Experimental and numerical study of the forward and inverse models of an MR gel damper using a GA-optimized neural network

Abstract

In this paper, we present a series of experimental and numerical studies on the performance and modeling of a developed magnetorheological gel (MRG) damper. A bi-directional shear-type damper was designed and fabricated. The MRG damper, which utilizes the gel’s high viscosity, can effectively alleviate the settlement problem inherent in magnetorheological fluid damper applications. Then, dynamic performance experiments were carried out to obtain the damping force with sinusoidal and random displacement excitations. Based on the test results, the forward model of the damper was established using a backpropagation neural network. A genetic algorithm was employed to optimize both the network structure parameters and the initial weight and bias values. Different forward models generated using different training datasets were validated and compared with the RBFNN model and Bouc-Wen model using different test datasets. The validation results indicate that the neural network-based forward model greatly outperforms the RBFNN model and Bouc-Wen model in terms of the estimation performance. The influence of the inputs at previous time has also been investigated. Finally, to generate the command current for controlling the damper, inverse neural network models with optimized structure parameters were established using different training datasets. Validation results with different test datasets indicate that, although the predicted current generated by the inverse models has many high-frequency components, it can still act as an effective damper controller, with the resulting damping force calculated using the predicted current coinciding well with the desired behavior.