Optimization of hybrid control systems with seismic isolationand tuned liquid damper and prediction with machine learning

Seismic base isolators are effective control devices in reducing structural acceleration and displacement control. In multi-storey structures, the isolator is forced by the weight of the structure, and the energy is not sufficiently absorbed in the isolation layer and is transferred to the superstructure. Strong vibrations that will occur to seismic isolators may result in excessive movement of the isolator, causing mechanical damage at the isolator connection points or material deformation in the isolation layer. In earthquakes with destructive features such as near fault and causing large displacements, the addition of a damper to isolation systems is recommended. Tuned liquid damper (TLD) systems are control devices that utilize the sloshing of the liquid within them in reducing structural displacements. The hybrid use of these two systems is effective in reducing the acceleration and displacements accumulated in the structure. In this thesis, it is aimed to create a machine learning model that estimates the system properties by considering the optimum design process of a hybrid system consisting of a seismic base isolator and TLD and the size limits of the area where the TLD will be placed. In the first application, a seismic isolator was placed at the base of a structure under near-fault earthquakes and TLD was placed at the top floor, and the effect of TLD addition to the isolator on the structural control was investigated for different damping and mobility capacities of the isolator system. Isolator and damper properties were optimized with an adaptive harmony search algorithm to reduce the maximum acceleration of the sturcture. It has been determined that the addition of TLD has preventive effects on the isolator from being damaged by limiting the displacements at the isolator level of the near fault. In the second application carried out in the thesis study, a machine learning models was developed that estimates the design parameters of a structure with a hybrid control system with seismic isolator and TLD, independent of the building mass. The earthquake behavior of a standard structure under pulsed near-fault earthquakes was simulated via Matlab Simulink, and the optimum design parameters and structure behavior were recorded with the Jaya algorithm optimization. A data set was created with the optimum design properties and various parameters obtained. Considering the size limits of the area where the control system will be placed and the liquid properties, machine learning models that predict TLD tank dimensions, isolator and TLD damping ratios and periods were produced. The most effective algorithms were determined among the developed learning models, and it was determined that among the predicted variables, TLD tank radius, liquid height and isolator period properties could be predicted with an accuracy exceeding 96% and a mean absolute error around 8%.