Abstract: Rolling bearings are widely used in all kinds of rotating machinery,and their health condition is very important for the normal operation of equipment.Mastering the Remaining Useful Life (RUL) of rolling bearings can better ensure the security and efficiency of production activities.The present deep learning-based RUL prediction methods of bearings are all devoted to the overall trend features,while ignoring the interdependence features during each rotation period.To solve this problem,a bearing RUL prediction network considering rotation period which called two-channel network model was proposed.The Convolution Neural Network (CNN) and the Bi-directional Long-Short Term Memory Network (BiLSTM) were used to extract the overall trend features of bearing vibration data,and attention mechanism was attached to enhance the feature extraction ability of the model.A skip recurrent neural network component based on rotation period was used to capture the interdependent pattern of data during each rotation period.The effectiveness of the proposed network was verified by the data of accelerated degradation test of rolling bearings,and comparing experiments with some existing improved algorithms.The result showed that the prediction performance of the algorithm was outstanding.

Key words: remaining useful life prediction, deep learning, bi-directional long-short term memory network, squeeze and excitation operations, rolling bearings

摘要： 滚动轴承广泛应用于各类旋转机械设备,其健康状况对机器设备的正常运行至关重要,掌握其剩余使用寿命(RUL)可以更好地保证生产活动安全有效的进行。目前,基于深度学习的轴承RUL预测方法均致力于对整体趋势特征的把控,而忽略了对各转动周期间相互依赖特征的挖掘。针对这一问题,提出一种考虑转动周期的轴承RUL预测网络——双通道网络模型。该预测网络使用卷积神经网络(CNN)和双向长短期记忆网络(BiLSTM)来提取轴承振动数据的整体趋势特征,并引入注意力机制来增强模型的特征提取能力。利用基于转动周期的跳越循环神经网络组件来捕捉各转动周期之间的相互依赖模式。通过滚动轴承加速退化实验的数据,验证了所提网络的有效性,并与一些智能算法进行了对比实验,预测精度表现优异。

关键词: 剩余使用寿命预测, 深度学习, 双向长短期记忆网络, 压缩激励机制, 滚动轴承