摘要
利用深度学习方法实现故障诊断过程中存在模型的超参数配置困难问题,为了有效优选模型的超参数集合,提出一种正态分布与莱维飞行(NLF)相结合的优化算法,并以此提出NLF-LSTM滚动轴承故障诊断方法。通过正态分布的集中性与莱维飞行的分散性对超参数集合多次取值,并用长短时记忆深度网络(LSTM)训练以获取最优超参数集合,再利用最优超参数集合构建故障诊断模型,实现对滚动轴承3种部位、10种状态的有效诊断。实验结果表明,与基于SVM等故障诊断方法相比,提出的方法能够有效提高识别滚动轴承故障状况的能力。这也说明基于NLF-LSTM故障诊断方法在降低研究人员配置超参数难度的同时,也使得故障诊断方法更加端对端化与智能化。
In the process of fault diagnose,it is difficult to use the deep learning method to perform model hyperparameter configuration. To effectively optimize the hyperparameter sets of the model, an optimization algorithm combining normal distribution and Levy flight(NLF)is proposed,based on which,a fault diagnosis method of NLF-LSTM(long-and short-term memory)rolling bearing is proposed. The hyperparameter sets are valued many times and subjected to LSTM training to obtain the optimal hyperparameter set by the centralization of normal distribution and the dispersion of Lévy flight. And then,the optimal hyperparameter set is used to construct the fault diagnosis model,so as to realize the effective diagnosis of 3 positions and 10 states of the rolling bearing. The results of experiments show that the proposed method can effectively improve the ability of identifying the fault status of rolling bearings in comparison with other fault diagnosis methods based on SVM. This also indicates that the fault diagnosis method based on NLF-LSTM can not only reduce the difficulty for researchers to configure hyperparameters,but also make the fault diagnosis method end-to-end and more intelligent.
作者
彭成
蒋金元
李凤娟
PENG Cheng;JIANG Jinyuan;LI Fengjuan(School of Computer Science,Hunan University of Technology,Zhuzhou 412007,China;School of Automation,Central South University,Changsha 410083,China)
出处
《现代电子技术》
2022年第1期142-148,共7页
Modern Electronics Technique
基金
国家自然科学基金资助项目(61871432)
湖南省自然科学基金资助项目(2020JJ4275)。