Inter-shaft Bearing Fault Diagnosis Based on Aero-engine System: A Benchmarking Dataset Study

Lei Hou; Haiming Yi; Yuhong Jin; Min Gui; Lianzheng Sui; Jianwei Zhang; Yushu Chen

doi:10.37965/jdmd.2023.314

Authors

Lei Hou School of Astronautics, Harbin Institute of Technology, Harbin 150001, China https://orcid.org/0000-0003-0271-7323
Haiming Yi School of Astronautics, Harbin Institute of Technology, Harbin 150001, China https://orcid.org/0009-0000-5806-0653
Yuhong Jin School of Astronautics, Harbin Institute of Technology, Harbin 150001, China
Min Gui Factory of Xiang Yang Hang Tai Power Machinery, Xiangyang 441000, China
Lianzheng Sui Factory of Xiang Yang Hang Tai Power Machinery, Xiangyang 441000, China
Jianwei Zhang Factory of Xiang Yang Hang Tai Power Machinery, Xiangyang 441000, China https://orcid.org/0009-0004-3907-5197
Yushu Chen School of Astronautics, Harbin Institute of Technology, Harbin 150001, China

DOI:

https://doi.org/10.37965/jdmd.2023.314

Keywords:

inter-shaft bearing, fault diagnosis, aero-engine test, dataset

Abstract

In this paper, the aero-engine test with inter-shaft bearing fault is carried out, and a dataset is proposed for the first time based on the vibration signal of rotors and casings. First, a test rig based on a real aero-engine is established, driven by motors and equipped with a lubricating system. Then, the aero-engine is disassembled and assembled following the specification process, and the inter-shaft bearing with artificial fault is replaced. Next, the aero-engine test is conducted at 28 groups of high and low pressure speeds. Six measuring points are arranged, including two displacement sensors to test the displacement vibration signals of the low pressure rotor and four acceleration sensors to test the acceleration vibration signals of the casing. The test results are integrated into an inter-shaft bearing fault dataset. Finally, based on the dataset in this paper, frequency spectrum, envelope spectrum, CNN, LSTM and TST are used for fault diagnosis, and the results are compared with those of CWRU and XJTU datasets. The results show that the characteristic fault frequency cannot be found directly in the spectrum and envelope spectrum corresponding to this paper's dataset but in CWRU and XJTU datasets. Using CNN, LSTM and TST for fault diagnosis of the dataset in this paper, the accuracy is 83.13%, 85.41% and 71.07%, respectively, much lower than the diagnosis results of CWRU and XJTU datasets. It can be seen that the dataset in this paper is closer to the actual fault diagnosis situation and is a more challenging dataset. This dataset provides a new benchmark for the validation of fault diagnosis methods. Mendeley data: https://github.com/HouLeiHIT/HIT-dataset.

Conflict of Interest Statement

The authors declare no conflicts of interest.

Corrections

https://ojs.istp-press.com/dmd/article/view/523