Nondestructive rail neutral temperature estimation based on low-frequency vibrations and machine learning

Abstract

This paper describes the latest advancements of a non-invasive inspection/monitoring technique to estimate the rail neutral temperature (RNT), which is the temperature at which the net longitudinal force in continuous welded rails is zero. The technique is based on the measurement of the lowest modes of vibration of the rail of interest, and the application of machine learning (ML) to predict the RNT from the spectral densities of the vibrations’ frequencies. In the study presented in this article, the proposed technique was tested twice in the field. Vibrations were induced with an instrumented hammer on a 5° curved rail on concrete ties, and detected with a few wireless and wired accelerometers. The time-series were post-processed to extract the power spectral densities across the 0–700 Hz range. These densities constituted the input vector of a ML algorithm trained with experimental data and then tested with different experimental data. The RNT predicted by the algorithm were compared and showed very good agreement with the temperature estimated by an independent party and based on conventional strain-gage rosettes.