Deep Learning-Based Fault Diagnosis for Rotating Machinery Using Multiscale Convolutional Networks

Abstract

Fault diagnosis of rotating machinery plays a crucial role in modern industrial systems, where unexpected equipment failures may lead to severe economic losses and safety risks. Traditional signal-processing-based diagnostic methods rely heavily on handcrafted features and domain expertise, limiting their adaptability to complex operating environments. With the rapid advancement of deep learning techniques, data-driven fault diagnosis has become an effective alternative for intelligent machinery monitoring. This study proposes a multiscale convolutional neural network (MSCNN) framework for vibration-based fault diagnosis of rotating machinery. The proposed model extracts hierarchical features from vibration signals through multiscale convolutional kernels, enabling the detection of both local transient patterns and global temporal structures. Experimental evaluations conducted on benchmark bearing datasets demonstrate that the proposed method significantly improves diagnostic accuracy compared with traditional machine learning approaches. The results confirm that deep learning-based feature learning can effectively enhance reliability and robustness in mechanical fault detection systems.