Attention-Aware Meta-Reweighted Optimization for Enhanced Intelligent Fault Diagnosis

Due to stringent aircraft safety requirements and the high cost of experiments, there is a scarcity of failure samples, which creates a gap between existing diagnostic models and practical applications. To address this issue, we have developed a small-sample civil aircraft fault diagnosis method. This method combines a meta-learning approach to tackle data imbalance with a channel attention mechanism to enhance feature extraction efficiency. Specifically, our approach integrates the advantages of meta-learning and attention regularization, effectively addressing both the imbalance in training sample distribution and the need for human interaction to enhance feature representation. We then evaluated five data imbalances and introduced a fault diagnosis algorithm based on a one-dimensional convolutional network, which has been successfully applied to solve small sample yield tasks in two datasets. Additionally, we provide baseline accuracy under the same conditions for comprehensive comparison and reference. Through extensive experiments, our method achieves competitive performance and demonstrates its superiority in solving imbalanced distribution experimental configurations.