LAUSR

Abstract The performance of tuned mass dampers (TMDs) relies on a tuning process in which the parameters are optimized for one or several objectives. For this purpose, several formulations and numerical optimization methods have been developed. The existing closed form formulation may not be suitable for a desired objective, and numerical optimization methods must be applied to every different system. For that reason, optimum TMD parameters such as period and damping ratio can be estimated by using machine learning. Since the use of machine learning in TMD optimization is a novel field of research, a well-known Artificial Neural Network (ANN) model is proposed. Training of the ANN was done by using optimum TMD values of several sets of single degree of freedom (SDOF) structures. The optimum values were found according to an optimization process using a flower pollination algorithm (FPA) and transfer function amplitude as an objective in the frequency domain. The generated ANN model is generally effective to find the same optimum results as the FPA approach up to two decimal places. Also, the ANN model was used to generate three basic tuning formulations, which are tested on SDOF and multiple degrees of freedom (MDOF) structural models. Including the responses under seismic excitation by considering stroke of the TMD, the optimum TMD parameters defined according to the developed formulations are more effective than optimum results found by existing formulations for reduction of maximum structural responses.