MR-FuSN: A Multi-Resolution Selective Fusion Approach for Bearing Fault Diagnosis

TL;DR

The paper introduces MR-FuSN, a new network for diagnosing bearing faults in noisy industrial environments, achieving high accuracy even with low signal quality.

Contribution

The novel MR-FuSN architecture uses multi-resolution feature extraction and adaptive kernel convolution to improve fault diagnosis in noisy settings.

Findings

MR-FuSN achieves 99.97% diagnostic accuracy under 0 dB signal-to-noise ratio conditions.

The method performs well in environments with SNRs ranging from -5 dB to 10 dB.

The architecture enhances robustness against environmental noise through adaptive kernel convolution.

Abstract

Vibration signals serve as the primary data source for bearing fault diagnosis. However, when collected in complex industrial environments, these signals are often contaminated by noise interference, posing significant challenges to fault feature extraction and diagnostic accuracy. To address these issues, this paper proposes a novel bearing fault diagnosis network architecture: the Multi-Resolution Fusion Selection Network (MR-FuSN). The MR-FuSN effectively extracts domain-invariant features from input data through multi-resolution feature extraction and incorporates an adaptive kernel convolution strategy, thereby enhancing its robustness against environmental noise. Experimental results demonstrate that the MR-FuSN achieves outstanding performance in noisy environments with signal-to-noise ratios (SNRs) ranging from −5 dB to 10 dB, particularly attaining a diagnostic accuracy of 99.97% under 0 dB conditions. This study provides technical support for practical fault diagnosis applications.

Figures