Monitoring of bridge displacements is critical for safe and cost-effective railroad operations. One of the essential parameters for determining the serviceability of the bridges is the dynamic displacement of the bridge during train-crossing events. The traditional methods for bridge displacement measurement often utilize linear variable differential transducers (LVDTs). However, irregular terrain, remote and inaccessible locations, and the height of the railroad bridges make implementation of these sensors for displacement measurements inadequate, risky, and time-consuming, and sometimes not possible altogether. In recent years, the use of laser Doppler vibrometers (LDVs) in the field of bridge displacement measurement has drawn attention as an alternative. In these applications, the vibrometer is generally placed on a fixed-point reference close to the bridge. However, it is not always possible to locate a fixed reference perpendicular to the bridge span to measure transverse displacements, especially when the bridge spans over a large opening. Furthermore, LDV sensors require calibration for every unique, different setup and are cumbersome to implement across a variety of different bridges. This paper presents a novel concept for bridge displacement measurement that enables the use of noncontact and reference-free moving vibrometers in the field without the need for calibration. The concept discussed herein proposes a method of compensating for measurement errors due to the angular and linear movement of the vibrometer to obtain accurate transverse displacement measurements of bridges. The results of this study showed that the signal difference between the measured outputs of a moving LDV system and an LVDT was between 10% and 15% peak-to-peak and between 2% and 5% root-mean square (RMS), which are generally considered acceptable levels of accuracy by railroad managers for field applications. An outdoor test was conducted in which an LDV mounted on an unmanned aerial system (UAS) was used to collect displacement measurements of a moving target structure representing a railroad bridge train-crossing event. Researchers conducted three outdoor experiments by collecting LDV and LVDT measurements from the UAS as it was hovering. The results from this outdoor testing showed that the signal difference between the measured dynamic data from the moving LDV and the LVDT was less than 5% (peak) and 10% (RMS). The reference-free dynamic displacement is generally accepted by railroad managers as an important index related to bridge deterioration, so this application can inform railroads in the field of prioritization decisions without the need for installed sensors. The ultimate goal of this research was to validate the applicability of the proposed method by mounting one LDV on a UAS and measuring dynamic bridge displacements with minimal setup and calibration.

Measurement of bridge displacements is important for ensuring the safe operation of railway bridges. Traditionally, contact sensors such as Linear Variable Displacement Transducers (LVDT) and accelerometers have been used to measure the displacement of the railway bridges. However, these sensors need significant effort in installation and maintenance. Therefore, railroad management agencies are interested in new means to measure bridge displacements. This research focuses on mounting Laser Doppler Vibrometer (LDV) on an Unmanned Aerial System (UAS) to enable contact-free transverse dynamic displacement of railroad bridges. Researchers conducted three field tests by flying the Unmanned Aerial Systems Laser Doppler Vibrometer (UAS-LDV) 1.5 m away from the ground and measured the displacement of a moving target at various distances. The accuracy of the UAS-LDV measurements was compared to the Linear Variable Differential Transducer (LVDT) measurements. The results of the three field tests showed that the proposed system could measure non-contact, reference-free dynamic displacement with an average peak and root mean square (RMS) error for the three experiments of 10% and 8% compared to LVDT, respectively. Such errors are acceptable for field measurements in railroads, as the interest prior to bridge monitoring implementation of a new approach is to demonstrate similar success for different flights, as reported in the three results. This study also identified barriers for industrial adoption of this technology and proposed operational development practices for both technical and cost-effective implementation.

