Rail defect detection is crucial to rail operations safety. Addressing the problem of high false alarm rates and missed detection rates in rail defect detection, this paper proposes a deep learning method using B-scan image recognition of rail defects with an improved YOLO (you only look once) V3 algorithm. Specifically, the developed model can automatically position a box in B-scan images and recognize EFBWs (electric flash butt welds), normal bolt holes, BHBs (bolt hole breaks), and SSCs (shells, spalling, or corrugation). First, the network structure of the YOLO V3 model is modified to enlarge the receptive field of the model, thus improving the detection accuracy of the model for small-scale objects. Second, B-scan image data are analyzed and standardized. Third, the initial training parameters of the improved YOLO V3 model are adjusted. Finally, the experiments are performed on 453 B-scan images as the test data set. Results show that the B-scan image recognition model based on the improved YOLO V3 algorithm reached high performance in its precision. Additionally, the detection accuracy and efficiency are improved compared with the original model and the final mean average precision can reach 87.41%.
Rail wear occurs continuously owing to the rolling contact load of trains and is fundamental for railway operational safety. A point‐based manual rail wear inspection cannot satisfy the increasing demand for rapid, low‐cost, and continuous monitoring. This paper proposes a depth‐plus‐region fusion network for detecting rail wear on a running band, which is a collection of wheel–rail interaction traces. The following steps are involved in the proposed method. (i) A depth map estimated by a modified MiDaS model is utilized as guidance for exploiting the depth information of the running band for rail wear detection. (ii) The running band of a rail is segmented and extracted from images using an improved mask region‐based convolutional neural network that uses the scale and ratio information to perform instance segmentation of the running band images. (iii) A two‐channel attention–fusion network that classifies rail wear is constructed. In this study, we collected real‐world running band images and rail wear‐related data to validate our approach using a high‐accuracy rail‐profile measurement tool. The case‐study results demonstrated that the proposed method can rapidly and accurately detect rail wear under different ambient light conditions. Moreover, the recall rate of severe wear detection was 84.21%.
This study developed a multi-classification model for vehicle interior noise from the subway system, collected on smartphones. The proposed model has the potential to be used to analyze the causes of abnormal noise using statistical methods and evaluate the effect of rail maintenance work. To this end, first, we developed a multi-source data (audio, acceleration, and angle rate) collection framework via smartphone built-in sensors. Then, considering the Shannon entropy, a 1-second window was selected to segment the time-series signals. This study extracted 45 features from the time- and frequency-domains to establish the classifier. Next, we investigated the effects of balancing the training dataset with the Synthetic Minority Oversampling Technique (SMOTE). By comparing and analyzing the classification results of importance-based and mutual information-based feature selection methods, the study employed a feature set consisting of the top 10 features by importance score. Comparisons with other classifiers indicated that the proposed XGBoost-based classifier runs fast while maintaining good accuracy. Finally, case studies were provided to extend the applications of this classifier to the analysis of abnormal vehicle interior noise events and evaluate the effects of rail grinding.
Accurately estimating the quality of transmission (QoT) in modern transport optical networks has been regarded as one of the most critical factors to reduce the design margins. In recent years, machine learning (ML) based models have exhibited a powerful capacity for various kinds of QoT estimation tasks. However, the existing ML-based QoT estimators suffer from two kinds of phenomena that are hard to bypass in real optical networks. The first conundrum is the variation of the number of parameters in transmission features introduced by the changeable link configurations. The second conundrum is the distribution drift of the transmission parameters relative to the training dataset. To mitigate the above two problems, we propose an invariant convolutional neural network predictor (ICNNP), which consists of a fixed-length encoder for encoding variable-length link features, and a robust neural network predictor, which can adapt to the changing transmission conditions with limited additional data. To alleviate the time dependence and link length dependence of the QoT estimator, we trained the model with a joint training algorithm. We validate our method experimentally by collecting datasets under different transmission configurations. The proposed ICNNP exhibits significant advantages in comparison with the four benchmark algorithms. When the span numbers vary from 9 to 12 and the evaluation period is expanded from 12 to 72 h, the standard deviation of the signal-to-noise ratio prediction error of our model holds below 0.4 dB and 0.25 dB, respectively. We also propose a continual learning workflow with an evaluation-update framework, with which our model can perform QoT estimation with the highest efficiency and the lowest training cost. The ensemble of components in this paper builds a deployment-oriented reliable QoT estimation tool.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.