As a key component of the high-speed railway catenary, the insulator plays an important role in supporting the catenary and maintaining the insulation between the catenary and earth. Regular inspection using computer vision technology is an effective way to detect insulator defects and improve the catenary operation safety. However, achieving full automation of insulator defects detection is still a difficult task due to the defective sample scarcity and subtle defect features. To overcome the problem, this paper proposes a defect detection GANs (DefGAN) that consists of a denoising auto-encoder, a discriminator and a classifier to detect defects, which can generate defective samples from pitted latent representations, thereby improving the reliability of defect detection classifier. The proposed DefGAN can pit the latent representations of normal samples to generate defective samples. Our proposal is mainly consists of two stages. In the first stage, we use a cascaded deep segmentation networks (CDSNets) to extract the insulator from the catenary images. Then the insulator area is sampled into small patches. In the second stage, the patches are fed into the proposed DefGAN to detect defects. Moreover, the defect score are determined by the anomaly probability of the classifier and reconstruction error of the denoising auto-encoder. The effectiveness of our work is measured across the catenary datasets.
15 October 2021 in IEEE Transactions on Intelligent Transportation Systems
Automatic inspections of overhead contact lines (OCLs) are developed to implement anomaly detection during normal operation. It is an essential prerequisite for efficient maintenance of railway electrification system. This paper presents a comprehensive survey on the inspections of OCLs with an emphasis on computer vision technology, which has developed rapidly due to its ability to understand images. Our survey begins with a brief introduction on anomalies in OCL inspections and generic procedures of computer vision inspection for anomaly detection. Subsequently, for detecting deviations of parameters and defective components during OCL inspections, the existing techniques involving stereo vision and object vision, especially convolution neural networks are described in detail from two aspects: measurement of OCL parameters and identification of OCL conditions. Some interference factors in OCL inspection are analyzed. Actual cases of the inspections are also briefly shown. Challenges and suggestions for further research on OCL inspection are drawn toward the end of the paper.
11 August 2022 in IEEE Transactions on Transportation Electrification
Due to being completely exposed outdoors without backup, overhead contact lines (OCLs) are more vulnerable to external weather conditions. To perceive the weather-related failure risk and evaluate the associated potential risks, a multilayer Bayesian network (MLBN) approach is proposed for predictive probabilistic risk assessment for OCLs. More specifically, the weather-driven failure probabilities of OCLs are predicted using a first-layer Bayesian network (BN), including lightning strikes, windstorms, and fog-haze. Then, the weather-driven failure patterns of OCLs are investigated and captured, and furthermore, the second-layer BN is embedded to evaluate the risk propagation. Furthermore, the third-layer BN is implemented to calculate and quantify weather-driven risk consequences from perspectives of economic loss and social trust loss constrained by power outage time and train timetable. The three-layer BN models are built by connecting the nodes with causal relationships, avoiding complicated structure learning. The experiments on the actual OCL failure records and weather data demonstrate that the proposed MLBN approach can identify weather-involved risk activities and comprehensively analyze the weather-driven cumulative risks of OCLs from spatiotemporal perspectives. In addition, it is capable of giving reasonable guiding information for mitigating the operational risks of OCLs against external weather conditions.
29 September 2020 in IEEE Transactions on Intelligent Transportation Systems
The swivel clevis (SC) is a vulnerable part of the Overhead Catenary System (OCS). Regular inspection using computer vision technology is an effective way to detect SC defects and improve the OCS operation safety. However, achieving full automation of SC defects detection is still a difficult task due to defective sample scarcity and data distribution shift. To overcome these problems, this paper proposes a novel defects detection method that combines an adaptive SC segmentation network (Adaptive SSN) and local operators. During the inspection process, an unreliability index defined by the model uncertainty and prior knowledge is used to monitor the reliability of the Adaptive SSN. When data distribution shift causes the Adaptive SSN to be unreliable, human annotator will be asked to update the training set and retrain the Adaptive SSN to adapt to the new data distribution. Then the geometric features obtained from segmentation masks and the local features extracted by local operators are used to detect the SC defects. Effectiveness of the proposed method is demonstrated by the experimental results on the data from several high-speed railway lines.
The contact wire support (CWS) is an important catenary component that maintains the contact wire height and stagger. The direct impact of the pantographs makes the CWS a vulnerable part of the catenary. Recently, automatic catenary inspection using computer vision and pattern recognition has been introduced to improve railway operation safety. However, the automated detection of CWS defects remains to be further studied. This paper proposes a novel CWS defect detection system that consists of three stages. First, the Faster R-CNN network is adopted to localize the key catenary components, and the image areas that contain CWS components are obtained. Then, the CWS components are segmented using a Bayesian fully convolutional catenary components segmentation network (CCSN) that fuses different level features of the backbone network. The CCSN is not only able to perform accurate CWS components segmentation, but also capable of evaluating the model uncertainty by Monte Carlo dropout. Finally, the defect status is determined using the proposed criteria, which are defined according to the geometries of the components. Experiments on the Hefei-Fuzhou high-speed railway line indicate that this approach can be applied to the CWS defect detection.