English

Journal of Chinese Agricultural Mechanization

Journal of Chinese Agricultural Mechanization ›› 2024, Vol. 45 ›› Issue (5): 140-146.DOI: 10.13733/j.jcam.issn.2095-5553.2024.05.022

Previous Articles     Next Articles

Fault diagnosis of agricultural machinery bearing based on hierarchical feature attention decoupling

Qinggui1, Wu Kai2, Zhou Hongbin3   

  • Online:2024-05-15 Published:2024-05-22

基于分层特征注意力解耦的农机轴承故障诊断

邢清桂1,吴凯2,周洪斌3   

  1. 邢清桂1,吴凯2,周洪斌3
  • 基金资助:
    国家自然科学基金(51575498)

Abstract: In view of the timevarying working conditions in the actual operation for agricultural machinery equipment, a fault diagnosis algorithm of agricultural  machinery bearing based on hierarchical feature attention decoupling was proposed. Firstly, a Transformer network improved by Long ShortTerm Memory (LSTM) neural network was used as the backbone network, and a hierarchical feature set of agricultural machinery bearing fault data was constructed according to the Multihead mechanism of Transformer. Then, a crossattention mechanism was employed to explore the correlations between different layers of features, and enhance the expression ability of agricultural machinery bearing fault features. Finally, by employing the multilabel diagnosis of agricultural  machinery bearing faults, the mixed features were decoupled into multiple independent sets of bearing fault features. The decoupled features were used to predict corresponding labels, and achieve the diagnosis of various types of agricultural machinery bearing faults. The experimental results showed that the proposed model can achieve an average recognition accuracy of 96.58% and can diagnose multiple types of bearing faults in a finegrained manner.

Key words: agricultural machinery bearings, fault diagnosis, hierarchical feature, crossattention, feature decoupling

摘要: 鉴于农机设备实际运行中的工况具有时变性,提出一种基于分层特征注意力解耦的农机轴承故障诊断算法。利用长短时记忆神经网络改进的Transformer网络作为主干网络,并按照Transformer的Multihead机制构造农机轴承故障数据的分层特征集;利用交叉注意力机制挖掘不同层特征间的关联关系,强化农机轴承故障特征的表达能力;借助农机轴承故障诊断多标签将混合特征解耦为多个独立的轴承故障特征集,并利用解耦后的特征预测对应的标签,实现待测农机轴承故障类型的诊断。结果表明,所提出的模型可以实现平均96.58%的识别精度,并且可以细粒度地对多种轴承故障进行诊断。

关键词: 农机轴承, 故障诊断, 分层特征, 交叉注意力, 特征解耦

CLC Number: