Research on civil aircraft fault diagnosis method based on selective integration

Abstract

Based on the research of the existing methods, this paper applies the performance parameters of the base learner and proposes the overall performance metrics of the base learner subset based on the diversity measurement method, the diagnostic accuracy of the base learner, and the running time. The performance metric as the fitness function of the genetic algorithm establishes the selection method of the subset of the base learner; the deep neural network is used to learn the complex nonlinear relationship between the base learners and a selective integration framework based on the stacking method is proposed to improve the overall diagnostic performance of integration model. Then, the data from engine operation to failure are used to verify the method performance proposed in this paper. The comparison results find that the method has a higher fault diagnosis accuracy and robustness. Finally, the method is streamlined to make it more applicable and suitable for engineering problems.