航空学报 > 2022, Vol. 43 Issue (9): 625056-625056   doi: 10.7527/S1000-6893.2021.25056

航空发动机轴承故障结构化贝叶斯稀疏表示

张烁, 刘治汶   

  1. 电子科技大学 自动化工程学院, 成都 611731
  • 收稿日期:2020-12-04 修回日期:2020-12-30 出版日期:2022-09-15 发布日期:2021-04-29
  • 通讯作者: 刘治汶,E-mail:lzw1682007@126.com E-mail:lzw1682007@126.com
  • 基金资助:
    国家自然科学基金(52075080, U1733107);航空科学基金(20173319003)

Structured Bayesian sparse representation of aero-engine bearing failures

ZHANG Shuo, LIU Zhiwen   

  1. School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
  • Received:2020-12-04 Revised:2020-12-30 Online:2022-09-15 Published:2021-04-29
  • Supported by:
    National Natural Science Foundation of China (52075080, U1733107); Aeronautical Science Foundation of China (20173319003)

摘要: 航空发动机轴承振动信号中与故障关联的瞬时冲击成分在时频变换域上不仅具有稀疏性, 还具有某些结构特征, 而传统的以正交匹配追踪(OMP) 算法为代表的贪婪类及其改进重构方法, 通常仅利用了信号整体的稀疏性, 未考虑结构性干扰可能造成的影响, 导致算法求解效率较低。针对这一问题, 提出了一种参数优化字典的结构化贝叶斯稀疏表示方法。首先, 在OMP算法基础上, 基于贝叶斯概率模型, 研究了一种能够促进稀疏重构效果的结构化贝叶斯正交匹配追踪(SBOMP) 稀疏表示模型, 实现对信号的稀疏表示求解。其次, 针对轴承故障振动信号的特性, 构建能更好的匹配分析信号的时频冲击原子库, 降低了字典的冗余程度, 并将灰狼优化算法(GWO) 引入到基于时频冲击字典的SBOMP模型中, 为SBOMP模型提供高效的原子选取策略, 降低了稀疏模型的复杂度。仿真与实验结果表明: 所提方法能够更有效降低背景噪声和杂质频率的干扰, 验证了所提方法对航空发动机轴承故障特征提取的有效性和适用性。

关键词: 故障诊断, 结构化贝叶斯正交匹配追踪, 灰狼优化算法, 时频冲击字典, 航空发动机轴承

Abstract: The instantaneous impact component associated with the fault in the aero-engine bearing vibration signal has not only sparseness in the time-frequency transform domain but also specific structural characteristics. Traditional greedy algorithms and their improved reconstruction methods, represented by the Orthogonal Matching Pursuit (OMP) algorithm, usually only use the sparseness of the signal as a whole, but not the possible impact of structural interference, resulting in lower solution efficiency. This paper proposes a structured Bayesian sparse representation method for the parameter optimization dictionary. First, on the basis of the OMP algorithm, a Structured Bayesian Orthogonal Matching Pursuit (SBOMP) sparse representation model is developed based on the Bayesian probability model to promote the sparse reconstruction effect, so as to obtain the sparse representation of signals. Second, considering the characteristics of bearing fault vibration signals, a time-frequency impulse atomic library that can better match the analysis signals is constructed to reduce redundancy of the dictionary. The Grey Wolf Optimization (GWO) algorithm is introduced into the SBOMP model based on the time-frequency impulse dictionary to provide an efficient atom selection strategy for the SBOMP model and reduce complexity of the sparse model. Simulation and experimental results show that the method proposed in this paper can more effectively reduce interference of background noise and impurity frequencies, which verifies the effectiveness and applicability of the method proposed for extraction of aero-engine bearing fault features.

Key words: fault diagnosis, structured Bayesian orthogonal matching tracking, gray wolf optimization algorithm, time-frequency impact dictionary, aero-engine bearing

中图分类号: