基于蒸馏学习的滚动轴承异常诊断方法
Fault Anomaly Detection Method of Rolling Bearing Based on Distillation Learning
张春青 1毕剑 1高月1
作者信息
- 1. 中国航发沈阳黎明航空发动机有限责任公司,沈阳 110862
- 折叠
摘要
针对航空发动机滚动轴承异常难以诊断的问题,提出了一种基于蒸馏学习的轴承智能诊断方法.首先,为提高计算效率,采用经过压缩后的vision transformer(ViT)作为蒸馏学习的主干特征提取网络;其次,采用滚动轴承振动试验台数据完成蒸馏学习中教师网络的预训练;在蒸馏学习模型的训练过程中,为充分利用教师网络的知识,采用了特征知识约束的损失函数,且利用了频谱变换的数据预处理方法,用于提高模型的诊断精度;最后,在某型多台航空发动机滚动轴承数据上对所提模型进行了充分的对比验证.结果表明,所提方法能够准确地实现滚动轴承运行状态的诊断,且诊断精度能够达到96%以上,充分表明了所提方法具有很高的实用价值.
Abstract
Aiming at the problem that it is difficult to diagnose the fault of aero-engine rolling bearing,an intelligent diagnosis method of bearing based on distillation learning is proposed.Firstly,in order to improve the computational efficiency,the compressed vision transformer(ViT)is used as the backbone feature ex-traction network for distillation learning.Secondly,the data of rolling bearing vibration test bench is used to complete the pre-training of teacher network in distillation learning.in the training process of the distillation learning model,in order to make full use of the knowledge of the teacher network,the loss function of the feature knowledge constraint is adopted,and the data preprocessing method of the spectrum transformation is used to improve the diagnostic accuracy of the model.Finally,the proposed model is fully compared and verified on the fault data of rolling bearings of multiple aero-engines.The results show that the proposed method can accurately realize the intelligent diagnosis of rolling bearing faults,and the diagnosis accuracy can reach more than 96%,which fully shows that the proposed method has high practical value.
关键词
航空发动机/滚动轴承/智能诊断/蒸馏学习Key words
aero-engine/rolling bearing/fault diagnosis/distillation learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据