Modelling of Frictional Phenomena using Neural Networks: Friction Coefficient Estimation

Yannis L. Karnavas and Achilles Vairis


Friction modelling, Artificial neural networks, Ti6Al4V, Function Approximation, power system


In this work, an effort is made to model the friction coefficient of sliding surfaces under a variety of temperature, stress and sliding velocity conditions using an artificial neural network (ANN) methodology. First, friction coefficient measurements were obtained for unlubricated similar metal couples of the most commonly used titanium alloy Ti 6Al 4V, for interface temperatures of 20°C up to 900°C, normal stress conditions up to 30 MPa and rubbing velocity between the specimens of 178 mm/s up to 700 mm/s. Next, these measured friction coefficients along with the relevant measured conditions were used to train, in an efficient way, appropriate neural network architecture and further tests were also conducted in order to validate the artificial neural network performance. Two of the most widely known neural network model architectures are being examined in this work and the relevant conclusions and results are discussed and also shown. Through an exhaustive search procedure it is found that, the radial basis function (RBF) type of neural network exhibits the more satisfactory results and seems to be the most appropriate architecture for the friction coefficient estimation of sliding surfaces.

Important Links:

Go Back