Metamodeling and Optimization of a Blister Copper Two-Stage Production Process

It is often difficult to estimate parameters for a two-stage production process of blister copper (containing 99.4 wt.% of Cu metal) as well as those for most industrial processes with high accuracy, which leads to problems related to process modeling and control. The first objective of this study was to model flash smelting and converting of Cu matte stages using three different techniques: artificial neural networks, support vector machines, and random forests, which utilized noisy technological data. Subsequently, more advanced models were applied to optimize the entire process (which was the second goal of this research). The obtained optimal solution was a Pareto-optimal one because the process consisted of two stages, making the optimization problem a multi-criteria one. A sequential optimization strategy was employed, which aimed for optimal control parameters consecutively for both stages. The obtained optimal output parameters for the first smelting stage were used as input parameters for the second converting stage. Finally, a search for another optimal set of control parameters for the second stage of a Kennecott-Outokumpu process was performed. The optimization process was modeled using a Monte-Carlo method, and both modeling parameters and computed optimal solutions are discussed.