Novel identification technique of moving loads using the random response power spectral density and deep transfer learning

Abstract

Accurate identification of vehicle loads plays a great significance in design, monitoring and evaluation of bridges. However, it is hard to acquire them via conventional methods because of their weak practicality or low modeling accuracy. To this end, this study proposes an innovative method to perform load identification based on the random response power spectral density (PSD) and deep transfer learning strategy. This method belongs to a data-driven model, which avoids the expensive facility requirement or ill-posed inverse problem in conventional methods. Specifically, a vehicle-bridge interaction dynamic system considering the disturbance of road surface excitation is firstly established to generate structural response for any specified vehicle loads. Then, the corresponding response PSD is transformed to a three-dimensional matrix in the form of color image, by which the deep transfer learning strategy is incorporated in deep convolutional neural network for describing the mapping relationship between structural response and these specified vehicle loads. On this basis, load identification can be performed via inputting any interested response PSD into the established relationship. Finally, the effectiveness of the proposed method is validated numerically. The results demonstrate that it can accurately identify the vehicle weight, vehicle speed and road surface roughness with an overall accuracy exceeding 98% and presents an excellent anti-noise capacity.