Virtual Axle Detector Based on Analysis of Bridge Acceleration Measurements by Fully Convolutional Network

Abstract

In the practical application of the Bridge Weigh-In-Motion (BWIM) methods, the position of the wheels or axles during the passage of a vehicle is a prerequisite in most cases. To avoid the use of conventional axle detectors and bridge type-specific methods, we propose a novel method for axle detection using accelerometers placed arbitrarily on a bridge. In order to develop a model that is as simple and comprehensible as possible, the axle detection task is implemented as a binary classification problem instead of a regression problem. The model is implemented as a Fully Convolutional Network to process signals in the form of Continuous Wavelet Transforms. This allows passages of any length to be processed in a single step with maximum efficiency while utilising multiple scales in a single evaluation. This allows our method to use acceleration signals from any location on the bridge structure and act as Virtual Axle Detectors (VADs) without being limited to specific structural types of bridges. To test the proposed method, we analysed 3787 train passages recorded on a steel trough railway bridge of a long-distance traffic line. Results of the measurement data show that our model detects 95% of the axles, which means that 128,599 out of 134,800 previously unseen axles were correctly detected. In total, 90% of the axles were detected with a maximum spatial error of 20 cm, at a maximum velocity of vmax=56.3m/s. The analysis shows that our developed model can use accelerometers as VADs even under real operating conditions.

Virtual Axle Detector: Train Axle Localization based on Bridge Vibrations

“…In order to apply the threshold, discrete time points representing the actual detection of an axle must be extracted from the multitude of pseudo-probabilities. For this purpose, we extracted the peaks from the individual pseudo probabilities with the same peak finding algorithm [4] and with the same settings as in VAD [3] to ensure comparability (0.25 for minimum height of the peak, 20 samples for minimum distance between two peaks, and 0.15 for prominence of the peak compared to the surrounding points).…”

“…The advantage of CWTs is that signals can be analysed in both the frequency domain and the time domain. Since the contribution of the structure-dependent natural vibration and the contribution of the load-induced vibration are mainly in different frequency ranges, VADs inputs were optimized accordingly [3]. It was assumed that the model should thus be able to learn the difference between the influence of the bridge and the influence of the loads more easily.…”

“…The positions can be determined with conventional axle detectors, but their lifetime is limited and their installation is expensive and often requires the blocking of roads or railroad tracks. To avoid these problems, we propose an improved version of the first method, to the authors' best knowledge, that should able to detect axles for all types of bridges using accelerometers that can be placed anywhere along the bridge [3].…”

“…The Virtual Axle Detector (VAD) is a Convolutional Neural Network (CNN) with exclusively input size independent layers [3]. By eliminating fully-connected or other layers constrained to a fixed input size, passages of any length can be processed in a single step without the need for a windowed approach.…”

“…The perceptive field of a CNN is defined by the kernel size of its convolutional layers. VADER was validated on the same data set [12,13] as VAD under real operating conditions to ensure comparability of the two models. The TensorFlow library [14] was used for the implementation of the models, and PlotNeu-ralNet [15] was used for its visualisation.…”