For high accuracy in mechanical control, the compensation of friction must be taken into account. When working at low velocities, the slip-stick phenomenon appears, introducing additive perturbation torque. Mechanical and robotics engineers use several models to simulate these torques. Here we select two dynamic friction laws (Dahl model and Reset-Integrator model), and we use Recurrent Neural Networks to model these dynamic systems in order to complete the a priori knowledge with what we don't know how to modelize. We use the canonical architecture to construct our network, and use a gradient based algorithm to train it. Results show that a general architecture for describing this family of friction laws can be obtained. This type of architecture may be used in regulation schemes.
CITATION STYLE
Dominguez, M., Michelin, J. M., & Martinez, J. M. (1995). Recurrent neural networks for identification of friction. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 930, pp. 1060–1067). Springer Verlag. https://doi.org/10.1007/3-540-59497-3_285
Mendeley helps you to discover research relevant for your work.