Study of Distribution of Magnetic Field Strength in Magnetorheological Elastomers using Artificial Neural Network

Abstract

Magnetorheological elastomers (MRE) find a large number of applications in vibration control. For effective working of MRE, proper design of a magnetic circuit is needed so that it can draw the maximum amount of magnetic field through MRE region and at the same time satisfy all geometric constraints for the related application. With the availability of B-H curve data for MRE, finite element (FE) modelling of the full magnetic circuit is possible. FE modelling presents a good perspective on the distribution of the magnetic field inside the circuit. However, FE results cannot be obtained frequently for different geometries of the circuit. In the present study, the feedforward neural network (FNN) is used to predict the magnetic flux density in the MRE region of magnetic circuit setup. Training data for the FNN is generated from the results of the FE model where a log file has been coded keeping geometric parameters as variables and magnetic flux density as output. The FNN model is trained to have minimum mean square error. The FNN is further used for the optimization of dimensions of the magnetic circuit to have maximum magnetic field in the MRE region for the given geometric constraint.