A Satellite Fault Diagnosis Method Based on Focus Loss Function and Balance Factor

Fault diagnosis is an effective method to protect satellites in orbit. In recent years, researchers have used data-driven methods to provide some solutions, but most of them have not considered the data imbalance in the satellite fault diagnosis. In response to this problem, this paper analyzes the imbalance problem from two aspects: the imbalance of sample distribution and the imbalance of misclassification cost. Additionally, based on the analysis, this paper proposes an online fault diagnosis method based on focus loss function and balance factor, which can focus the loss on the fault samples in training process. Through experiments, the model greatly improves the detection accuracy of fault samples from 49.5% to 93.2%. At the same time, this method can also provide reference for solving the problem of category imbalance in other industrial field.