摘要
目的实现齿轮箱故障类型的智能识别诊断。方法针对传统故障诊断方法通用性不广、数据依赖强、泛化能力弱并需人工提取特征问题,提出一种基于条件变分自编码器的故障诊断方法。以故障类别概率分布为目标并将振动信号频谱作为条件,通过条件变分自编码器,建立齿轮箱振动信号频谱到对应各故障下的条件概率模型,并通过多层神经网络结合变分推断方法进行训练优化,实现对齿轮箱各类型故障的高精度分类诊断。结果在仅有少量训练数据条件下,实现了准确的故障识别。结论条件变分自编码器在齿轮箱振动信号频谱概率分布建模上具有优异性能,对故障信号数据量的依赖低、泛化能力强,无需人工提取特征。能有效实现齿轮箱故障的智能分类诊断。
The paper aims to realize intelligent fault type diagnosis of gearbox. A fault diagnostic method based on variational auto-encoder with condition(CVAE) was proposed to solve the shortcoming of traditional fault diagnosis methods of poor universality, strong data dependence, weak generalization ability and manual feature extraction demand. High accuracy identification of all kinds of gearbox faults were realized by building a conditional probability model of frequency spectrum of gearbox vibration signal through CVAE with the spectrum of vibration signal as condition, which was optimized by variational inference combined with multi-layer neural network. Accurate fault identification was realized with only a small amount of training data. CVAE has excellent performance in modeling frequency spectrum probability distribution of gearbox vibration signal with low dependence on fault signal data, strong ability of generalization, needlessness of manually extract features and can realize intelligent identification and diagnose of gearbox faults effectively.
作者
王昱
尹爱军
WANG Yu;YIN Ai-jun(State Key Laboratory of Mechanical Transmission,College of Mechanical Engineering,Chongqing University,Chongqing 400044,China)
出处
《装备环境工程》
CAS
2020年第7期64-69,共6页
Equipment Environmental Engineering
基金
重庆市重点研发项目(cstc2018jszx-cyztzxX0032)。