Detection method on pantograph-catenary arcing of electric locomotives based on multi-frequency-band characteristics of current signals
Detecting pantograph-catenary arcing on trains is crucial for ensuring safety in railway operation and maintenance. Most existing detection methods rely on optical instruments to capture pantograph-catenary images, followed by the analysis of these images to identify arc spectra as evidence of arcin...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Department of Electric Drive for Locomotives
2024-07-01
|
| Series: | 机车电传动 |
| Subjects: | |
| Online Access: | http://edl.csrzic.com/thesisDetails#10.13890/j.issn.1000-128X.2024.04.022 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Detecting pantograph-catenary arcing on trains is crucial for ensuring safety in railway operation and maintenance. Most existing detection methods rely on optical instruments to capture pantograph-catenary images, followed by the analysis of these images to identify arc spectra as evidence of arcing occurrences. However, these methods are limited by inadequate visibility in the external environments of trains, and maintenance access can be challenging. To address these issues, this paper proposed a detection method based on multi-frequency-band characteristics of current signals. First, based on the time-domain and frequency-domain characteristics of arcs, leveraging theoretically derived, simulated and measured waveforms, the following characteristic components of pantograph-catenary arcs were demonstrated: extremely low-frequency components caused by instantaneous ionization, harmonic components resulting from LC oscillation, and high-frequency components. These characteristic components were then utilized to devise a measurement scheme and data preprocessing algorithm, and historical data were incorporated, leading to the establishment of feature sets. Additionally, a random forest model was established, with feature vectors as inputs and detection results as outputs. The arcing labels and feature sets provided by 3C equipment were incorporated for training, to develop a classifier enabling real-time arcing detection. Its efficacy was demonstrated through on-board experiments, showcasing a detection precision up to 100% and a recall approximating 98.9%. In addition, the proposed method supports certain extensions for more application scenarios, after training using different event labels provided by users. |
|---|---|
| ISSN: | 1000-128X |