Application of Quaternion-Based Convolutional Neural Networks for Advanced Motor Fault Diagnosis

In industrial rotating machines, early detection of shaft misalignment is essential for stable operation and reduced maintenance costs. This study presents a method that uses Quaternion Convolutional Neural Networks (Quaternion CNN) to improve accuracy and efficiency in diagnosing faults in rotating machines. One-dimensional signal arrays are collected from an experimental model and go through preprocessing to filter out noise before being input into the classification model. The model, enhanced by Quaternion Convolutional Neural Networks and combined with Bidirectional Gated Recurrent Unit and Channel and Spatial Attention Feature Fusion, accurately identifies shaft misalignment faults, offering significant benefits for fault diagnosis in complex rotating machine systems.