A Comparison of Deep Recurrent Neural Networks and Bayesian Neural Networks for Detecting Electric Motor Damage Through Sound Signal Analysis

Fault detection in electric motors represents a critical challenge across various industries, as failures can lead to substantial operational disruptions. This study examines the application of deep neural networks (DNNs) and Bayesian neural networks (BNNs) for diagnosing motor faults through acoustic signal analysis. We propose a novel approach that leverages frequency-domain representations of sound signals to improve diagnostic accuracy. The architectures of both DNNs and BNNs are developed and evaluated using real-world acoustic data collected from household appliances via smartphones. Experimental results indicate that BNNs achieve superior fault detection performance, particularly in the context of imbalanced datasets, providing more robust and interpretable predictions compared to conventional methods. These findings suggest that BNNs, owing to their ability to incorporate uncertainty, are well-suited for industrial diagnostic applications. Further analysis and benchmarking are recommended to assess the resource efficiency and classification capabilities of these architectures.