ÂÌñÏ×ÆÞ

The Augsburg AI (artificial intelligence) Production Network — founded in January 2021 and based in Augsburg, Germany — brings together the University of Augsburg, the Fraunhofer Institute for Foundry, Composites and Processing Technology (Fraunhofer IGCV) and the Center for Lightweight Production Technology of the German Aerospace Center (DLR ZLP). The aim is joint research into AI-based production technologies at the interface between materials, manufacturing technologies and data-based modeling. One application example where AI can support production processes is the machining of fiber-reinforced composite materials.

Structure-borne sound and machine learning

Industrial CNC milling machine used for ultrasound sensor research. Photo Credit: Copyright University of Augsburg

Most modern CNC milling machines have some basic sensors already built-in, which record energy consumption, feed force and torque, for example. However, this data is not always sufficient for the resolution of fine details in the milling process. For this reason, ultrasonic sensors for analyzing structure-borne sound were developed at the University of Augsburg and integrated into an industrial CNC milling machine. These sensors detect structure-borne sound signals in the ultrasonic range that are generated during the milling process and then propagate through the system to the sensor.

Featured Content

Automated Composites Knowledge Center READ MORE

Market report: Composites and carbon fiber in hydrogen storage READ MORE

Tour NIAR at Carbon Fiber 2025 READ MORE

The structure-borne sound allows conclusions to be drawn about the state of the machining process. “This is an indicator that is as meaningful to us as bow strokes are for a violin,” explains Prof. Dr. Markus Sause, director of the AI ​​Production Network. “Music professionals can tell immediately from a violin’s sound whether it is in tune and how well the person playing it has mastered the instrument.” But how does this approach work for a CNC machine? Machine learning is the key.

In order to optimize the CNC milling process based on recorded data from the ultrasonic sensors, researchers working with Sause make use of what is known as machine learning. Certain characteristics of the acoustic signal can indicate an unfavorable process control, which points to poor quality of the milled component. As a result, the milling process can be directly adjusted and improved with this information. For this purpose, an algorithm is trained with recorded data and the corresponding states (e. g. good or bad machining). The person who operates the milling machine can then react to the system status information that is presented or the system can be programmed to react automatically.

Predictive maintenance - acting with foresight

Machine learning can not only optimize the milling process directly on the workpiece, but can also plan the maintenance cycles of a production plant as economically as possible. Functional parts need to work in the machine for as long as possible in order to increase economic efficiency, but spontaneous failures due to damaged parts must be avoided.

Predictive maintenance is an approach in which the AI, using the collected sensor data, calculates when a part should be replaced. In the case of the CNC milling machine under investigation, an algorithm recognizes when certain characteristics of the sound signal change. In this way, it not only identifies the degree of wear on the machining tool, but also predicts the right time to replace the tool. This and other AI processes are being incorporated into the AI ​​Production Network in Augsburg. The three main partner organizations are working with other production facilities to create a manufacturing network that can be reconfigured in a modular and material-optimized manner.