Machine Fault Detection Via Audio Signal Analysis

<jats:p>Machine fault detection is an essential component of predictive maintenance in modern industrial systems, where early identification of abnormal machine behaviour can reduce downtime, maintenance cost, and operational risks. This paper presents a deep learning–based machine fault detection system using audio signal analysis for intelligent monitoring of industrial machinery. The proposed system utilizes machine sound recordings as input and performs automated fault classification using Convolutional Neural Networks (CNNs). In the proposed approach, raw machine audio signals are converted into Mel spectrogram representations to capture time–frequency characteristics of machine sounds. Transfer learning using the MobileNetV2 architecture is employed to improve feature extraction and classification performance while reducing computational complexity and training time. The model is trained to classify machine conditions into normal and faulty categories using industrial audio datasets. A real-time web application is also developed to demonstrate practical deployment of the system. The application captures machine audio through a smartphone or microphone, performs live prediction, displays confidence scores, and visualizes spectrogram outputs. Experimental results show that the proposed model achieves high classification accuracy with minimal overfitting, demonstrating the effectiveness of deep learning for non-invasive industrial fault diagnosis. The proposed system provides a scalable, cost-effective, and intelligent predictive maintenance solution suitable for modern smart manufacturing environments.</jats:p>