Multivariable predictive neuronal control applied to grinding plants

This work investigates the use of a direct neural network predictive controller applied to a grinding plant. A phenomenological model of the grinding plant is used to simulate the control strategies. The model is based on a mass balance and power consumption of the mill containing 32 particle size intervals. The controller neural network is trained by using an estimation of the error. Several tests are performed driving the nonlinear process to an operation point and then controlling it by training the net online, which enables monitoring of the range over which the neural controller is still valid, without having to conceive a linear model of the process.