欢迎访问《兵工学报》官方网站,今天是

兵工学报 ›› 2020, Vol. 41 ›› Issue (11): 2170-2178.doi: 10.3969/j.issn.1000-1093.2020.11.003

• 论文 • 上一篇    下一篇

基于条件深度循环生成对抗网络和动作探索的行星轮轴承剩余寿命预测

于军1,2,3, 刘可1,3, 郭帅4, 于广滨1, 郭振宇2   

  1. (1.哈尔滨理工大学 先进制造智能化技术教育部重点实验室, 黑龙江 哈尔滨 150080;2.矿冶过程自动控制技术国家重点实验室, 北京 100089;3.哈尔滨理工大学 自动化学院, 黑龙江 哈尔滨 150080;4.中国船舶重工集团公司第703研究所 蒸汽轮机事业部, 黑龙江 哈尔滨 150078)
  • 上线日期:2020-12-04
  • 作者简介:于军(1984—),男,讲师,硕士生导师,博士。E-mail: shengda1302@126.com
  • 基金资助:
    国家重点研发计划项目(2019YFB2006400);矿冶过程自动控制技术国家重点实验室开放基金项目(BGRIMM-2020-06); 国家自然科学基金项目(61806060);黑龙江省科技重大专项项目(2019ZX03A02);黑龙江省杰出青年基金项目(JC2014020);哈尔滨市杰出青年人才基金项目(2017RAYXJ011)

Remaining Useful Life Prediction of Planet Bearings Based on Conditional Deep Recurrent Generative Adversarial Network and Action Discovery

YU Jun1,2,3, LIU Ke1,3, GUO Shuai4, YU Guangbin1, GUO Zhenyu2   

  1. (1.Key Laboratory of Advanced Manufacturing and Intelligent Technology, Ministry of Education, Harbin University of Science and Technology, Harbin 150080, Heilongjiang, China; 2.State Key Laboratory of Process Automation in Mining & Metallurgy, Beijing 100089, China; 3.School of Automation, Harbin University of Science and Technology, Harbin 150080, Heilongjiang, China; 4.Division of Steam Turbine, The No. 703 Research Institute, China State Shipbuilding Corporation Limited, Harbin 150078, Heilongjiang, China)
  • Online:2020-12-04

摘要: 为解决小样本和变工况下行星轮轴承剩余寿命预测准确率低的问题,提出一种基于条件深度循环生成对抗网络(C-DRGAN)和动作探索(AD)的行星轮轴承剩余寿命预测方法。将门控循环单元神经网络与条件生成对抗网络相结合,构建C-DRGAN,从非静态和非线性信号中提取故障特征,实现小样本和变工况下行星轮轴承的剩余寿命预测;采用基于AD的训练算法训练C-DRGAN,提高收敛速度,降低训练时间;根据训练后的C-DRGAN,利用多元线性回归分类器预测测试样本中行星轮轴承的剩余寿命。通过行星轮轴承加速疲劳寿命试验验证该方法的有效性。结果表明,该方法具有较强的非静态和非线性信号处理能力,并能在小样本情况下取得出色的行星轮轴承剩余寿命预测效果。

关键词: 行星轮轴承, 剩余寿命预测, 门控循环单元神经网络, 条件生成对抗网络, 动作探索, 小样本

Abstract: In order to address the problem about the remaining useful life (RUL) prediction of planet bearings under small samples and varying operating conditions, a RUL prediction method of planet bearings based on conditional deep recurrent generative adversarial network (C-DRGAN) and action discovery (AD) is presented. The gated recurrent unit neural network is integrated with conditional generative adversarial network to construct the C-DRGAN, which extracts the fault features from the nonlinear and non-stationary signals so as to realize the RUL prediction of planet bearings under small samples and varying operating conditions conditions. Then the training algorithm based on AD is employed to train the C-DRGAN to enhance the convergence speed and reduce the training time. Finally, according to the C-DRGAN after training, a multiple linear regression classifier is employed to predict the RUL of planet bearings in test samples. The effectiveness of the proposed method is validated through an accelerated fatigue life experiment of planet bearings. The results show that the proposed method processes strong processing adaptability to the nonlinear and non-stationary signals and obtains the excellent performance in the case of small samples.

Key words: planetbearing, remainingusefullifeprediction, gatedrecurrentunitneuralnetwork, conditionalgenerativeadversarialnetwork, actiondiscovery, smallsample

中图分类号: