Bridge frequency estimation strategies using smartphones

Abstract

Bridges are susceptible to deterioration and damage as they age and should be routinely assessed to evaluate their integrity and safety for service. Traditionally, structural monitoring has comprised visual inspections, however this is both time and labor intensive. Researchers have shown that sensors on moving vehicles may provide insight into the dynamic behavior of bridges. Accelerometers within smartphones may serve as the sensors from which data is collected; thus, enabling massive data collection from a fleet of potential monitoring vehicles. This paper presents four postprocessing strategies for estimating bridge frequencies from smartphone acceleration data streams with no a priori information about the mass or stiffness of the bridge or vehicle. These techniques utilize the DFT and MUSIC algorithms to calculate vehicle acceleration frequency spectrums from which the fundamental bridge vibration frequency may be estimated. Both single-vehicle and crowdsourced postprocessing techniques are investigated. Utilizing the MUSIC algorithm within a crowdsourcing framework, the correct bridge frequency was identified in all analytical simulations within 4% error, representing a significant increase in performance over single-vehicle estimations made using MUSIC. The effect of user interaction with the smartphone is studied by including superimposed acceleration signals on 25–100% of analytical results; the superimposed user events included a dropped smartphone and talking on a smartphone. Increasing the percentage of noisy signals in the pool of evaluated accelerations generally reduces performance with the exception of crowdsourced estimations made using the MUSIC algorithm, which proved to be robust against user interaction with the smartphone.