Fault diagnosis of aircraft fuel pump based on transfer learning

Sometimes there could be sufficient data in used aircraft fuel pump. A few data in a specific new pump, which low accuracy of diagnosis, are obtained. In this paper, a fault diagnosis method of centrifugal aircraft fuel pump based on migration learning is proposed. The features from the fault data of similar pumps, which are used as auxiliary data set, are extracted. Then, the training data set, which is composed of auxiliary data and a few amount of diagnostic target data, is established and is trained by TrAdaboost transfer learning algorithm. Finally, compared with traditional machine learning, it shows that the transfer learning has obvious diagnostic advantages in the situation of insufficient diagnostic target data.