Project: Wave - Development of integrated MEMS sensors for AI-supported maintenance

The aim of the WAVE (Wide Application Vibration Element) project was to develop an intelligent sensor node for monitoring machine status in real time. By combining airborne and structure-borne sound sensors and integrating active ultrasonic measurement methods, machine defects can be detected at an early stage and maintenance requirements reliably predicted. The complex sensor signals are evaluated using statistical methods and machine learning techniques.

Practical test scenarios and system validation

Two model environments were set up to develop and validate the system:

• Model 1: A rotating shaft with ball bearings in various states (intact and defective) – a realistic test situation, but with unknown dynamic boundary conditions.
• Model 2: A vibration platform with adjustable frequency and deflection – ideal for reproducible tests to verify the data processing chain.

In addition, a software concept for data acquisition, processing, and AI-supported analysis was created based on previous projects.

Sensors and signal processing

The sensor assembly was developed in several iterations and continuously improved. It comprises:

• Airborne sound sensor (Knowles SPH18C3LM4H-1): Detects sound signals from 40 Hz to 100 kHz.
• Acceleration sensor (ADXL1005): Measures structure-borne sound up to 23 kHz.
• Active ultrasonic sensor (LCMUT from Fraunhofer IPMS): Transmits a constant signal; reflected echoes are analyzed using Doppler effects.

A key innovation is the use of active ultrasonic measurements to detect fine vibrations through changes in the reflected signal pattern – an approach that has proven to be particularly sensitive to changes in machine behavior.

Demonstrator in real-world use

Thanks to the more efficient project process, it was also possible to create a demonstrator on a crosscut saw. Here, the sensor node was used to classify the condition of circular saw blades – in particular, to distinguish between sharp and blunt blades.

• Measurement of sound signals from saw blades with different defects
• Significant differences in the recorded data depending on condition
• Successful classification using machine learning models

Findings and outlook

• The sensor node provides a modular platform for acoustic and mechanical condition monitoring of machines.
• The combination of passive and active measurement methods allows a wide range of possible defects to be detected.
• The knowledge gained provides a solid foundation for future applications in predictive maintenance and quality assurance.

In-depth investigations in the high-frequency range were deliberately avoided, as the technical effort involved would have been too high in comparison to the expected benefits. Instead, the focus was on optimizing the overall system for industrial applications in the kHz range.