Evaluation of artificial neural network-based formulations for tuned mass dampers

Abstract

Tuned mass dampers (TMDs) are used to damp vibration of mechanical systems. TMDs are also used on structures to reduce the effects of strong forces such as winds and earthquakes. For the efficiency of TMD, optimization of TMD parameters is needed. Several classical formulations were proposed, but metaheuristic methods are generally used to find the best result. In addition, the metaheuristic based optimum results are used in machine learning of artificial intelligence-based models like artificial neural networks (ANN). These ANN models are also used in development of tuning equation via curve fitting. The classical and ANN-based formulations were found according to frequency domain responses. In the present study, the classical and ANN-based formulations were evaluated by comparing on time-history responses of seismic structure. In comparison, near-fault ground motion records including directivity pulses are used. The ANN based methods have advantages by providing smaller stroke requirement and damping for TMDs.