航空学报 > 2022, Vol. 43 Issue (7): 325403-325403   doi: 10.7527/S1000-6893.2021.25403

基于CDBN与BiLSTM的多元退化设备剩余寿命预测

牟含笑1, 郑建飞1, 胡昌华1, 赵瑞星2, 董青1   

  1. 1. 火箭军工程大学 导弹工程学院, 西安 710025;
    2. 火箭军装备部驻西安地区第三军事代表室, 西安 710100
  • 收稿日期:2021-02-18 修回日期:2021-05-11 发布日期:2021-05-10
  • 通讯作者: 郑建飞,E-mail:zjf302@126.com E-mail:zjf302@126.com
  • 基金资助:
    国家自然科学基金(61922089,61773386,61833016);陕西省自然科学基金(2020JM-360)

Remaining useful life prediction of multivariate degradation equipment based on CDBN and BiLSTM

MU Hanxiao1, ZHENG Jianfei1, HU Changhua1, ZHAO Ruixing2, DONG Qing1   

  1. 1. College of Missile Engineering, Rocket Force University of Engineering, Xi'an 710025, China;
    2. The Third Military Representative Office of the Rocket Force Equipment Department in Xi'an, Xi'an 710100, China
  • Received:2021-02-18 Revised:2021-05-11 Published:2021-05-10
  • Supported by:
    National Natural Science Foundation of China (61922089, 61773386, 61833016); Natural Science Foundation of Shaanxi Province (2020JM-360)

摘要: 基于多传感器对复杂工业设备的多元健康状态进行监测,进而实现设备更全面准确的性能评估、剩余寿命预测与健康管理已逐渐推广应用。针对一类监测数据呈现大规模、非线性、高维化等特点的多元退化设备,提出了一种基于连续深度置信网络(CDBN)与双向长短期记忆(BiLSTM)网络的剩余寿命预测方法。首先,通过CDBN对监测到的性能退化数据进行分析,提取出反映多元退化设备隐含深层次特征的健康指标;然后,根据构造的健康指标,利用BiLSTM网络挖掘其时序信息和退化趋势,预测多元退化设备的剩余寿命;最后,利用蒙特卡洛仿真技术得到剩余寿命的区间估计,并通过商用模块化航空推进系统数据集验证所提方法的有效性和先进性。结果表明:所提方法能够有效提高此类设备的剩余寿命预测准确度,具有潜在的应用价值。

关键词: 多元退化设备, 剩余寿命预测, 健康指标, 连续深度置信网络(CDBN), 双向长短期记忆(BiLSTM)网络

Abstract: Monitoring the multiple health status of complex industrial equipment based on multiple sensors has been gradually applied to achieve more comprehensive and accurate performance evaluation, remaining useful life prediction and health management of the equipment. For a type of multivariate degradation equipment with the monitoring data of large-scale, non-linear, high-dimensional characteristics, a new method based on the Continuous Deep Belief Network (CDBN) and Bidirectional Long Short-Term Memory (BiLSTM) network to predict the remaining useful life of the equipment is proposed. First, the degradation data obtained through the CDBN are analyzed, and the health indexes that reflect the hidden deep-level features of the multivariate degradation equipment are extracted. Then, according to the constructed health indexes, the BiLSTM network is used to mine the timing information and degradation trends, so as to predict the remaining useful life. Finally, the Monte Carlo simulation technology is used to obtain the interval estimation of the remaining useful life. The validity the proposed method is verified through the commercial modular aviation propulsion system dataset. The results show that the method proposed can effectively improve the accuracy of the remaining useful life prediction of this type of equipment, and has potential applicability.

Key words: multivariate degradation equipment, remaining useful life prediction, health index, Continuous Deep Belief Network (CDBN), Bidirectional Long Short-Term Memory(BiLSTM) networkhttp

中图分类号: