Fault diagnosis for submerged combustion vaporizer blower bearings based on deep residual shrinkage network

This research addresses the critical need for accurate fault diagnosis of rolling bearings in blowers within submerged combustion gasifiers, a key component ensuring the operational safety of such systems. To tackle the challenge of noise interference during fault feature extraction, a novel diagnostic method based on deep residual shrinkage networks is proposed. In this framework, residual learning modules leverage non-linear mappings to enable efficient training of deep network architectures; moreover, each module integrates a sub-network with soft-threshold non-linear transformations, which, combined with an attention mechanism, significantly mitigates the influence of noise in input data. The proposed method is verified through experimental evaluations on a large dataset of bearing fault signals. Results show that the enhanced network consistently achieves high diagnostic accuracy, even in environments with substantial noise contamination.