Removing spike noise from railway geometry measures with a fuzzy filter

This paper describes an algorithm based on fuzzy logic for the real-time detection and removal of impulsive spike noise from measures of track geometry in railway infrastructures. Spike noise is caused by random events, including anomalous reflections of external lights, detection or transmission errors within the measuring systems, and singular parts of tracks (switches, level crossings) hiding the rail profile. Its effects are extremely undesirable because they can generate erroneous alerts about inexistent railway defects. The real-time detection of spike noise is generally difficult, as the inspected tracks have extremely variable characteristics, and the measurement is performed in dynamic and turbulent conditions. By incorporating in a fuzzy system the typical rules used for offline analysis of the measured data, we develop a filtering algorithm that is able to perform spike removal in real time. Moreover, we also propose a variant of the fuzzy filter that improves the reconstruction of the corrupted measures. The proposed algorithm is validated on an extensive set of experimental investigations, which include a comparison with other algorithms selected from recent literature. The results confirm the effectiveness of the filtering algorithm, which has been recently implemented on some measuring vehicles for high-speed lines