Model reference control of a magneto-rheological damper using neural networks

By actively adjusting its magnetic field strength via control voltage, an MR damper can generate damping forces in a desired fashion, for example, behaving like a Coulomb friction damper. The hysteresis in magneto-rheological damper needs to be compensated in order to make its control more accurate. This paper presents a new approach to control of MR damper using model reference control and neural networks. A modified Bouc-Wen model of the hysteresis in an MR damper is used to generate training data for the forward neural network model of the hysteresis. Upon successfully training of the neural network model, neural network controllers are designed to control the MR damper so its output force follows a reference model without hysteresis. Using this model reference control and neural networks approach, piecewise linear relationship between damping force and applied voltage is achieved and this piecewise linear relationship greatly simplifies the control of an MR damper.