Novel Transfer Learning Based on Support Vector Data Description for Aeroengine Fault Detection

Abstract

Fault detection is an important part of aeroengine health management. Intelligent fault detection methods represented by machine learning have been widely studied. However, most studies assume that training and test data follow the same distribution, which is unrealistic. Due to the degradation of engine performance or change of engine operating environment, the historical operation data of aeroengines are different from the current operation data of the engine. If the engine history operation data are directly used to train a fault detection model, the fault detection of the current engine may lead to low efficiency and affect the reliability of fault detection. In order to overcome this problem, transfer learning is introduced into aircraft engine fault detection in this paper. This paper combines transfer learning with support vector data description (SVDD), a common fault detection algorithm, and proposes SVDD-based transfer learning (SVDD-TL). This algorithm takes the spherical center of the SVDD as the knowledge structure to transfer from the source domain to the target domain, which can improve the detection accuracy of the model in the target domain. A fault detection experiment for an aeroengine was designed. Single and mixed fault data were used in the experiment, and the variation of fault data quantity was considered. Experimental results showed that the proposed method can improve the fault detection accuracy of the model in the target domain and still have good detection performance when the amount of fault data changes.