Efficacy of ANN and ANFIS as an AI Technique for the Prediction of COF at Finger Pad Interface in Manipulative Tasks

Abstract

Current work intends to compare the modelling ability of two popular artificial intelligence (AI) techniques, namely artificial neural network (ANN) and adaptive-neuro fuzzy inference system (ANFIS). Outcome of study is useful in prediction and further optimization of the coefficient of friction in the design of assistive devices for an ergonomics and comfort of the user. Experiments were conducted using Taguchi L16 design of experiments (DOE). Total of 16 experimental runs were conducted. Two extrinsic factors normal load (2, 4,6, & 8 N) and sliding velocity (4, 6, 8 & 10 cm/s) that affect the finger pad friction are taken as input variables, while coefficient of friction (COF) between finger pad and the stainless steel (SS) probe is the output variable. ANN with 2 inputs, 10 hidden, and 1 output layer is trained by three algorithms, viz. Levenberg–Marquardt (R2 = 0.96), Bayesian Regularization (R2 = 0.93), and Scaled Conjugate Gradient (R2 = 0.98) based on the correlation coefficient. Although, both the techniques highlight significant predictability and accuracy, ANFIS results shows overfitting of the data. Hence, ANN technique is relatively better than ANFIS.