HMN 2026: How Rotating acoustic filter isolates machine fault sounds in 100 dB noise

Seoul National University College of Engineering announced that a research team led by Prof. Sung-Hoon Ahn of the Department of Mechanical Engineering has developed, for the first time in the world, an “interference acoustic band-pass filter” capable of selectively filtering and amplifying specific frequencies without the use of electronic circuits.

By utilizing a single microphone and an interference-based metamaterial structure, the researchers have demonstrated a technology that enables selective listening to desired frequencies, opening new possibilities for diagnosing machine failures in high-noise industrial environments.

The research findings were published in Mechanical Systems and Signal Processing.

Industrial environments such as factories, power plants, and aircraft engine rooms are filled with intense noise levels ranging from 80 to 100 decibels (dB). In such environments, subtle “abnormal sounds” emitted by machines just before failure are often masked by overwhelming background noise, making it difficult to detect early warning signs such as minor cracks or mechanical wear. This frequently leads to major accidents, resulting in casualties, significant repair costs, and production disruptions.

To address this issue, technologies have emerged that diagnose machine failures using sound in high-noise environments. These approaches rely on the fact that machines emit different sound frequencies when operating normally versus when malfunctioning. By isolating specific “abnormal frequency” components associated with faults, these systems can detect anomalies.

However, such techniques typically require band-pass filters and complex microphone arrays that rely on electronic circuits or computer software to extract specific frequencies from sensor signals. These conventional approaches suffer from high computational costs and require redesigning expensive circuits or structures each time a different type of machine fault (i.e., different frequency) needs to be detected.

To overcome these limitations, Prof. Ahn’s research team developed, for the first time in the world, a hardware-based acoustic band-pass filter capable of selectively filtering and amplifying sound in the 1.8–22 kHz range using only a single microphone and an ultra-compact metamaterial structure with a volume of just 0.2 liters.

The research originated from the idea that sound can be filtered purely through interference-based structures without relying on electronics. To realize this concept, the team designed a cylindrical “interference structure” featuring multiple slits arranged at regular intervals. Sounds entering through these slits are engineered to interfere constructively or destructively, enabling frequency selection. As a result, the band-pass filter enhances specific frequencies and directions by utilizing phase differences in sound waves rather than electronic signal processing.

Furthermore, the system allows frequency selection simply by rotating the structure. The frequency that is selectively amplified changes depending on the angle (e.g., 2 kHz ? 71°, 5 kHz ? 20°, 10 kHz ? 11°). This means that users can physically rotate the device to selectively listen to desired frequencies. Unlike conventional approaches, this design eliminates the need to redesign or rebuild filters when detecting different types of machine faults. In other words, the researchers have realized a hardware acoustic band-pass filter that overcomes the lack of versatility inherent in existing systems.

In field experiments validating the effectiveness of the band-pass filter, the researchers observed that even under 100 dB noise conditions—comparable to construction sites, club music, or train noise—the target frequency signal was amplified by 4.82 times. In additional experiments using a CNC (Computerized Numerical Control) machine, the abnormal frequency associated with faults (2041 Hz) was amplified 19.9 times, demonstrating the outstanding performance of the newly developed filter.

Moreover, the study achieved a significant improvement in AI-based fault detection performance. While conventional band-pass filters failed entirely (0% detection rate) under noisy conditions, the new system increased detection accuracy to 78.6%, indicating that failures previously undetectable can now be diagnosed in advance. These results demonstrate that a single hardware structure can outperform the roles traditionally handled by complex electronic filters and multi-microphone arrays.

The interference-based metamaterial structure developed in this study builds upon the team’s previous work on single-sensor-based 3D acoustic range (3DAR). In that earlier research, the team demonstrated an innovative acoustic sensor capable of estimating sound location using only a single microphone and a rotating structure. In the present study, the researchers extended this concept beyond sound localization to enable selective filtering of sound, marking a significant advancement in acoustic sensing technology.

This technology is expected to be applied as a hardware-based system for detecting critical abnormal signals in high-noise industrial environments such as smart factories, robotics systems, aircraft, and wind turbines.

For example, it could automatically detect abnormal sounds from CNC machines or motors in factories to prevent accidents, or be extended to systems capable of identifying pipeline leaks or collision sounds in continuous noisy environments. Because the system operates purely through hardware, it offers key advantages including zero power consumption, low failure risk, and reduced maintenance costs, increasing its potential for widespread adoption in various applications.

Prof. Sung-Hoon Ahn stated, “This research is significant in that it demonstrates ‘mechano-intelligence,’ where machines process and interpret information through physical structures before electronic computation. By embedding physical knowledge into structures, the system can reduce computational burden and operate more efficiently.”

Semin Ahn, the first author of the study and a Ph.D. candidate, added, “The world’s first hardware acoustic filter allows frequency control simply through angle adjustment. When combined with artificial intelligence, it is expected to accelerate the era of intelligent machines capable of making accurate decisions even in noisy environments.”

Semin Ahn, a Ph.D. candidate in the Department of Mechanical Engineering at SNU, is currently developing foundation model–based cognitive, decision-making, and action systems that enable machines to understand the meaning of sound in a human-like manner at the Innovative Design and Integrated Manufacturing Laboratory. He is also conducting research on multi-robot collaboration.

The content is provided for information purposes only.