Low-Cost Embedded System for Acoustic Fault Detection

Abstract

In recent years, acoustic signal analysis has emerged as a promising approach for fault detection in electromechanical systems, providing a non-invasive, low-cost, and real-time alternative to traditional monitoring techniques. This paper presents the design and implementation of a low-cost embedded system for acoustic fault detection. The proposed system is built on a Raspberry Pi Nano platform, integrating a low-power microphone, real-time audio signal preprocessing, and lightweight machine learning models for classification of normal versus faulty operating conditions. The focus is placed on developing an efficient signal processing pipeline that includes noise reduction, feature extraction (time and frequency domain descriptors, Mel-frequency cepstral coefficients), and on-device classification using compact neural network architectures. The embedded setup enables autonomous monitoring without the need for external computation resources, making it suitable for edge deployment in industrial and IoT environments. Experimental validation is carried out using publicly available datasets such as MIMII (Malfunctioning Industrial Machine Investigation and Inspection), as well as preliminary real-time recordings. The results demonstrate that the system achieves reliable fault detection accuracy while maintaining low computational and energy costs, highlighting its potential for scalable deployment in smart maintenance applications.