摘要
传统智能数据驱动机械剩余寿命预测存在两个问题:(1)预测起始时间的确定精度不高(2)实体之间数据分布的显著差异,为此提出一种基于数据对齐的深度对抗性神经网络机械寿命预测。利用生成性对抗性神经网络学习机器健康状态下的数据分布,建立系统异常的有效指标,确定预测的预测起始时间。将学习到的特征进一步用于剩余寿命估计,并对不同实体的数据比对引入对抗训练。在学习子空间中通过数据对齐策略来提取实体不变特征,弥补数据分布差异,提高了数据驱动程序方法的泛化能力。最后利用两组加速滚动轴承退化测试数据集验证了所提出的预测方法,结果表明提出的方法能够有效提升预测精度,并且有效解决了数据差异问题。
There were two problems in the traditional intelligent data-driven mechanical residual life prediction:(1)the low accuracy of the prediction start time;(2)the significant difference of data distribution among entities. Therefore,a deep adversarial neural network based on data alignment was proposed for mechanical life prediction. Using the deep adversarial neural networks to learn the data distribution of machine health state,the effective index of system anomaly is established,and the prediction start time was determined. The learned features were further used for residual life estimation,and the data comparison of different entities was introduced into confrontation training. In the learning subspace,the data alignment strategy was used to extract entity invariant features to make up for the difference of data distribution and improve the generalization ability of data driver method. Finally,two sets of accelerated rolling bearing degradation test data sets were used to verify the proposed prediction method. The results show that the proposed method can effectively improve the prediction accuracy and effectively solve the problem of data difference.
作者
吴涛
马驰
WU Tao;MA Chi(Ordos branch of China Shenhua International Engineering Co.,Ltd.,Ordos,Inner Mongolia Ordos 017000,China;School of Mechanical and Electrical Engineering,China University of Mining and Technology,Jiangsu Xuzhou 221008,China)
出处
《机械设计与制造》
北大核心
2023年第2期63-70,共8页
Machinery Design & Manufacture
基金
国家自然科学基金项目(51975569)。
关键词
剩余寿命预测
深度对抗性神经网络
数据差异
预测起始时间
Residual Life Prediction
Deep Adversarial Neural Networks
Data Difference
Prediction Start Time
