Session: 01-19 Aero-engine Control and Diagnostics II

Paper Number: 129080

129080 - Gas Path Fault Diagnosis Method Under Small Samples With Inter-Class Imbalance 

Gas path fault diagnosis plays a critical role in the security guarantee and maintenance of aero-engines. At present, most of the data-based diagnosis methods have high requirements for fault sample sets. However, fault samples are often expensive and difficult to obtain in the actual process. Even if limited faulty samples are obtained, the imbalance between the inter-class is another important problem. Therefore, we expect to adopt a Variational Auto-Encoder method based on transfer learning (VAE-TL) to achieve the production technology of fault samples and optimize the impact of inter-class imbalance on the performance of fault diagnosis model. VAE obtains its deep characteristics by pre-training the normal operating samples. Combined with transfer learning, the VAE is retrained using a small number of fault samples, i.e., some of the VAE weights are fine-tuned. Thereby, the operating state features in the source domain are combined with the fault features in the target domain, so that the VAE-TL has the ability to act as a generator of fault samples. The subsequent fault diagnosis strategy is based on the fault features and the generated fault data to implement the fault detection and isolation process. We verify the effectiveness and superiority of the method proposed in this paper through the simulation process, and the experimental results show that the method proposed in this paper can significantly improve the fault diagnosis results, weaken the influence of the noise within the data, and have excellent fault diagnosis capability under small samples with inter-class imbalance situation.

Presenting Author: Kun Wang Tsinghua University

Presenting Author Biography: B.S. and M.S. degrees from Northwestern Polytechnical University, School of Power and Energy
Research interests include aero-engine health management

Authors:

Kun Wang Tsinghua University
Ai He Tsinghua University
Yuan Liu AECC Hunan Aviation Powerplant Research Institute
Tingyu Xiao Tsinghua University
Zhixiong Chen AECC Hunan Aviation Powerplant Research Institute
Zhongzhi Hu Tsinghua University