Novel methodology for identifying the weight of moving vehicles on bridges using structural response pattern extraction and deep learning algorithms

Abstract

Acquiring the weights of moving vehicles on bridges is crucial for structural performance assessment, as well as for bridge operating and maintenance strategies. Due to the high cost and complex maintenance structure of the bridge weigh-in-motion (BWIM) system, it is necessary to develop an economic and intelligent technology to replace it. This paper proposes a novel methodology for acquiring the weights of vehicles in motion on a bridge, based on the extraction of structural dynamic response patterns and deep learning algorithms that allow the input of a vehicle to be directly distinguished from the bridge’s output response. The response patterns induced by vehicles crossing a bridge are captured by accelerometers, and the raw signals transformed into two-dimensional spectrogram images via time–frequency analysis to extract corresponding visual patterns. Object classification is then conducted using a deep convolutional neural network (DCNN) to automatically learn the features, while the application of optimal analysis enhances performance of the trained network. Experiments in both a laboratory and in the field were conducted to evaluate the effectiveness of the proposed method with results demonstrating that vehicle loads could be directly distinguished from a bridge’s structural responses, even in a noisy environment.