NONPARAMETRIC MODELING OF MAGNETORHEOLOGICAL DAMPER

With the unique advantages such as low power requirement and adequately fast response rate, Magneto-Rheological (MR) dampers have been employed in vibration control for civil engineering structures. An accurate and efficient model of MR damper is needed for effective control system design. Even several mathematical parametric models of MR damper have been proposed in the last years, it is still a challenging problem to model the behaviour of MR damper accurately. Moreover, some existing parametric models are too complex for control system design. Because of its nonlinear mapping capability, parallel computation, and adaptability, artificial neural network (ANN) provides an alternative way to describe the actual performance of MR dampers. In this study, two types of three-layer neural networks with different input variables were constructed respectively to model a MR damper at different excitation currents using teat data. The performance of the two proposed ANN based models is validated. The results show that the predicted damping forces of the two neural network models match well with the forces measurement, which demonstrates that the proposed neural network model can provide a computationally efficient way to model the behaviour of a MR damper.