A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals

doi:10.1016/j.measurement.2022.111782

	A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals
	Du, Xianjun; Jia, Wenchao; Yu, Ping; Shi, Yaoke; Cheng, Shengyi
刊名	Measurement: Journal of the International Measurement Confederation
	2022-10-01
卷号	202
关键词	Convolution Forecasting Preventive maintenance Rotating machinery Vibrations (mechanical) Wavelet transforms Attention mechanisms Continuous Wavelet Transform Convolutional neural network Degradation trend Feature map Remaining useful life predictions Rolling bearings Spectral feature Time frequency Time–frequency spectral feature map
ISSN号	0263-2241
DOI	10.1016/j.measurement.2022.111782
英文摘要	As a key component of the rotating machines, rolling bearings are widely used in mechanical engineering, aerospace and other fields. The health condition is closely related to the safe operation of the equipment. Predicting the degradation trend and remaining useful life of rolling bearings can enable effective preventive maintenance of rotating machinery. Therefore, an attention mechanism based multiscale convolutional neural network prediction model is proposed in this paper. Firstly, the continuous wavelet transform (CWT) is used to transform the one-dimensional vibration signal collected by the sensor into a two-dimensional time–frequency spectral feature map. Secondly, the quadratic degradation function is selected to determine the health indices of the bearings. Thirdly, the multi-scale convolutional neural network (MSCNN) is employed to realize the deep feature extraction. The multi-scale fusion features are constructed by extracting different degradation features of the signal using convolutional kernels of different sizes, and the necessary degradation features extracted are further enhanced and non-essential features are suppressed through a convolutional attention mechanism. Finally, the proposed model is verified on the PRONOSTIA dataset and compared with other prediction methods. The results indicate that the proposed one achieves better performances than other algorithms with the lowest prediction error and the highest prediction score. It is verified that this method can effectively improve the prediction accuracy and generalization performance, which could provide a certain theoretical basis and for RUL prediction of bearings and other equipment. © 2022 Elsevier Ltd
语种	英语
出版者	Elsevier B.V.
内容类型	期刊论文
源URL	[http://ir.lut.edu.cn/handle/2XXMBERH/159751]
专题	电气工程与信息工程学院
作者单位	College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou; 730050, China
推荐引用方式 GB/T 7714	Du, Xianjun,Jia, Wenchao,Yu, Ping,et al. A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals[J]. Measurement: Journal of the International Measurement Confederation,2022,202.
APA	Du, Xianjun,Jia, Wenchao,Yu, Ping,Shi, Yaoke,&Cheng, Shengyi.(2022).A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals.Measurement: Journal of the International Measurement Confederation,202.
MLA	Du, Xianjun,et al."A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals".Measurement: Journal of the International Measurement Confederation 202(2022).