MODELLING OF FRICTIONAL PHENOMENA USING NEURAL NETWORKS:FRICTION COEFFICIENT ESTIMATION

摘要

In this work, an effort is made to model the frictionrncoefficient of sliding surfaces under a variety ofrntemperature, stress and sliding velocity conditions usingrnan artificial neural network (ANN) methodology. First,rnfriction coefficient measurements were obtained forrnunlubricated similar metal couples of the most commonlyrnused titanium alloy Ti 6Al 4V, for interface temperaturesrnof 20°C up to 900°C, normal stress conditions up to 30rnMPa and rubbing velocity between the specimens of 178rnmm/s up to 700 mm/s. Next, these measured frictionrncoefficients along with the relevant measured conditionsrnwere used to train, in an efficient way, appropriate neuralrnnetwork architecture and further tests were also conductedrnin order to validate the artificial neural networkrnperformance. Two of the most widely known neuralrnnetwork model architectures are being examined in thisrnwork and the relevant conclusions and results arerndiscussed and also shown. Through an exhaustive searchrnprocedure it is found that, the radial basis function (RBF)rntype of neural network exhibits the more satisfactoryrnresults and seems to be the most appropriate architecturernfor the friction coefficient estimation of sliding surfaces.