基于对抗域增强领域泛化的轴承剩余使用寿命预测  

Remaining Useful Life Prediction for Bearings Based on Adversarial Domain-Augmented Domain Generalization

作　　者：柴立平 何昊昱[3] 段石誉 刘璇 陈为伟 CHAI Liping;HE Haoyu;DUAN Shiyu;LIU Xuan;CHEN Weiwei(Luoyang Bearing Research Institute Co.,Ltd.,Luoyang 471039,China;Henan Key Laboratory of High Performance Bearing Technology,Luoyang 471039,China;School of Mechanical Engineering,,Hefei University of Technology,Hefei 230009,China;School of Computer Science and Information Engineering,Hefei University of Technology,Hefei 230009,China;Shanghai Aerospace Control Technology Research Institute,Shanghai 201109,China)

机构地区：[1]洛阳轴承研究所有限公司,河南洛阳471039 [2]河南省高性能轴承技术重点实验室,河南洛阳471039 [3]合肥工业大学机械工程学院,合肥230009 [4]合肥工业大学计算机与信息学院,合肥230009 [5]上海航天控制技术研究院,上海201109

出　　处：《轴承》2025年第5期86-95,共10页Bearing

基　　金：河南省高性能轴承技术重点实验室开放基金资助项目(ZYSKF202307);国家自然科学基金资助项目(52375089)。

摘　　要：针对现有基于深度学习的轴承剩余使用寿命预测方法在测试数据是未知轴承或未知工况下使用训练轴承数据得到的模型由于分布外泛化能力的限制性能会严重下降的问题,提出了一种基于对抗域增强的领域泛化方法,包括回归预测模块和对抗训练域增强模块。回归预测模块中设计了卷积增强自注意力的门控神经网络用于捕捉时序数据的特征以输出预测值,对抗训练域增强模块中设计了基于长短时记忆网络的生成对抗网络用于数据增强,还引入了任务损失函数和域增强损失函数分别用于指导模型在预测准确性与域增强能力之间的平衡,损失函数的联合作用使模型能够更好地处理未知轴承和未知工况的数据并提高整体性能,实现模型对未知轴承和未知工况的性能泛化。在PHM 2012 Challenge数据集和XJTU-SY数据集上的试验结果证明了该方法在寿命预测性能上的优越性。与其他方法的对比试验结果显示,在不使用复合故障样本的情况下,该方法预测轴承剩余使用寿命的均方误差可低至0.0004,平均绝对误差可低至0.0050。The test data is collected from unknown bearings or under unseen operating conditions,and the models obtained from training data of bearings suffer from severe performance degradation due to limited out-of-distribution generalization capability.A domain generalization method is proposed based on adversarial domain augmentation,which includes regression prediction module and adversarial training domain augmentation module.In regression prediction module,a convolutional augmented self-attention gated neural network is designed to capture the features of time series data and output the predicted value.In adversarial training domain augmentation module,a generative adversarial network based on long short-term memory network is designed for data augmentation.In addition,the task loss function and domain augmentation loss function are introduced to guide the balance between prediction accuracy and domain augmentation capability of the model respectively.The combined effect of these loss functions enables the model to process the data from unknown bearings and under unseen operating conditions better and improve the overall performance,thus realizing the performance generalization of the model for unknown bearings and unseen operating conditions.The test results on PHM2012 Challenge dataset and XJTU-SY dataset demonstrate the superiority of the proposed method in life prediction performance.The comparative test results with other methods show that the mean square error and mean absolute error of the proposed method in predicting the remaining useful life of bearings can be as low as 0.0004 and 0.0050 respectively without using compound fault samples.

关 键 词：滚动轴承 剩余寿命预测 深度学习 测试数据 损失函数 均方误差 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP391[自动化与计算机技术—计算机应用技术]