Prediction of the tractor tire contact area, contact volume and rolling resistance using regression model and artificial neural network

Abstract

A novel method to estimate the contact area and contact volume was developed with molding the tire footprint by liquid plaster and converting these molds to three-dimensional models using a 3D scanner. A 12.4-28, 6 ply tractor tire was operated under three levels of vertical load, three levels of inflation pressure and three levels of soil moisture content. To analyses the obtained data regression and Artificial Neural Network (ANN) models were used and the accuracy of predicted results were compared with measured data. A multi-layer perceptron feed-forward ANN with back propagation (BP) learning algorithm was employed. Two hidden layers were used in network architecture and the best number of neuron for each hidden layer was selected with attention to minimum RMSE criterion. The results showed that tire contact volume is a better parameter than tire contact area to predict rolling resistance. The comparison of the results of regression and ANN models to predict the contact area, contact volume and rolling resistance showed that ANN predictions had a closer agreement with the measured data than the regression model predictions.