融合时空信息的刀具健康状态评估及寿命预测
Fusion of spatio-temporal information on assessment of tools'health status and prediction of their useful time
黄秋豪 1吴兴富 2黄强飞 3杨骅 4牟全臣 5李子瑞4
作者信息
- 1. 河北工业大学人工智能与数据科学学院,天津 300401
- 2. 湖南大学机械与运载工程学院,湖南长沙 410082
- 3. 中国航发南方工业有限公司,湖南株洲 412004
- 4. 河北工业大学机械工程学院,天津 300401
- 5. 苏州数设科技有限公司,江苏苏州 215000
- 折叠
摘要
刀具的健康状态评估和寿命预测对于刀具的预测性维护、刀具生产准备及制定刀具需求计划等方面具有重要影响.为了准确监测刀具健康状态和预测剩余寿命,文中提出一种融合时空信息的刀具健康状态监测及寿命预测方法.该方法利用频谱时间图神经网络StemGNN(Spectral Temporal Graph Neural Network),通过图结构表征时间序列间的关系,在频域和谱域对传感器数据的时空关系进行建模,重构学习刀具健康状态下的数据分布,并向模型输入运行周期数据,输出重构数据与原始数据的误差作为刀具退化过程的健康指标(Health Indicator,HI),形成刀具健康状态曲线;然后,将刀具健康状态指标作为输入构建基于通道注意力机制(ECANet)和时序卷积网络(TCN)的刀具寿命预测方法,学习健康指标序列时间依赖,实现了刀具的剩余寿命(Remaining Useful Life,RUL)预测.在PHM2010数据集上进行试验验证,结果表明,相较于对比方法,所提方法更好地反映了刀具的退化趋势,提高了刀具寿命的预测精度.
Abstract
Both the assessment of tools'health status and the prediction of their useful life are crucial for predictive ma-intenance,preparation for production,and formulation of demand plans.In this article,in order to accurately assess the tools'health status and predict their remaining useful life,a method which fuses the spatio-temporal information on assess-ment of the tools'health status and prediction of their useful life is proposed.This method employs the StemGNN(Spectral Temporal Graph Neural Network)to represent the relationship between the time series through a graph structure;the spatio-temporal relationship of the sensor data is modeled in the frequency and spectral domains;the data distribution is reconstruc-ted when the tools are in good health.Besides,the operational cycle data is input into the model,while the error between the reconstructed data and the original data is output as a Health Indicator(HI),thus forming the curve of the tools'health status.Subsequently,the indicators of the tools'health status are used as the input to propose the method of predicting the tools'useful time,based on the Efficient Channel Attention Network(ECANet)and the Temporal Convolutional Network(TCN).The time dependencies of the health indicator sequences are explored;as a result,the tools'remaining useful life(RUL)is predicted.A series of experiment are conducted on the PHM2010 dataset.It is shown that this method,com-pared with its counterpart,can reflect the tools'trend of degradation more accurately and has a higher standard of accuracy in prediction of their useful life.
关键词
刀具/健康状态评估/剩余寿命预测/频谱时间图神经网络/时序卷积神经网络Key words
tool/assessment of health status/prediction of remaining useful life/Spectral Temporal Graph Neural Network/Temporal Convolutional Neural Network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据