Enhancing Sound-Based Anomaly Detection Using Deep Denoising Autoencoder

Abstract

Traditional sensor-based methods that detect machine malfunctions in industrial environments are often costly and complex; sound-based anomaly detection offers a simpler alternative. Such methods, however, must contend with industrial noise that masks the sound patterns necessary for effective diagnosis. This study proposes a method that uses a deep denoising autoencoder to filter out various levels of industrial noise from audio data and employs unsupervised learning models for rapid and accurate anomaly detection. The primary novelty of the proposed methodology lies in the audio data preprocessing techniques and the customized denoising process that is tailored to the noise levels of various industrial environments. Several experiments using different types of industrial machinery, such as pumps, valves, and slide rails, demonstrated the efficiency, effectiveness, and rapid processing capabilities of the proposed methodology. Specifically, the experimental results show that the proposed methodology afforded an average area under the curve performance improvement of approximately 18% compared to previous studies.