摘要
行星齿轮箱的工作环境复杂、转速和载荷多变,而常规故障诊断算法模型仅适用于单一工况,在多工况下诊断效果不佳。针对此问题,提出一种结合深度学习和迁移学习的故障诊断算法模型,用于多工况下行星齿轮箱的故障诊断。建立卷积神经网络(CNN)和长短期记忆网络(LSTM)为主体的模型框架,融入通道注意力机制(CAM)和自注意力机制(Self-Attention)完成模型的搭建。将传感器采集的数据划分为源域和目标域,采用数据充足的源域样本训练模型并保存,采用少量带标记的目标域样本对预训练模型进行微调。为了防止样本数量过少造成模型过拟合、分类效果不佳等问题,通过生成一定长度的随机噪声数据替换微调样本上部分数据点,达到扩充微调样本数量的目的。结果表明:所提方法能够较好地完成变工况下行星齿轮箱的故障诊断任务,在定工况下的诊断精度接近100%,在跨工况下的故障诊断精度均达99%以上,证明了采用所提方法时模型整体更加稳定,为齿轮箱的故障诊断提供了新的方法和思路。
The working environment of planetary gearbox is complex,the speed and load are variable,and the conventional fault diagnosis algorithm model is only suitable for a single working condition,and diagnostic performance is poor under multiple working conditions.In order to solve this problem,a fault diagnosis algorithm combining deep learning and transfer learning was proposed for fault diagnosis of planetary gearboxes under multiple operating conditions.Convolutional neural network(CNN)and long short-term memory network(LSTM)were established as the main model frameworks,and channel attention mechanism(CAM)and self-attention mechanism were integrated to complete the model construction.The data collected by sensors were divided into source domain and target domain,the model was trained and saved by using source domain samples with sufficient data,and a small number of labeled target domain samples were used to fine-tune the pre-trained model.In order to prevent problems such as overfitting of the model and poor classification effect caused by too small sample size,some data points on the fine-tuning samples were replaced by generating random noise data with a certain length to expand the number of fine-tuning samples.The results show that the proposed method can effectively complete the fault diagnosis task of planetary gearboxes under variable working conditions,and the diagnostic accuracy is close to 100%under fixed working conditions,and the fault diagnosis accuracy is more than 99%under cross-working conditions,which proves that the model is relatively stable when the proposed method is adopted,providing new methods and ideas for fault diagnosis of gearboxes.
作者
陈超
许琦
宋正华
高虎
CHEN Chao;XU Qi;SONG Zhenghua;GAO Hu(School of Mechanical Engineering,Yancheng Institute of Technology,Yancheng Jiangsu 224000,China;Jiangsu Kunlun Interconnection New Energy Group Co.,Ltd.,Yancheng Jiangsu 224001,China)
出处
《机床与液压》
北大核心
2025年第10期40-49,共10页
Machine Tool & Hydraulics
关键词
行星齿轮箱
深度学习
迁移学习
样本扩充
planetary gearbox
deep learning
transfer learning
sample expansion
