A Lightweight Enhanced MobileNetV3 Method for Bearing Fault Diagnosis Integrating Vibration Signals and Multi-Sensor Features
TL;DR
Experimental validation demonstrates that by using the GADFMap and the Enhanced MobileNetV3 model, the proposed method achieves higher diagnostic accuracy with fewer parameters and lower computational complexity compared with mainstream models.
Abstract
Rolling bearings are vital for the reliable operation of mechanical systems. However, accurately and efficiently identifying faults under complex working conditions remains a significant challenge. This paper proposes a lightweight and high-precision diagnostic framework specifically designed for industrial edge computing. Specifically, a multi-sensor data fusion method, termed GADFMap, is introduced. First, the Wavelet Packet Decomposition (WPD) is applied to extract three sub-band signals with the highest kurtosis values, which are then weighted and combined to generate a new signal. Subsequently, the resulting signals are encoded using Gramian Angular Difference Field (GADF) to effectively integrate the multi-sensor data. Moreover, an improved lightweight diagnostic network, Enhanced MobileNetV3, is developed by augmenting MobileNetV3-Small with Efficient Channel Attention (ECA) modules. This improvement reduces model parameters and computational complexity, while strengthening the model’s focus on salient features. Experimental validation demonstrates that by using the GADFMap and the Enhanced MobileNetV3 model, the proposed method achieves higher diagnostic accuracy with fewer parameters and lower computational complexity compared with mainstream models.