Magnetostriction behavior modeling of magnetorheological foam using data-driven neural network algorithm

Abstract

Magnetorheological (MR) foam is a magnetic polymer composite (MPC) that has the potential to be used for the application of soft sensors and actuators in robotics due to its tuneable mechanical properties and magnetostriction. Material development has recently become challenging since it is both time-consuming and costly. As such, it is crucial to model the mechanical properties and magnetostriction of MR foam to expedite the development of MR foam devices. As a consequence, extreme learning machine (ELM) and artificial neural network (ANN) machine learning models for predicting the magnetostriction behavior are performed. These models were developed to describe the non-linear relationship between different carbonyl iron particles (CIP) compositions and magnetic field as inputs, whereas strain and normal force as outputs. The model had variation hyperparameters, such as different learning algorithms and activation functions. For ANN, RMSProp and ADAM learning algorithms were applied with two different activation functions, sigmoid and ReLU. The ELM model, on the other hand, considered the Hard limit (HL), ReLU and sigmoid activation function. Then, the model was assessed for both training and testing datasets. Based on the results, RMSProp with activation function sigmoid of ANN model showed an agreeable accuracy with the experimental data compared to the other models. However, the correlation analysis and comparison between prediction and experimental data showed that ELM HL was more generalized in predicting strain and normal force with R2, 0.999 and root mean square error (RMSE) less than 0.002 respectively. In conclusion, the ELM HL model successfully predicts the magnetostriction behavior of MR foam at various compositions that could be applied in the development of MR foam devices in the near future.