The study of establishing the NN models for operating conditions inside grinding mill and response factors outside

The grinding process has many uncertain factors and nonlinear characteristics that are hard to describe precisely. This dissertation is based on our researches about operating conditions inside grinding mill (which are the charge volume ratio of balls to the mill, the weight ratio of solid to water, the fraction of the spaces between the balls at rest which is filled with powder) and response factors (sound intensity and spectrum distributions) outside, and has been established the neural network models of the grinding mill by using radial basis networks with the ability of approximating the nonlinear function. Besides we have done further investigation of the shortages of conventional models, we use the trial results to train the neural networks again with its learning capability, to give the networks a more precise approximate ability (sum-squared error less than 0.02). Compared with the conventional models, the simulating results had showed that the neural networks not only have achieved very good approximation capability at the associated targets, but also have very good approximation capability between them