Interpretable diagnosis of aircraft actuator imbalance faults based on stacking integration strategy

Faults in the actuators of aircraft flight control systems directly impact flight safety. To address the challenges of unbalanced fault diagnosis and the complex interpretation of diagnostic results in the intelligent fault diagnosis of actuators in engineering practice, this paper proposes a fault diagnosis strategy based on stacking ensemble learning and the Gaussian Naive Bayes (GNB) method. Feature engineering methods, such as feature correlation analysis, are employed to select features from the collected data of the actuator sensors, and the optimal primary learner of the stacking integrated learner is determined through setup comparison verification. The experimental results show that the proposed diagnostic strategy has significant superiority in both balanced and unbalanced data situations, especially in unbalanced data situations. The fault diagnosis results are comprehensively interpreted by analyzing the prior and posterior probabilities of GNB; the features are evaluated in depth based on the two metrics of mean and variance, and mutual corroboration with the results of feature correlation analysis is achieved. The proposed strategy for fault diagnosis of aircraft flight control system actuators provides essential theoretical and methodological support for the much-needed unbalanced fault diagnosis and explainable fault diagnosis techniques in engineering practice.