Application of Decision Trees for Wave-Breaking Prediction of Tuned Liquid Dampers

摘要

Owing to their low cost, ease of design and implementation, Tuned Liquid Dampers (TLDs) are becoming increasingly popular in suppressing structural vibration under wind and seismic loads. The ability to accurately predict their dynamic properties is a key task in their optimum design. However, due to the highly nonlinear behavior (e.g. wave-breaking during the sloshing of the liquid inside the tank) of these devices, it is difficult to establish accurate analytical or numerical models for a wide range of operation. Currently, some of the existing models use either the wave height or a ratio of the tank length to the amplitude of the displacement it goes through as parameters to reveal the occurrences of wave-breaking. There is a need to understand the parameters that control the wave-breaking occurrence, so that the models can be improved accordingly. By carrying out 125 tests that consider TLDs with different dimensions, height of water, etc. under sinusoidal excitations with different amplitudes and frequency ratios, an experimental database is developed where the wave-breaking incidences were identified with the help of a high-speed camera. Subsequently, the data is processed using decision trees to identify the parameters that influence wave-breaking. The main advantage of the decision trees over other soft computing tools is their easier use and the comprehensible mathematical rules they produce. The decision tree results in predicting the wave-breaking occurrence were compared to earlier studies and based on these results recommendations have been made to modify the models in one of these studies.