航空学报 > 2021, Vol. 42 Issue (3): 423876-423876   doi: 10.7527/S1000-6893.2020.23876

基于LSTM和CNN的高速柱塞泵故障诊断

魏晓良, 潮群, 陶建峰, 刘成良, 王立尧   

  1. 上海交通大学 机械系统与振动国家重点实验室, 上海 200240
  • 收稿日期:2020-02-14 修回日期:2020-02-29 发布日期:2020-05-21
  • 通讯作者: 陶建峰 E-mail:jftao@sjtu.edu.cn
  • 基金资助:
    国家重点研发计划基金(2017YFD0700602);中国博士后科学基金(2019M660086)

Cavitation fault diagnosis method for high-speed plunger pumps based on LSTM and CNN

WEI Xiaoliang, CHAO Qun, TAO Jianfeng, LIU Chengliang, WANG Liyao   

  1. State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai 200240, China
  • Received:2020-02-14 Revised:2020-02-29 Published:2020-05-21
  • Supported by:
    National Key R&D Program of China (2017YFD0700602); China Postdoctoral Science Foundation (2019M660086)

摘要: 针对高速轴向柱塞泵容易发生空化,且目前空化故障诊断方法存在依赖手工特征提取、鲁棒性不高的问题,提出了一种基于长短时记忆(LSTM)和一维卷积神经网络(1D-CNN)相结合的空化故障诊断方法。搭建了柱塞泵故障实验台,采集柱塞泵在不同空化等级下的壳体振动信号。利用LSTM和1D-CNN搭建的分类模型对不同进口压力情况下的振动信号进行空化等级识别。实验结果表明:提出的方法能够准确地识别出4类不同的空化等级,准确率高达99.5%,同时在不附加降噪方法的情况下,具有良好的鲁棒性,在0 dB信噪比的情况下,识别准确率高达87.3%。

关键词: 高速轴向柱塞泵, 空化等级识别, 长短时记忆, 卷积神经网络, 故障诊断

Abstract: Aiming at the cavitation of high-speed axial piston pumps and the drawbacks of conventional cavitation diagnosis methods such as relying on manual feature extraction and lack of robustness, this paper proposes a new method based on the combination of Long Short-Term Memory (LSTM) and one-Dimensional Convolutional Neural Network (1D-CNN). A test bench for axial piston pumps is built to collect the vibration signals of the pump housing at different cavitation levels. The classification network constructed by LSTM and CNN is used to identify the cavitation levels based on the vibration signals under different inlet pressure conditions. The experimental results show that the proposed method can accurately identify four different types of cavitation levels and that the accuracy rate can reach 99.5%. Furthermore, it has good robustness without the contribution of a noise reduction method. In the case of 0 dB signal to noise rate, the recognition accuracy is up to 87.3%.

Key words: high-speed axial piston pumps, cavitation intensity identification, long short-term memory (LSTM), convolutional neural network (CNN), fault diagnosis

中图分类号: