The detection of rail surface defects is an important tool to ensure the safe operation of rail transit. Due to the complex diversity of track surface defect features and the small size of the defect area, it is difficult to obtain satisfying detection results by traditional machine vision methods. The existing deep learning-based methods have the problems of large model sizes, excessive parameters, low accuracy and slow speed. Therefore, this paper proposes a new method based on an improved YOLOv4 (You Only Look Once, YOLO) for railway surface defect detection. In this method, MobileNetv3 is used as the backbone network of YOLOv4 to extract image features, and at the same time, deep separable convolution is applied on the PANet layer in YOLOv4, which realizes the lightweight network and real-time detection of the railway surface. The test results show that, compared with YOLOv4, the study can reduce the amount of the parameters by 78.04%, speed up the detection by 10.36 frames per second and decrease the model volume by 78%. Compared with other methods, the proposed method can achieve a higher detection accuracy, making it suitable for the fast and accurate detection of railway surface defects.

The rolling bearing is an important part of the train’s running gear, and its operating state determines the safety during the running of the train. Therefore, it is important to monitor and diagnose the health status of rolling bearings. A convolutional neural network is widely used in the field of fault diagnosis because it does not require feature extraction. Considering that the size of the network model is large and the requirements for monitoring equipment are high. This study proposes a novel bearing fault diagnosis method based on lightweight network ShuffleNet V2 with batch normalization and L2 regularization. In the experiment, the one-dimensional time-domain signal is converted into a two-dimensional Time-Frequency Graph (TFG) using a short-time Fourier transform, though the principle of graphics to enhance the TFG dataset. The model mainly consists of two units, one for extracting features and one for spatial down-sampling. The building units are repeatedly stacked to construct the whole model. By comparing the proposed method with the origin ShuffleNet V2, machine learning model and state-of-the-art fault diagnosis model, the generalization of the proposed method for bearing fault diagnosis is verified.

The traction drive system of a high-speed train has a vital role in the safe and efficient operation of the train. This paper established an electromechanical coupling model of a high-speed train. The model considers the interaction of the gear pair, the equivalent connecting device of the transmission system, the equivalent circuit of the traction motor, and the direct torque control strategy. Moreover, the numerical simulation of the high-speed train model includes constant speed, traction, and braking conditions. The results indicate that the meshing frequency and the high harmonics rotation frequency constitute the stator current. Furthermore, both frequencies are evident during constant speed. However, they are blurry among other conditions except for twice the rotation frequency. Meanwhile, the rotor and stator currents’ root-mean-square (RMS) values during traction are less than the RMS value during braking. The initiation of traction and braking causes a significant increase in current. During the traction and braking process, the RMS value of the current gradually decreases. Therefore, it is necessary to pay attention to the impact of the transition process on system reliability.

In recent years, the lightweight neural network models have been gradually applied to fault diagnosis. In order to solve the problems about computation bottleneck of the pointwise convolution module which is widely used in lightweight networks, and explore how to effectively evaluate the quality of extracted features as well as deeply merge traditional fault diagnosis methods into deep learning, this paper proposed a diagnosis model named butterfly-transform (BFT)-MobileNet V3. BFT-MobileNet V3 was based on MobileNet V3, and consisted of BFT module and a novel algorithm called Deep-SHAP. This model not only had the advantages of low time complexity and high accuracy compared with the original network, but also had a novel feature that was able to automatically figure out the fault characteristic frequency and visualize the quality of extracted features. The experimental results showed that the time complexity of the BFT-MobileNet V3 model proposed in this paper decreases from O n 2 to O n logn while keeping a high accuracy rate. With the same time complexity, BFT-MobileNet V3 also had a higher accuracy rate than other networks. Meanwhile, with the Deep SHAP algorithm, the proposed model can accurately calculate the fault feature frequency of the roller bearings as well as intuitively visualize the quality of extracted features.

The problem of distribution shift during the degradation of rolling bearings can lead to a decrease in the prediction accuracy of the remaining useful life (RUL). Moreover, the traditional sequential long short-term memory (LSTM) model has limitations in its ability to extract relevant information, which hinders its prediction performance. This paper presents a novel strategy called Channel Dependent and Channel Independent (CD-CI) to alleviate distribution shift by improving robustness. In the CD strategy, a fusion of the residual LSTM and self-attention (RLSA) module is proposed. Subsequently, residuals connect multiple RLSAs to create the RRLSA stack, enhancing the learning of feature dependencies. Meanwhile, the CI strategy fuses Dlinear and self-attention (DLSA) to intensify attention toward the degradation information of a single feature. Ultimately, this paper introduces an adaptive channel-time attention shrinkage (CTAS) module to reduce internal noise within the model. The validation and analysis of the CD-CI strategy uses the IEEE PHM 2012 bearing dataset and XJTU-SY bearing dataset. The CD-CI strategy exhibits minimal (non-)robustness variation, indicating a robust performance. The root mean square error (RMSE), mean absolute error (MAE), and Score predicted by RUL are improved by 9.9%, 16.4%, and 15.9%, respectively, compared with the advanced model. These experiments indicate that the CD-CI strategy can alleviate distribution shift and improve prediction accuracy by enhancing robustness.