Modeling and semi-active control performance analysis of MRE isolator based on BPNN optimized by GA using different evaluation indexes

Abstract

The utilization of isolation systems employing Magnetorheological Elastomer (MRE) devices holds significant promise for structural vibration applications due to their customizable stiffness and damping characteristics. However, the nonlinear dynamics inherent in MRE isolators present formidable obstacles for the establishment of accurate models and development of effective control strategies for practical implementation. In this work, the dynamic properties of a self-made MRE isolator under different loading conditions are tested and analyzed. Then the nonparametric forward model and inverse model of MRE isolator based on BPNN (back propagation neural network) are established respectively, and the GA (genetic algorithm) is used to optimize the neural structure of BPNN. The precision and accuracy of the forward and inverse model is verified by comparing the predicted and experimental data. Simulation and experimental results show that the BPNN optimized by GA can efficiently and accurately model the nonlinear behavior of MRE isolators. Based on this, we take an eight-story shear frame building based on the proposed MRE vibration isolator model as the research object and numerically studied the vibration suppression effect under the control of three typical control algorithms, that is, LQR, FC, and FC-PID. In the process of evaluating the effect of vibration isolation control, in addition to utilizing the traditional displacement amplitude, layer distance and acceleration as evaluation indexes, we also propose a new comprehensive evaluation index which has weighting coefficients and considers the input cost. It is shown that the newly proposed comprehensive index can more conveniently compare the advantages and disadvantages of different control algorithms, and the fuzzy PID is the most suitable among the three control algorithms.