Inverse model identification of magnetorheological damper using extreme learning machine method

This paper presents a method to build an inverse model for predicting electric current as the input of a magnetorheological damper using extreme learning machine method. The proposed method can overcome the previous methods’ drawbacks, such as the longer training time and possibly to be trapped in local solution. A modified Bouc-Wen model is employed to generate data training and testing at various operating condition. The inverse model inputs are the past force, displacement, and velocity, while electrical current will be the target prediction. The best hyperparameters values for the proposed model will be found by performing several variations of hyperparameters, such as activation functions, and the hidden node numbers. The data is split into 80% of training data and 20% test data. The activation function that best fits this model is sigmoid. The effective number of neuron hidden layer is 2000 neurons. From the variation that has been selected, it is found that this model root mean square error (RMSE) and the R-squared value of and 0.03 and 0.99 for the training process, respectively. Meanwhile, in the testing process, the RMSE and R-Squared values were obtained at 0.03 and 0.99, respectively.