quality control without black-box effects. If the AI detects an error, it notifies the user while the component is still in the laser cell. This increases the first-pass yield, that is, the number of parts that meet quality standards on the first run. No programming knowledge is required for this. Users train the AI using image data of correct and defective weld seams.

For Sarah Engel, the specific requirements of customers are the priority: “As a solution provider, we comprehensively master our process chains, from hardware to service.”

Not suitable for safety-critical systems

As a heavyweight in the industry, Trumpf is certainly a pioneer. Laser processing also opens up degrees of freedom that machining processes do not have. In principle, the use of AI in machine tools is limited to non-safety-critical systems and functions. “There is currently no safe AI that learns in the field. For safety-critical applications in the field, it must always be traceable what the software does and why,” emphasizes Gereon Weiß, head of the automation systems department at the Fraunhofer Institute for Cognitive Systems IKS.

Manufacturers such as Index Werke share this view. With its Index and Traub brands, the company is one of the world’s leading manufacturers of CNC lathes. “Our machine tools have to operate with absolute precision and maximum repeatability over many years. AI technologies, on the other hand, are evolving extremely quickly, and their statistical nature can lead to results that are factually incorrect or unreliable,” explains Dr. Christopher Kohl, digitalization project manager at Index Werke. “That is why we use AI where it creates real added value: in mobile assistance solutions such as iXmobile and in cloud-based value-added services like our IIoT platform iX4.0, but not as an integrated function in the machine itself.”

Assistance functions evaluate process data and reveal optimization potential or increase availability. Such solutions can be updated quickly and further developed independently of the machine life cycle. Specialized AI applications, such as predictions of workpiece quality or tool life, are highly process- and user-specific. From the machine manufacturer’s point of view, they are therefore ideally developed by specialized third-party providers that focus exclusively on such data-driven solutions.

Always plan for retraining

Before using AI functions in day-to-day production, however, you need to look closely at the fine print. While production systems – once commissioned – often run for many years with unchanged software, innovation cycles in AI are measured in months. Even when frozen versions are used, the AI functions and their tasks require continuous monitoring. “I always have to plan for retraining,” warns Gereon Weiß, head of the automation systems department at the Fraunhofer Institute for Cognitive Systems, IKS. “I have to be able to work with it myself or take care of suitable service providers.”

Proprietary AI models still dominate today. But what happens if a provider discontinues its service? “Users should run through this scenario before making a decision and rely on standards, give preference to open-source models, or secure access to the source code,” explains Dr. Gereon Weiß as possible solutions.

Risk management as with safety

For the decision-making process, he recommends risk management strategies such as those known from safety. “The point is to configure the system so that errors are reliably caught. This reduces error costs for users on the one hand, but on the other hand they may not fully exploit the entire potential of AI,” is how he describes the necessary trade-off. For this, it is necessary to know how reliably the system operates. One option at the beginning is a “shadow mode.” 

In this case, the AI function runs passively in the background but does not trigger any actions. Humans do not monitor the system during live operation, but instead evaluate how well the AI works before it is actually switched to active mode in production. As a matter of principle, critical decisions must always be made by humans. “There is no zero-risk scenario, even in non-safety-critical systems. In each individual case, we must precisely identify the benefits and define which costs are acceptable for them,” says the researcher from the Fraunhofer Institute IKS.

Human-centered human-machine interaction

The Institute for Production Management, Technology and Machine Tools (PTW) at TU Darmstadt is also intensively involved with artificial intelligence. “We research data-driven approaches and develop AI solutions for various manufacturing use cases,” explains Gilbert Ely Engert, team leader of the manufacturing technology research group at PTW, describing their approach. “AI-based optimization of process planning and in-process control using machine learning, transfer learning approaches for transferring models to other machines and processes, and the human-centered development of contemporary intelligent systems for interaction between humans and machines with AI agents.”

The focus is on use in small and medium-sized enterprises and in small-batch and one-off production, as opposed to fully automated high-volume manufacturing. “Here you typically find heterogeneous machine landscapes, small batch sizes, and a shortage of skilled workers.” Most recently, PTW has presented new results for human-machine interaction. The new operating concept aims at more efficient processes and fewer errors on existing machines. The researchers pay particular attention to the next generation of operators who grew up with smartphones. 

"As part of a research project with industry, we have developed a novel AI‑supported system for operating machine tools, which is intended to make operation as easy as using smartphones," he says, presenting a demonstrator at PTW in Darmstadt. Interested parties can try out how the new operation works themselves during a one-hour evaluation in June at PTW (see box).

Hey AI, manufacture my component for me

Can a machine tool be operated as easily as a smartphone? As part of a research project with industry, Gilbert Engert has developed a novel AI‑supported system for operating machine tools and implemented it in a demonstrator. For his dissertation, Gilbert Engert is looking for test participants to evaluate the new human‑machine interface, with and without prior experience in machine operation. The test duration is one hour.

Dates: Every Wednesday, Thursday and Friday in June 2026 from 09:00 to 17:30 in the TEC Lab of PTW at TU Darmstadt.

You can find the registration here.

Contextualized data required

Whether the available AI models are good or bad plays a subordinate role for Dr.-Ing. Marcel Fey from RWTH Aachen from the users’ point of view. The senior engineer in the machine technology department at the WZL machine tool laboratory sees a different core challenge: "In the heterogeneous production landscape, data is rarely cleanly usable because it is not sufficiently contextualized." Users do record raw and time series data from machines, such as axis positions, speeds, feed rates, alarms, or energy curves.

"On their own, however, these signals are usually just ‘noise’ as long as it is not clear which job, which workpiece, which tool, which NC program, which machine or kinematics, or which process phase they belong to," the researcher explains. In reality, this contextual information often exists, but in higher-level systems outside the machine tools.

For him, therefore, the real task in many cases is to link the data base and data logic: clear identifications, consistent time references, clean event and process segmentation, and traceable assignment of context to time series. This effort is worthwhile, Dr.-Ing. Marcel Fey is certain: “Even on the way to a clean data base, companies often solve a large part of their problems without AI – simply through transparency, better traceability, consistent process data, clear KPI definitions, and reliable evaluations.”

Siemens AI data alliance for open standard

These assessments do not come out of thin air. Siemens, together with various machine tool manufacturers and the machine tool laboratory at RWTH Aachen University, has founded a comprehensive AI data alliance. The goal is an industry-specific foundation model for artificial intelligence. “Our goal is to support engineers in handling time-consuming tasks more productively, to ensure compliance with industry standards, and to improve and accelerate the product development process in the areas of design, planning, engineering, operation, and service,” explains Dr. Stefanie Frank, head of machine tool systems, Siemens Digital Industries, by way of background.

One planned use case is the automated creation of part programs for machine tools. This makes it possible to create part programs significantly faster while reducing the error rate in code generation. The prerequisite for this is the contextualized processing of machine data. Among other things, this data is used to develop and train AI models that are specifically tailored to the requirements of industrial manufacturing. “The partnership includes the exchange of anonymized machine data under strict compliance with data protection and security standards,” emphasizes Dr. Stefanie Frank.

In the long term, the alliance is intended to become an open platform approach that will also be accessible to other industrial companies in order to establish cross-industry standards for AI-based manufacturing. This confirms the following: The short-term effects of AI may be overestimated by many. But most people certainly underestimate how fundamental the long-term impact of AI will be.