Journals / IASC / Vol.29, No.3

Research Article

BEST PAPER 2021

Surge Fault Detection of Aeroengines Based on Fusion Neural Network

Desheng Zheng1, Xiaolan Tang1,*, Xinlong Wu1, Kexin Zhang1, Chao Lu2, Lulu Tian3
1 School of Computer Science, Southwest Petroleum University, Chengdu, 610500, China
2 AECC Sichuan Gas Turbine Establishment, Mianyang, 621700, China
3 Department of Computer Science, Brunel University London, Middlesex, UB8 3PH, United Kingdom
* Corresponding Author: Xiaolan Tang. Email:

Abstract

Aeroengine surge fault is one of the main causes of flight accidents. When a surge occurs, it is hard to detect it in time and take anti-surge measures correctly. Recently, people have been applying detection methods based on mathematical models and expert knowledge. Due to difficult modeling and limited failure-mode coverage of these methods, early surge detection cannot be achieved. To address these problems, firstly, this paper introduced the data of six main sensors related to the aeroengine surge fault, such as, total pressure at compressor (high pressure rotor) outlet (Pt3), high pressure compressor rotor speed (N2), power level angle (PLA), etc. Secondly, aiming at preprocessing of data sets, this paper proposed a data standardization preprocessing algorithm based on batch sliding window (DSPABSW) to build a training set, validation set and test set. Thirdly, aeroengine surge fault detection fusion neural network (ASFDFNN) was provided and optimized to improve the detection accuracy of aeroengine surge faults. Finally, the experimental results showed that the model achieved 95.7%, 93.6% and 94.7% in precision rate, recall rate and F1_Score respectively and consequently it can detect the aeroengine surge fault 260 ms in advance.

Keywords

Sliding window; time series prediction; aeroengine surge fault detection; fusion neural network
  • 366

    View

  • 310

    Download

  • 100

    Like

Share Link

WeChat scan