| Summary: | A meandering magnetorheological damper's produced damping force or pressure drop is correlated with the inputted current. Although the relationship can be considered proportional, the predicted force's hysteresis influence led to the difficulty of the modeling process. Artificial Neural Networks (ANN) can be a solution to predict the behavior of a meandering magnetorheological damper. This paper presents an investigation of the application of ANN on the force prediction of a magnetorheological damper using various hidden layer numbers. The model contains three inputs: displacement, velocity, and electrical current, and one output, which is force. Firstly, the experimental data is prepared and pre-processed to eliminate outliers and to normalize the input range. The prepared data is divided into training and testing data and inputted into the training algorithm. Various hidden layer numbers' effects on the model accuracy have been investigated. The visual observations on the figures of force-displacement and force-velocity also have been conducted to check the predicted hysteresis and pattern at various currents. The results show that layers two, three, and four can produce better results than one layer ANN with mean square errors above 0.010. In conclusion, the proposed method can accurately predict the damping force, considering the R-squared value of 0.9997.
|
|---|