Multistage Physics-Informed Signal Processing Framework for In-Motion Detection of Track Stiffness Irregularities Using Onboard Vibration Sensors

With nearly 140,000 mi of track, railroads are central to North America’s transportation infrastructure, carrying over 40% of freight by ton-miles and serving millions of passengers annually. Maintaining the structural integrity of the track is essential for operational safety and economic efficiency. Integrity is compromised by stiffness changes caused by ballast degradation, subgrade settlement, aging ties, temperature-induced stresses, and repeated loading. Proactive track health monitoring systems are needed to detect such changes continuously under dynamic train loads. This paper presents a multistage physics-informed framework for the in-motion detection of track stiffness irregularities (TSIs), serving as a proxy for identifying potential defects. The proposed system uses onboard vibration measurements, processed through advanced signal processing techniques. The system consists of three modules, for data acquisition, change detection, and change classification. It operates on an edge-computing platform, allowing real-time processing, and achieves over 95% data compression. The change-detection module, emphasized in this paper, combines wavelet packet analysis, variational mode decomposition, and the Hilbert transform to extract instantaneous energy features from vertical acceleration signals. These features act as sensitive indicators of track stiffness variation. The method was validated through both simulation and offline field data. Simulations captured a wide range of TSIs, including abrupt changes, gradual transitions, and localized weak zones. Field validation confirmed the model’s ability to consistently detect recurring irregularity patterns along the track, without requiring any assumptions about their physical origin. These results demonstrate the robustness, scalability, and real-world applicability of the proposed approach for continuous rail infrastructure monitoring.