RBFNN Hybrid Modeling And Optimizing Of Cobalt Oxalate Particle Size Distributions

In the thesis, the modeling and optimizing of cobalt oxalate particle size distributions (PSD) in its synthetic process is studied, which is a subproject of the development of the optimal control software system in hydrometallurgical process-one of "863" National High-tech Research Projects. Hydrometallurgy is a metallurgical method, which is used for the extraction and separation of metal in the raw material by chemical reactions.Cobalt is a strategic metal, which is one of the main raw materials for hard alloy. With the ultra-fine hard alloy industrial development, not only is a higher quality of chemical composition of cobalt powder demanded, but also the physical properties of cobalt powder such as particle size, PSD, are required. Not only is the particle size needed to meet the requirements, but also the distribution is required to be concentrated. Cobalt powder is generally acquired through decomposition of cobalt oxalate. The shape and particle size of cobalt oxalate decide the shape and particle size of cobalt powder. At present in China the size of cobalt powder and cobalt oxalate is described by fill density, which is inherited from former Soviet Union. But this method can not meet the increasing needs of the development of hard alloy. So, researching on the rules of changes in the size of cobalt powder and cobalt oxalate is becoming a new issue. So far, researches on modeling and optimizing of cobalt oxalate PSD have not been reported in China. In most cases, appropriate operating conditions corresponding to different fill density are obtained by experimenting, which is high cost and low efficient.Oxalate cobalt is acquired by precipitation and crystallization in the synthetic process of cobalt hydrometallurgy. In the thesis, with deeply understanding of cobalt oxalate synthetic process and according to the practical production, the key parameter and operational variables in the synthetic process of cobalt oxalate are established. The key parameter is PSD of cobalt oxalate and operational variables are crystallization temperature, stirring speed, the concentration and flow rate of ammonium oxalate.In the thesis,with deeply understanding of cobalt oxalate synthetic process and applying reaction kinetics, crystallization kinetics, the population balance theory and the mass balance theory of reactive precipitation process, the mechanism model of cobalt oxalate PSD corresponding to operational variables is established. The operational variables include crystallization temperature, stirring speed, the concentration and flow rate of ammonium oxalate. State equations are the population balance equations and the mass balance equation. The outputs are average particle size and relative volume-based variance. The process described by the model is dynamic, but eventually tends to be steady and the outputs during the process are difficult to obtain in the practice manufacturing field, so only the steady state value is chosen in simulation. Based on the theory of particle size classification, the mass balance equation and the population balance equations are converted into ordinary differential equations and simulation is conducted to find the rules of cobalt oxalate PSD corresponding to operational variables changes. The trends are abtained according to the simulation, which are verified by practical production.Because of the complexity of the crystallization process, the crystallization parameters being incomplete and the mechanism model being constructed by many simplified assumptions, the accuracy of the mechanism model is difficult to meet the requirements. So in this thesis, the mechanism-radial basis function neural network (RBFNN) parallel hybrid model is established combining mechanism model and RBFNN, which is more practical. RBFNN centers are selected by dynamic adaptive clustering method.The width of RBF is the standard deviation of each cluster. The linear output weights are obtained by least square.The final RBFNN is simplified by delete policy. The simulation is conducted in the process of hybrid modeling, and the results indicate the prediction accuracy of the mechanism-RBFNN parallel hybrid model is better than the prediction accuracy of the mechanism model and the black-box model.Based on the hybrid model and deeply understanding of the synthetic process, the optimization model of cobalt oxalate PSD is obtained by establishing the objective function and constraints including model constraints, production rules and capacity constraints, quality and yield constraints. Optimization model is converted into a new optimization model based on the theory of penalty function. The cobalt oxalate PSD is optimized by adaptive genetic algorithm. The optimization results show that the optimization model established is effective to optimize the PSD of cobalt oxalate to obtain better products whose average particle size meet the requirements and relative volume-based variance are smaller. The optimization of cobalt oxalate PSD is of great importance to the industrial production of cobalt oxalate.