A Machine Learning Framework for Rail Neutral Temperature Estimation Using Impulse Vibrational Responses

Longitudinal rail force management of continuous welded rail (CWR) is important for safe and efficient railroad operation. A key parameter to measure and monitor is the rail neutral temperature (RNT) or the stress-free temperature. The team proposed a supervised learning framework to estimate the RNT using impulse vibrational responses from CWRs. We first established an instrumented field site on a revenue-service line and collected impulse vibrational response data covering a wide range of temperature and thermal stress. Then, we trained a data-driven model that uses rail temperatures and modal frequencies as the input for in-situ RNT prediction. The results demonstrated that the proposed framework could provide RNT estimation with a reasonable precision (±5 ºF)