Comparison between PID controller and a nonlinear control based on neural network modelling applied to a ball and plate system
In this work we present two proposals to control the position and the path of a sphere on a two degree of freedom flat surface. This pair is known as ball and plate and the first approach consists in a linear approximation for the original dynamical model in order to apply a PID controller to get some metrics and compare to the second approach which is a neural network evolutionary training that make possible include the non linearity of the ball and plate dynamics . Both of theses strategies are applied to a virtual simulation where we can predict the behavior of the sphere based on a set of constitutive equations derived taking into account all rigid body and non-slip bearing simplifications. As the controllers are adjusted, we enter as input the actual position of the ball with a pattern recognition algorithm to mimic a sensing done with cameras commonly seen in experimental benches for this system.