Prediction of Radio Signal Failures of Communication Based Train Operating Systems by Machine Learning Methods

Arslan, Burak; Tiryaki, Hasan · 2024 · Crossref

DOI: 10.18185/erzifbed.1196965

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical need for predictive maintenance in Communication Based Train Operating Systems (CBTC), which rely on continuous wireless radio communication for safe and efficient urban rail transit. Radio signal failures in CBTC systems can cause emergency braking, unexpected train stops, and significant operational delays, impacting passenger comfort and equipment longevity. The authors aim to develop a decision support model that predicts these signal failures before they disrupt operations, allowing maintainers to intervene proactively. This work represents the first study to utilize radio signal level data for failure prediction in railway signaling systems. The researchers collected data from trains operating on a CBTC-enabled railway line, focusing on wireless signal levels received from trackside radios. They defined ideal operating conditions as receiving a signal stronger than -85 dB from at least one radio with a packet loss rate below 1%. Based on deviations from these norms, they established six maintenance recommendations, ranging from normal operation to specific checks for fiber optic/LAN connections, antenna orientation, or radio output power. To predict these states, the study employed 29 different machine learning methods, including Decision Trees, Support Vector Machines, Naive Bayes, and various Artificial Neural Network (ANN) configurations. The models were trained and validated using MATLAB R2022a Classification Learner on a dataset of 502 training samples and 101 test samples, utilizing 10-fold cross-validation. Performance was evaluated using Correlation Coefficient ($R^2$), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The results demonstrated that the Bilayered Neural Network method outperformed all other algorithms in both training and testing phases. During training, this model achieved an $R^2$ of 0.972, an RMSE of 0.0333, and an MAE of 0.0238. In the testing phase, it achieved an $R^2$ of 0.980, an RMSE of 0.0232, and an MAE of 0.0137. The model’s effectiveness was further confirmed by a validation Area Under the Curve (AUC) of 0.99 and a test AUC of 0.97, indicating high predictive accuracy. Confusion matrices showed that True Positive Rates and Positive Predictive Values were consistently higher than their respective false rate counterparts, confirming the model's reliability in identifying signal issues. The significance of this study lies in its successful application of machine learning to enhance railway maintenance management. By accurately predicting radio signal failures, the proposed model enables early intervention, thereby preventing unscheduled stops and reducing repair times. This approach supports a shift from reactive to preventive maintenance, ensuring operational continuity and minimizing costs associated with downtime and equipment wear. The authors conclude that the high success rates validate the use of artificial neural networks for this specific application and suggest future development of interface software to integrate this decision support model into active preventative maintenance workflows.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction