Differential contrast guidance for aeroengine fault diagnosis with limited data

Data-driven methods have high requirements for data samples and the ideal state is to have sufficient samples and labels for model training. However, due to the limited sample of aeroengine fault data, existing methods often cannot achieve good classification results. To solve this problem, a contrastive learning strategy guided by fault type differences for aeroengine fault diagnosis with limited samples is proposed. Different from the traditional contrastive learning paradigm using data augmentation, the proposed method uses the fault data to construct sample pairs, uses similarity comparison to learn fault features from limited data, and uses the learned fault features for fault diagnosis. A deep learning model for joint training of feature extractor and classifier is built to improve the fault diagnosis accuracy. Finally, the aeroengine dataset and bearing dataset are used to verify the effectiveness of the proposed method in the case of limited data. The experimental results show that compared with the most advanced methods, the proposed method can achieve higher fault diagnosis accuracy.