SRCAE-STCBiGRU: a fused deep learning model for remaining useful life prediction of rolling bearings

• Published in: Signal, image and video processing image and video processing, 2024-12, Vol.18 (12), p.9119-9140
• Main Authors: Deng, Linfeng, Yan, Xinhui, Li, Wei
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Bearings; Computer Imaging; Computer Science; Continuous wavelet transform; Deep learning; Feature extraction; Image acquisition; Image Processing and Computer Vision; Life prediction; Machine learning; Mechanical systems; Multimedia Information Systems; Original Paper; Pattern Recognition and Graphics; Roller bearings; Service life; Signal,Image and Speech Processing; Spatiotemporal data; System reliability; Useful life; Vision; Wavelet transforms
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-024-03534-1

The intelligent prediction of bearing remaining useful life (RUL) plays a critical role in bearing maintenance. Therefore, it is particularly significant to accurately estimate the RUL of bearings in order to ensure the reliability and safety of mechanical systems. And deep learning techniques have been successfully applied in the RUL prediction. However, there are unresolved problems of information loss during feature extraction and hardly effectively extracting spatio-temporal sequence information during bearing degradation process for the convolutional neural networks. To solve the problem, this paper proposes a RUL prediction framework based on stacked residual convolutional autoencoder and spatio-temporal convolutional bidirectional gated recurrent unit. The method adopts continuous wavelet transform technology to convert the acquired raw vibration signals into two-dimensional time–frequency images, constructs a deep network using stacked residual convolutional networks to extract feature information at different levels, and learns the spatio-temporal information in the time series information in the past and future states through spatio-temporal convolutional bi-directional gated recurrent units to more accurately predict the remaining service life of rolling bearings. In experimental verification, by comparing with existing RUL prediction methods and utilizing the PHM2012 and XJYU-SY public datasets, the superiority and effectiveness of our proposed method were well validated. The experimental results indicate that the proposed RUL prediction approach exhibits excellent performance in terms of accuracy and generalization.