Affiliations: Université de Lyon, Laboratoire de Tribologie et de Dynamique des Systèmes CNRS UMR5513 ECL, 36 avenue Guy de Collongue, Ecully, France | Université de Lyon, Laboratoire de Tribologie et de Dynamique des Systèmes CNRS UMR5513 ECL/ENISE, 58 rue Jean Parot Saint-Etienne, France | SAGEM Défense Sécurité, BP 72, 72-74 rue de la Tour Billy Argenteuil, France
Note: [] Corresponding author. Rosario Toscano, Université de Lyon, Laboratoire de Tribologie et de Dynamique des Systèmes CNRS UMR5513 ECL/ENISE, 58 rue Jean Parot 42023 Saint-Etienne cedex 2, France. E-mail: [email protected]
Abstract: In this paper it is shown that we can estimate the fretting wear evolution via an artificial neural network (ANN) model without making use of the back-propagation learning algorithm and without using any regularization method. This can be done by integrating in the ANN model all the available knowledge about the wear mechanism. This kind of model is referred to as a semi-physical neural model. One of the main advantages in building a semi-physical neural model is that its number of parameters is reduced compared with a standard ANN model. This is a very favourable property against the over-fitting inconvenience. In addition, via appropriate nonlinear transformations, the semi-physical neural model can be rendered linear with respect to the parameters that are to be determined. Consequently, a simple least square approximation can be used to determine the unknown parameters.
Keywords: Fretting Wear Estimation, Neural Network, Semi-Physical Neural Model
