The machine thinks for itself

AI in machine tools means much more than just automation. It enables machines to learn from data, make decisions and optimize processes. This involves the use of sensors, data analysis, machine learning and intelligent assistance systems - both at the control level and in interaction with humans.

There are many potential applications for manufacturing companies. "Typical examples are the prediction of process properties in real-time operation for inline quality control and the monitoring of processes and their properties," says Prof. Philipp Klimant, Business Unit Manager Process Digitization and Manufacturing Automation at the Fraunhofer Institute for Machine Tools and Forming Technology (IWU) in Chemnitz. "The advantage over traditional approaches here is the ability to include a particularly large number of parameters in the monitoring process," emphasizes Klimant. However, there are also numerous other fields of application, such as AI assistance models for training and artificial intelligence to support maintenance.

The Fraunhofer IWU, which specializes in the production-related adaptation of machine learning methods, is headed by Martin Dix, Welf-Guntram Drossel and Steffen Ihlenfeldt. All three are members of the Wissenschaftliche Gesellschaft für Produktionstechnik (WGP), an association of German professors of production science. Since January, the WGP has consolidated the ProKI initiative, which was originally funded by the BMBF, under its umbrella and has since been offering practical expertise and demonstrators, particularly for small and medium-sized companies.

Great leverage for efficiency gains

The IWU researcher's tip is to start by asking yourself: How high are the efficiency gains that are actually possible through AI in my production? "The question of possible efficiency gains cannot be answered in general terms without further ado," says Klimant. The potential depends heavily on the actual process and the associated optimization options. "In the field of plastics processing, reject rates of 20 to 30% can occur in rare cases." This is a major lever for efficiency gains with AI. It can also be useful for processes that are already running stably, for example for predictive maintenance and to achieve longer tool service lives.

According to Klimant, artificial intelligence can also make an important contribution to countering the shortage of skilled workers. "We store knowledge implicitly in AI. This knowledge can be used to train new employees, especially when older colleagues retire and important knowledge leaves the company," explains the scientist. "This AI knowledge store also offers new opportunities for automation, not least for automated quality controls," says Klimant.

The researcher defines artificial intelligence as follows: "When we talk about AI, we usually mean machine learning as a subgroup of AI. This is able to learn independently from training data. This is an empirical process that learns correlations without us knowing the analytical correlations. Simply put, we learn from experience." AI is used to optimize process parameters in production and feed them back into process control via an automated control system. "Artificial intelligence is like a black box, input values go in and forecasts come out," says Klimant. "One example is a forming process where we measure an acoustic signal and the AI tells us whether the process was successful or not." Ultimately, it is a digital system that can be connected to control systems via existing interfaces. This would allow AI to influence control algorithms at various points.

In order for artificial intelligence to be used successfully in production, hardware with very high computing power is sometimes required. "First of all, a distinction needs to be made between the training phase and the utilization phase (inference). The training phase is always more computationally intensive, but is carried out offline. In the utilization phase, edge devices are often sufficient for classic methods such as the support vector machine," says Klimant. The situation is different when it comes to image processing. These AI models require more computing power, both in the training and usage phases. "The application cycle also plays a decisive role here," explains the researcher. The evaluation of language models is an exception here. These require powerful hardware, from high-performance consumer graphics cards to special AI cards.

Self-learning machine tool for autonomous production

Artificial intelligence makes the self-learning machine tool possible. Milling machine manufacturer and EMO exhibitor Datron is relying on this innovation, in which the machine draws on learned knowledge and adapts the production process. The Datron milling machine is set to develop into an adaptive production cell that automatically adapts to component requirements and environmental conditions. "This will not only reduce set-up and machining times, but also increase process stability - a decisive step towards autonomous production," says Jonas Gillmann, Chief Technology Officer (CTO) of the machine manufacturer. AI is thus shifting the focus away from rigid programming towards assisted, learning and adaptive manufacturing. "Machines are becoming partners in the manufacturing process that adapt to humans - not the other way around. In mechanical engineering, this is no longer a vision, but is increasingly becoming a reality," says Gillmann. According to him, AI in production offers high efficiency gains: "In CNC production with Datron machines, it can reduce set-up times by up to 60%, significantly reduce waste and extend tool life - while at the same time increasing process reliability."