Prediction of Tuned Liquid Damper Damping Force Based on Back Propagation Neural Network

Tuned liquid dampers (TLDs) become more and more popular in suppressing structural vibration under wind, seismic, and wave loads recently due to its low cost, easy adjustment of damping frequency and relatively high efficiency. However, it is difficult to establish the analytical or numerical model because of the highly non-linear behavior of TLDs, such as wave breaking, swirling, and overturning. This paper presents an accurate prediction of the damping capability of TLDs with shallow water depth. Back propagation neural network (BPNN) is applied to create the model of TLDs damping forces. Firstly, a general approach of BPNN with multi-hidden layer and iterative steps is introduced. Then, different combinations of input parameters are investigated to predict the damping force time series. Both harmonic and random vibration inputs are studied with different training functions and node numbers. Based on the model developed from BPNN, the predicted damping force time series is compared to the experimental results. Good agreements are obtained from the comparisons, which validates the prediction of TLDs damping force.