PREDICTION OF FRICTION COEFFICIENT USING ARTIFICIAL NEURAL NETWORK

Abstract

In the present study, friction coefficient of fibre reinforced polymeric composites is predicted by developing neural network model as a function of several input parameters, i.e. the sliding distance which is (0-50km), the applied load (30-50-70-100) N and the sliding velocity (2.8-1.1-2.1-3.1-3.5) M/S. Data sets of all the process parameters of the studied Kenaf fibre are received from an experimental study consists of 1096 data for the friction coefficient at different operating parameters. Several models are developed and the optimum is selected which is based on two hidden layers and a training function of GDM, and transfer function of Purelin-Tangsig and 40 neurons. 70% of the experimental data is used for the training purposes and the rest was used for the verification. In the verification, the error percentage was 20%. The prediction was performed for parameters different than the experimental ones and the data is collected and presented. The correlation between the ANN result and the experimental data have been recorded as 0.8195 which reflect only 20% of error.

Key words: Friction coefficient, Artificial Neural Network, GDM training function.