Accurate estimation of friction

The effect of friction in low velocity, precise, position controlled mechanisms can be dominant and is difficult to model. This study examines the use of neural networks to model friction in mechanical systems that use DC motors for actuation. Neural network designs alleviate the need to select a friction model and determine its parameters. In addition, since neural networks are inherently capable of learning, the method can be used on-line to compensate for changes in friction characteristics. Simulation studies are performed on a DC motor load case.