Abstract: Aiming at the phenomenon that the redundant and poorly performing base learners in the Bagging algorithm increase the consumption of diagnostic procedures, a SelBagging selective integrated learning algorithm is proposed, which measures the diversity of the integrated system by calculating the Kohavi-Wolpert variance of the subset of the base learner. According to the diversity of the integrated system and the classification performance of the base learner, a subset of the base learners is selected with large differences and good classification performance for integration. By modeling the processed engine fingerprint map fault identification data, the results show that the SelBagging algorithm can effectively improve the classification accuracy, and has better and more stable classification results than the traditional Bagging algorithm. Finally, the actual fault case verifies that the SelBagging fault diagnosis model can be used for fault diagnosis of aero-engines.

Key words: aero-engine, Bagging algorithm, selective integration, Kohavi-Wolpert variance