In order to realize engine fault detection, a fault detection algorithm EA-ENC based on Few-shot learning is proposed in this paper. In recent years, more and more deep learning algorithms have been applied to the field of industrial detection, and certain achievements have been achieved. However, a common problem is that it is difficult to obtain available samples in the field of fault detection, and the labeled samples are too few. This problem of sample imbalance leads to the difficulty of model optimization, and the proposed method is difficult to be applied in engineering. This paper proposes a fault detection scheme based on Few-shot learning. The data involved in training is augmented by Few-shot learning, and the distribution reference line of the category is obtained by training with custom N_ways and N_shot. The classification reference line is constantly optimized during training, and the learning ability of the algorithm for features is strengthened by increasing the network depth of feature extractors. Through comparison and testing, the proposed algorithm can be applied in current industrial fault detection engineering.
|