Multi-parameter Sensitivity Analysis And Hybrid Genetic Algorithm Based Parameter Estimation Method For Dynamic Model Of Ball Mill

Ball milling process is an important operation unit in the beneficiation process, which grinds the comminuted ores to small particles and liberates the valuable minerals from gangue minerals for classification process and then beneficiation process. This process has synthesized complex characteristics, such as nonlinear, large time-delay, slow time-varying, many key process parameters cannot be measured online. These characteristics greatly increase the modeling difficulty of ball mill.This paper is supported by National973Program "Complex Manufacturing Process Integration Control Theory and Technology Basic Research" of State Key Laboratory of Synthetical Automation for Process Industries in Northeastern University. For the dynamic characteristics of ball milling, parameter estimation method for dynamic model of ball mill is deeply studied so that the dynamic model of ball mill is established, which promotes the application of model-based control and optimization methods.The details are as follows:(1) Based on the analysis of dynamic characteristics of ball milling, empirical parameter-based dynamic model of the ball mill is established. It includes pulp flow equation of mill output, pulp volume equation inside the mill, pulp concentration equation inside the mill, population balance equation of each particle inside the mill, grinding kinetic equation etc. Experiment results show that the established dynamic model of the ball mill can accurately reflect the qualitative characteristic of mill discharge pulp concentration, particle size distribution, etc.(2) Based on the above dynamic model of ball mill, a parameter estimation method based on multi-parameter sensitivity analysis and hybrid genetic algorithm for dynamic model of ball mill is proposed. Firstly, multi-parameter sensitivity analysis is employed to evaluate the relative importance of the7parameters in selection function and breakage function. And then hybrid genetic algorithm is used to estimate the important model parameters using the practical data of grinding process. The unimportant model parameters are fixed by operation experience. Dynamic model of ball mill is then obtained. Experiment results show that the mean square error on test samples of the proposed method is small and the proposed method has a good prediction performance. Finally, by analyzing the estimation results of the hybrid genetic algorithm, generalized likelihood uncertainty estimation is applied to analyze the uncertainty of model parameters. Moreover, the proposed method can decrease the parameter number to be evaluated and thus reduce time taken by parameter estimation.(3) Experimental study is preformed. It includes step response of ball mill model for step changes in ore density, step response of ball mill model for step changes in ore flow, and step response of ball mill model for step changes in water flow. Experiment results show that the dynamic characteristics of mill discharge pulp concentration, particle size distribution and so on accord with practical process.