【摘要】 轴承是机械装备中的重要零件，其剩余使用寿命的准确预测是减少维护和避免事故的急切需求。提出一种基于数据增强的可分离卷积神经网络轴承寿命预测方法。首先，使用无监督对抗生成方法进行原始样本的数据增强。然后，针对轴承寿命预测模型拟合能力弱的问题，提出一种新的基于改进的可分离卷积神经网络的预测方法，此方法代替普通卷积进行特征提取，可大幅度减少计算量，从而可在相同计算量的条件下挖掘更深层次的数据特征。实验结果表明，所提方法均优于常见的深度学习方法，并在真实数据集上验证了所提方法的有效性。更多还原

【Abstract】 Bearings are important parts in mechanical equipment, and accurate prediction of their remaining service life is an urgent requirement to reduce maintenance and avoid accidents. A separable convolutional neural network bearing life prediction method based on data enhancement is proposed. First, data enhancement of the original samples is performed using an unsupervised adversarial generation method. Then, a new prediction method based on improved separable convolutional neural network is proposed to address the problem of weak fitting ability of bearing life prediction model, which replaces the ordinary convolution for feature extraction, and can significantly reduce the computation amount, thus mining deeper data features under the condition of the same computation amount. The experimental results show that all the methods outperform the common deep learning methods, and the effectiveness of the proposed method is verified on real datasets.更多还原