Study of extreme learning machine activation functions for magnetorheological fluid modelling in medical devices application

Magnetorheological (MR) fluid applications in various medical equipment have been widely studied because its easiness to utilize and fast response. The application can be divided into, at least, two forms, which are haptic and prosthetic devices. In each equipment design process, rheological models are essential to determine the required inputs to produce enough force or yield stress. However, each existing model has its own limitations, such as agreeable performance on limited inputs ranges of magnetic fields and shear rates. A modeling method using extreme learning machine (ELM) as an intelligent model may be able to solve this problem. Therefore, this paper aims to investigate the ELM performance to model MR fluids behavior using various activation functions. Five activation functions are applied, which are hard limit, sigmoid, sine, triangular basis and radial basis function. Then, the investigation is divided into two cases, which are a wide and low operating shear rate based on the medical devices applications. The comparisons with the experimental data show that the models produce considerably low RMSE or less than 3 kPa, especially for hard limit activation function. The paper also demonstrates the model capability to predict dynamic yield stress.