Today, railroads carry 40% of the United States’ freight tonnage, and the demand for this service is expected to double in 20 years. The railroad infrastructure contains more than 100,000 bridges spanning over 225,000 km of tracks. Half of those bridges are over 100 years old. Accordingly, safe and efficient railroad operations depend on the continuous maintenance of this aging bridge network. Recent research has demonstrated that bridge displacements can be utilized as an indication of structural integrity in the prioritization of maintenance, repair, and replacement (MRR) operations. However, existing response-measurement methods either require expensive technological instrumentation or are based on complex structural modeling assumptions. Furthermore, many of them do not accommodate real-time monitoring. This paper proposes a new methodology fusing multimetric measurement data obtained from reference-free sensors, such as accelerometers and tiltmeters, in real time to provide accurate transverse displacement measurements inexpensively. Twenty-four percent of railroad bridges in the United States are timber. Therefore, to study the performance of the proposed method, a realistic train-crossing event was simulated by using a numerical model of a timber pile bent. Timber piles were simplified as free-standing cantilever beams accounting for the soil–structure interaction based on literature and industry input. The model was excited dynamically with real train input loads collected at the site. The analysis reproduced nonsymmetric bridge responses, including pseudostatic displacements observed in the field, which conventional accelerometers cannot measure. The pseudostatic displacements from tiltmeter data were extracted using simple Euler-Bernoulli beam equations that can be estimated for any railroad timber pile bent, requiring only the length of the pile above ground as the input. These pseudostatic responses were combined with dynamic displacements derived from accelerometer data, allowing the estimation of the transverse displacement in the field without the need of a finite-element model. Researchers compared measured responses to displacements estimated by the proposed data-fusion method. The results indicate that transverse bridge displacements can be determined by combining pseudostatic and dynamic responses in real time more precisely compared to previous methods. Consequently, more effective MRR prioritization can be undertaken with accurate response measurement.

Currently, over half of the U.S.’s railroad bridges are more than 100 years old. Railroad managers ensure that the proper Maintenance, Repair, and Replacement (MRR) of rail infrastructure is prioritized to safely adapt to the increasing traffic demand. By 2035, the demand for U.S. railroad transportation will increase by 88%, which indicates that considerable expenditure is necessary to upgrade rail infrastructure. Railroad bridge managers need to use their limited funds for bridge MRR to make informed decisions about safety. Consequently, they require economical and reliable methods to receive objective data about bridge displacements under service loads. Current methods of measuring displacements are often expensive. Wired sensors, such as Linear Variable Differential Transformers (LVDTs), require time-consuming installation and involve high labor and maintenance costs. Wireless sensors (WS) are easier to install and maintain but are in general technologically complex and costly. This paper summarizes the development and validation of LEWIS2, the second version of the real-time, low-cost, efficient wireless intelligent sensor (LEWIS) for measuring and autonomously storing reference-free total transverse displacements. The new features of LEWIS2 include portability, accuracy, cost-effectiveness, and readiness for field application. This research evaluates the effectiveness of LEWIS2 for measuring displacements through a series of laboratory experiments. The experiments demonstrate that LEWIS2 can accurately estimate reference-free total displacements, with a maximum error of only 11% in comparison with the LVDT, while it costs less than 5% of the average price of commercial wireless sensors.

In the last decade, the interest in machine learning (ML) has grown significantly within the structural health monitoring (SHM) community. Traditional supervised ML approaches for detecting faults assume that the training and test data come from similar distributions. However, real-world applications, where an ML model is trained, for example, on numerical simulation data and tested on experimental data, are deemed to fail in detecting the damage. The deterioration in the prediction performance is mainly related to the fact that the numerical and experimental data are collected under different conditions and they do not share the same underlying features. This paper proposes a domain adaptation approach for ML-based damage detection and localization problems where the classifier has access to the labeled training (source) and unlabeled test (target) data, but the source and target domains are statistically different. The proposed domain adaptation method seeks to form a feature space that is capable of representing both source and target domains by implementing a domain-adversarial neural network. This neural network uses H-divergence criteria to minimize the discrepancy between the source and target domain in a latent feature space. To evaluate the performance, we present two case studies where we design a neural network model for classifying the health condition of a variety of systems. The effectiveness of the domain adaptation is shown by computing the classification accuracy of the unlabeled target data with and without domain adaptation. Furthermore, the performance gain of the domain adaptation over a well-known transfer knowledge approach called Transfer Component Analysis is also demonstrated. Overall, the results demonstrate that the domain adaption is a valid approach for damage detection applications where access to labeled experimental data is limited.