DCANet:Dense channel attention network based on SimCLR for fault diagnosis of aviation transmission magazine

Aero engines and their accessory subsystems operate under complex operating conditions in a production environment,and usually,the equipment will experience variable acceleration and deceleration.The existing models and methods are limited by many problems,such as network structure and poor feature extraction ability,resulting in low diagnostic accuracy under variable working condition datasets.A Dense channel attention network(DCANet)based on SimCLR is proposed to address the above problems.This method combines the advantages of DenseNet and SENet.It can enhance feature reuse,integrate the feature information of high,middle,and low layers,and realize recalibration in the feature extraction process.It effectively enhances the feature extraction ability of single-channel time-frequency images,recognizes the attention weighting of features,and extracts effective features.At the same time,SimCLR's twin network small-sample learning method is introduced to solve the problem of low accuracy due to the small amount of data.Many experiments show that the recognition rate of the proposed method is better than the existing methods under the variable working condition dataset.The highest accuracy rate is 97.12%,which has been obtained on the Dongan dataset.