ITSC 2025 Paper Abstract

Paper WE-EA-T9.3

Ding, Lu (Tongji University), Li, Wenhao (Tongji University), Zhou, Tingliang (CASCO Signal Ltd), Han, Dong (CASCO Signal Ltd.), Jin, Bo (Tongji University)

Multimodal Fault Diagnosis for Railway Switch Machine Based on Semantic Information Contrastive Learning

Scheduled for presentation during the Regular Session "S09b-Optimization for Multimodal and On-Demand Urban Mobility Systems" (WE-EA-T9), Wednesday, November 19, 2025, 14:10−14:30, Coolangata 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 19, 2025

Keywords Autonomous Rail Systems and Advanced Train Control Technologies, IoT-based Traffic Sensors and Real-time Data Processing Systems, AI, Machine Learning and Predictive Analytics for Traffic Incident Detection and Management

Abstract

The railway switch machine is a critical component of the railway signaling system, significantly influencing the efficiency and safety of train operations. However, most existing fault diagnosis technologies for switch machines struggle to differentiate samples with similar characteristics, leading to low diagnostic accuracy. To address this limitation, we propose a novel multimodal fault diagnosis method for railway switch machines. Via integrating current signals of switch machines (CSSM) with their corresponding semantic descriptions during training, this method, termed the Multi-modal CSSM-TEXT Fault Diagnosis (MCTFD) model, can notably improve the performance of fault classification only using numerical current signals during inference. MCTFD utilizes contrastive learning to achieve modal alignment between text and time series data. To the best of our knowledge, this is the first work that employs semantic information for the fault diagnosis of switch machines. Experimental results using real-world and generated data of Shanghai Metro demonstrate that the proposed MCTFD model achieves superior diagnostic accuracy, compared to existing machine learning and deep learning methods that rely on single-modal data. By effectively incorporating semantic information, our method shows strong potential for enhancing fault diagnosis capabilities in rail transit systems.