DEVELOPMENT OF AN INTELLIGENT CONTROL SYSTEM FOR THE ORE GRINDING PROCESS BASED ON AN ARTIFICIAL NEURAL NETWORK

This article addresses the design of an intelligent controller for ore grinding that is built around an artificial neural network. Two indicators are used to describe the state of a ball mill: the deviation of the mill load from its set value and the deviation of the product particle size from its target. On the basis of these two indicators, a five-layer feedforward network is proposed, with ReLU activation in the hidden layers and Tanh in the output layer. The weighted sums of the neurons, the matrices of synaptic weights and the bias terms are written out in mathematical form. For training, the backpropagation algorithm is combined with the Adam optimizer. The network output yields the control law Ut, which adjusts the ore feed rate.