航空学报 > 2024, Vol. 45 Issue (13): 629038-629038   doi: 10.7527/S1000-6893.2023.29038

基于实例迁移学习的跨工况刀具剩余寿命预测

强碧瑶1,2,3, 史恺宁1,2,3(), 任军学1,2,3, 史耀耀1,2,3   

  1. 1.西北工业大学 机电学院,西安 710072
    2.西北工业大学 航空发动机高性能制造工信部重点实验室,西安 710072
    3.西北工业大学 航空发动机先进制造技术教育部工程研究中心,西安 710072
  • 收稿日期:2023-05-24 修回日期:2023-06-12 接受日期:2023-07-30 出版日期:2024-07-15 发布日期:2023-09-06
  • 通讯作者: 史恺宁 E-mail:shikaining@nwpu.edu.cn
  • 基金资助:
    国家自然科学基金(51905442);国家科技重大专项(J2019-VII-0001-0141)

Instance transfer for tool remaining useful life prediction cross working conditions

Biyao QIANG1,2,3, Kaining SHI1,2,3(), Junxue REN1,2,3, Yaoyao SHI1,2,3   

  1. 1.School of Mechanical Engineering,Northwestern Polytechnical University,Xi’an 710072,China
    2.Key Laboratory of High Performance Manufacturing for Aero Engine,Ministry of Industry and Information Technology,Northwestern Polytechnical University,Xi’an 710072,China
    3.Engineering Research Center of Advanced Manufacturing Technology for Aero Engine,Ministry of Education,Northwestern Polytechnical University,Xi’an 710072,China
  • Received:2023-05-24 Revised:2023-06-12 Accepted:2023-07-30 Online:2024-07-15 Published:2023-09-06
  • Contact: Kaining SHI E-mail:shikaining@nwpu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(51905442);National Major Science and Technology Projects of China(J2019-VII-0001-0141)

摘要:

精确可靠的刀具剩余寿命预测可以减少加工过程中刀具过度使用和未充分使用的比率,从而最大限度地提高加工可靠性并降低生产成本。传统机器学习方法在预测刀具剩余寿命时依赖于训练数据和测试数据遵循相同分布的假设,以及广泛的离线测量数据。然而在实际加工过程中,由于加工条件的变化和有限的刀具磨损数据,致使传统方法在跨工况预测刀具剩余寿命时精度较差。针对该问题,提出一种基于实例迁移学习的刀具剩余寿命预测方法,以达到准确预测跨工况条件下刀具剩余寿命的目的。首先,利用迁移学习算法动态调整多个源域中所有实例的权重,充分利用与目标数据高度相关的源域信息来改善模型的泛化能力,从而利用少量目标域数据预测目标工况下的刀具剩余寿命。其次,为了提升迁移学习算法的时间序列预测能力,开发了递归高斯过程回归模型作为基学习器,通过延迟反馈对相邻时刻的刀具剩余寿命进行约束,与此同时还减少了特征准备工作并降低了模型复杂度。结果表明,该方法可以有效提升跨工况下刀具剩余寿命的预测精度,预测效果也证实了方法的稳定性和可靠性。

关键词: 跨工况, 实例迁移学习, 递归高斯过程回归, 时间序列预测, 刀具剩余寿命

Abstract:

Accurate and reliable predictions of tool remaining useful life could reduce the rate of over-utilization and under-utilization of tools during machining, thereby maximizing the machining reliability and reducing production costs. Traditional machine learning methods for tool remaining useful life prediction rely heavily on the assumption that training and test data follow the same distribution, as well as extensive offline measurement data. However, in actual machining process, prediction accuracy of the traditional methods is reduced due to the variation in machining conditions and limited tool wear data. To address this problem, an Instance-based Transfer Learning framework is proposed to accurately predict the tool remaining useful life cross different working conditions. Firstly, a transfer learning algorithm is used to dynamically adjust the weights of all instances in multiple source domains, which aims to make full use of the source domain information that is highly correlated with the target data. Thus, the generalization ability of the model is improved, and the remaining tool life of the target working conditions could be well predicted with only a small amount of target domain data. Secondly, recurrent Gaussian process regression model is further developed as the base learner to improve the time series prediction capability of the transfer learning algorithm. The model limits the tool remaining useful life at adjacent moments through delayed feedback, while reducing the feature preparation time and the model complexity are reduced. The results indicate that the proposed framework can effectively improve the prediction accuracy of the tool remaining useful life cross different working conditions, and the prediction effectiveness also confirms the stability and reliability of the framework.

Key words: cross working conditions, instance-based transfer learning, Recurrent Gaussian process regression, time series prediction, tool remaining useful life

中图分类号: