Artificial Neural Network (ANN) Approach to Predict LWST Values from Friction and Texture Measurements

Abstract

The paper aims to find whether friction values namely skid numbers obtained by the Locked Wheel Skid Trailer (LWST) device can be predicted using values obtained by the Dynamic Friction Tester (DFT) and the Circular Texture Meter (CTM). The last two measure the coefficient of dynamic friction (called DFTx) at different speeds (labeled x) and the Mean Profile Depth (MPD), they also measure the International Friction Index (IFI) parameters F60 and Sp. Artificial Neural Network (ANN) software was used to investigate the relationships. Twelve (12) different models were proposed with different input parameters and the best model giving the highest coefficient of determination (R2) was discussed in this paper. The results show that the most influential factors on LWST friction values are MPD, DFT0, DFT50, and DFT64 and MPD was the strongest among them. In addition, results show that the ANN approach is very efficient in predicting the LWST friction values for both training and validation sets with R2 values of 79% and 83%, respectively. It was also shown that the IFI parameters were relatively less influential on LWST values than DFT and MPD measurements.