With the rapid development of deep learning (DL) techniques, data-driven models have been increasingly used in remaining useful life (RUL) prediction, in which convolution neural network (CNN)-based RUL prognostics models have received special attention. However, there are still two main issues that need to be addressed: 1) traditional CNN is not suitable to extract the time-sequence characteristics from the long-term historical signals and 2) the receptive field range of convolution operation is fixed, thus only learning the feature information at a specific scale, which is insufficient for complex feature extraction. To address these two issues, a multiscale temporal convolutional network (MsTCN) that has powerful time-sequence characteristics is proposed for RUL prediction in this article. The MsTCN adopts the temporal convolutional network (TCN) framework, which is good at extracting time-sequence information. Based on this, a new multiscale dilated causal convolution residual block (MsDCCRB) is developed to constitute the RUL prognostics model, where multiple dilated convolutions (DCs) are based on different dilation factors that are put on each layer in parallel. Furthermore, the squeeze-and-excitation (SE) unit is embedded into the MsDCCRB to adaptively recalibrate the sequence feature responses and enhance the representation learning ability. Through stacking multiple MsDCCRBs, the historical condition monitoring data can be fed directly into the proposed model to realize the high-level representations of RUL estimation. Finally, the proposed approach is validated with the accelerated whole-life degradation dataset of rolling element bearings (REBs). The experimental results exhibit that the proposed MsTCN achieves a higher RUL prediction accuracy, which is superior to some state-of-the-art data-driven prognostics methods.

Intelligent mechanical fault diagnosis algorithms based on deep learning have achieved considerable success in recent years. However, degradation of the diagnostic accuracy and operational speed has been significant due to unfavorable working conditions and increasing network depth. An improved version of ResNets is proposed in this paper to address these issues. The advantages of the proposed network are presented as follows. Firstly, a multi-scale feature fusion block was designed, to extract multi-scale fault feature information. Secondly, an improved residual block based on depthwise separable convolution was used to improve the operational speed and alleviate the computational burden of the network. The effectiveness of the proposed network was validated by discriminating between diverse health states in a gearbox under normal and noisy conditions. The experimental results show that the proposed network model has a higher classification accuracy than the classical convolutional neural networks, LeNet-5, AlexNet and ResNets and a faster calculation speed than the classical deep neural networks. Furthermore, a visual study of the different stages of the network model was conducted, to effectively comprehend the operational processes of the proposed model.

A sensible maintenance strategy must take into account the remaining usable life (RUL) estimation to maximize equipment utilization and avoid costly unexpected breakdowns. In view of some inherent drawbacks of traditional CNN and LSTM-based RUL prognostics models, a novel combination model of the ConvLSTM and the Transformer, which is based on the idea of “Extracting spatiotemporal features and applying them to RUL prediction”, is proposed for RUL prediction. The ConvLSTM network can directly extract low-dimensional spatiotemporal features from long-time degradation signals. The Transformer, based entirely on attention mechanisms, can deeply explore the mapping law between deep-level nonlinear spatiotemporal feature information and equipment service performance degradation. The proposed approach is validated with the whole-life degradation dataset of bearings from the PHM 2012 Challenge dataset and the XJTU-SY public dataset. The detailed comparative analysis shows that the proposed method has higher RUL prediction accuracy and outstanding comprehensive prediction performance.

Remaining useful life (RUL) prediction of key components is an important influencing factor in making accurate maintenance decisions for mechanical systems. With the rapid development of deep learning (DL) techniques, the research on RUL prediction based on the data-driven model is increasingly widespread. Compared with the conventional convolution neural networks (CNNs), the multi-scale CNNs can extract different-scale feature information, which exhibits a better performance in the RUL prediction. However, the existing multi-scale CNNs employ multiple convolution kernels with different sizes to construct the network framework. There are two main shortcomings of this approach: (1) the convolution operation based on multiple size convolution kernels requires enormous computation and has a low operational efficiency, which severely restricts its application in practical engineering. (2) The convolutional layer with a large size convolution kernel needs a mass of weight parameters, leading to a dramatic increase in the network training time and making it prone to overfitting in the case of small datasets. To address the above issues, a multi-scale dilated convolution network (MsDCN) is proposed for RUL prediction in this article. The MsDCN adopts a new multi-scale dilation convolution fusion unit (MsDCFU), in which the multi-scale network framework is composed of convolution operations with different dilated factors. This effectively expands the range of receptive field (RF) for the convolution kernel without an additional computational burden. Moreover, the MsDCFU employs the depthwise separable convolution (DSC) to further improve the operational efficiency of the prognostics model. Finally, the proposed method was validated with the accelerated degradation test data of rolling element bearings (REBs). The experimental results demonstrate that the proposed MSDCN has a higher RUL prediction accuracy compared to some typical CNNs and better operational efficiency than the existing multi-scale CNNs based on different convolution kernel sizes.

Fault detection of axle bearings is crucial to promote the safe, efficient, and reliable running of high-speed trains. In recent decades, time−frequency analysis (TFA) techniques have been widely used in mechanical equipment fault diagnoses. Time-reassigned multisynchrosqueezing transform (TMSST), as a novel time−frequency representation (TFR) algorithm, is more suitable for dealing with strong frequency-varying signals. However, TMSST TFR results are subject to noise interference. It is difficult to extract the accurate time−frequency (TF) fault feature of the axle bearing under a complex working environment. In addition, determination of the TMSST algorithm parameters depends on the personnel’s subjective experience. Therefore, the TMSST result has a great randomicity and has the disadvantage of having a poor reliability. To address the above issues, a hybrid SVD-based denoising and self-adaptive TMSST is proposed for axle bearing fault detection in this paper. The main improvements of the proposed algorithm include the following two aspects: (1) An SVD-based denoising method using the maximum SV mean to determine the reasonable SV order is adopted to eliminate noise interference and to reserve useful fault impulse information. (2) A new evaluation metric, named time−frequency spectrum permutation entropy (TFS-PEn), is put forward for the quantitative evaluation of the performance of TFR for the TMSST, and then a water cycle algorithm (WCA)-based optimized TMSST can adaptively determine the optimal algorithm parameters. In both the simulation and experimental tests, the superiority and effectiveness of the proposed method is compared with the TMSST, short-time Fourier transform (STFT), MSST, wavelet transform (WT), and Hilbert-Huang transform (HHT) methods. The results show that the proposed algorithm has a better performance for extracting the weak fault features of axle bearing under a strong background noise environment.

Numerous convolutional neural network (CNN)- and Transformer-based models have made significant progress in fault diagnosis. However, challenges remain, notably limited feature extraction abilities and elevated computational expenses, especially when applied to the fault diagnosis of wheelset bearings under intricate operational scenarios. This study proposes an improved RSMamba network based on multi-domain image fusion for wheelset bearing fault diagnosis. We have devised an RGB-CF strategy that integrates time, frequency, and time-frequency domain features to convert 1D vibration signals into 2D images. The RSMamba network is enhanced through the introduction of dynamic multi-path Mamba blocks, which handle non-causal relationships, and the embedding of a CSRA module to boost the model’s capacity to recognize class-specific features. The experimental results show that the proposed model attains a classification accuracy exceeding 99% across six testing tasks utilizing two distinct real-world wheelset bearing datasets, outperforming existing CNN- and Transformer-based models substantially in diagnostic accuracy and computational efficiency. This study demonstrates the substantial potential of the proposed methodology in enhancing fault diagnosis for wheelset bearings, making it a viable option for practical implementation in the maintenance of high-speed trains.