Fault diagnosis method for small sample based on adaptive margin loss
Existing metric learning-based fault diagnosis methods lack consideration of margins while optimized by the loss function,making the model overly sensitive to the training data.To this end,an adaptive margin loss function is constructed to reduce overfitting and improve the model performance on new samples.This margin loss helps the model to learn the relative distances between samples and thus gain enough distance information to generalize to new samples.It automatically adjusts the margin size according to the distribution of training data and the model performance to adaptively distinguish different fault categories.To better meet the small sample,a small sample fault diagnosis method based on a meta-metric network with adaptive margin loss is proposed by combining the meta-learning episode training mode and the flexibility of the metric module in metric learning.The metric module introduces the cosine similarity to improve the expressiveness,then guides the optimization and model training to fit the small sample better.To validate the effectiveness of the proposed method,a dataset consisting of UAV flight log data with faults is used,and the proposed method is compared with the traditional methods based on metric learning.Experiment results show that the proposed method performs well in small sample fault diagnosis for UAV with good diagnostic performance and stability.The proposed method provides an effective solution to ensure the safe and reliable operation of UAV.
NETL
NSTL
万方数据
