Abstract: In mineral beneficiation,the crushing and grinding process is the key material preparation stage. Aiming at the problem that the intrinsic parameters of the equipment are not considered in the mathematical model modeling method of the traditional crushing and grinding process,the mature matrix model is adopted in the crushing stage. Meanwhile,the multi-feature fusion ability of neural networks is utilized to process the structural characteristics of the equipment and the distribution of feed particle size in the grinding process,and the Back Propagation-Matrix Model ( BP-MM ) is established. Taking the short process of crushing and grinding sample preparation as an example,the matrix model of jaw crusher and roll crusher is constructed based on the crushing experimental data. The BP neural network model of disc mill with the gap parameters of grinding disc is constructed by neural network method,and the BP-MM hybrid model of short process of crushing and grinding sample preparation is constructed. Taking the mean absolute error,root mean square error and determination coefficient as evaluation indicators,the prediction results of the BP-MM hybrid model were compared with the simulation results of JKSimmet. The results demonstrated that the BP-MM hybrid model achieved prediction errors below 3%. When the gap of the grinding disc is 0. 1 mm,the particle size of the grinding product in the short process of crushing and grinding sample preparation is ≤0. 15 mm. The BP-MM hybrid model modeling method can handle the input of multi-feature and multi-parameter fusion,effectively improve the modeling accuracy and prediction performance,and provide new ideas for the control optimization of crushing and grinding process. 

Key words: crushing and grinding process,matrix model,neural network,particle size distribution