Predictive evaluation of dynamic responses and frequencies of bridge using optimized VMD and genetic algorithm-back propagation approach

Abstract

The large amount of data collected by structural health monitoring systems deployed in the bridge contains dynamic information about the structure. To enhance the prediction accuracy of the structural dynamic responses and to evaluate the frequencies from predicted restructured responses, this paper develops an approach of optimized variational mode decomposition (OVMD) combined with a genetic algorithm-back propagation (GA-BP) neural network. The procedure is first to establish the OVMD algorithm using relative root mean square error (RRMSE) and correlation coefficient to determine reasonable decomposition and retention of the intrinsic mode function (IMF) components in the response decomposition. Then each retained IMF component is used as input to the GA-BP for prediction. Finally, the frequencies and their characteristics of the structure are estimated from the predicted restructured responses. A damaged arch bridge test shows that OVMD overcomes the shortcomings of VMD, decomposes and reconstructs the signals effectively, and outperforms the other three methods in denoising. The experimental results of the long-span cable-stayed bridge prove that OVMD combined with GA-BP has higher prediction accuracy for the dynamic responses with high sampling rates. The structural frequencies are correctly determined from predicted recombined displacement and acceleration responses. This approach provides a useful tool for bridge dynamic response decomposition, reconstruction, prediction, and structural frequency evaluation.