Multi-Scale Recursive Semi-Supervised Deep Learning Fault Diagnosis Method with Attention Gate

The efficiency of deep learning-based fault diagnosis methods for bearings is affected by the sample size labeled data, which might be insufficient in engineering field. Self-training a commonly used semi-supervised method, usually limited accuracy features unlabeled data screening. It significant to design an efficient training mechanism extract accurate and novel feature fusion ensure that fused capable A multi-scale recursion (MRAE) designed Autoencoder this article, can extraction with small amount data. An attention gate-based was constructed make full use all useful sense it incorporate distinguishing on different scales. Utilizing large numbers proposed recursive learning method gate (MRAE-AG) efficiently improve performance DNNs trained number benchmark dataset from Case Western Reserve University bearing center validate shows 7.76% improvement achieved case when only 10 samples available supervised DNN-based model.