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

Comparing the linear and logarithm normalized extreme learning machine in flow curve modeling of magnetorheological fluid

“…Both Lin and log has the same order in terms of accuracy. For linear case, this result is consistent with the result from the previous paper [25]. Meanwhile, the biggest gap between log and linear is achieved by ELM Sin.…”

“…The flow curve is obtained based on the rotational test on a parallel plate rheometer (Anton Paar) as reported in [25]. The tested MR fluid is manufactured by CK, South Korea called MRC C1L with the properties as described in [25]. The applied magnetic fields are 830 mT, 770 mT, 420 mT, 310 mT, 200 mT, and 0 mT.…”

“…If one wants to cover other magnetic fields, a polynomial model of yield stress needs to be applied [22], [23]. The limitation can be solved using machine learning, such as Artificial Neural Networks (ANN), Support Vector Machine (SVM) [24] and Extreme Learning Machine (ELM) [25]. ANN and SVM have been known for their limitations, which are slower training time and have more possibility to be trapped at local minima or extrema [26].…”

“…The possible improvement is the activation function, normalization method and the pre-processing of the inputs. Meanwhile, the activation function effects were investigated in [20], [25]. However, the normalization effect at the ELM-based rheological model is only discussed separately and not comprehensively.…”

“…However, the normalization effect at the ELM-based rheological model is only discussed separately and not comprehensively. While linear normalization has been utilized in [25], [27], [28], logarithm normalization has been introduced in [20]. Since these previous works focus only on the introduction of the extreme learning machine platform for various important parameters prediction, the logarithm normalization influence on the model performance has not been discussed in detail.…”