Artificial neural network prediction of the aluminum extraction from bauxite in the Bayer process

Abstract

This paper presents the results of statistical modeling of the bauxite leaching process, as part of Bayer technology for an alumina production. Based on the data, collected during the period between 2008 - 2009 (659 days) from the industrial production in the alumina factory Birac, Zvornik (Bosnia and Herzegovina), the statistical modeling of the above mentioned process was performed. The dependant variable, which was the main target of the modeling procedure, was the degree of Al2O3 recovery from boehmite bauxite during the leaching process. The statistical model was developed as an attempt to define the dependence of the Al2O3 degree of recovery as a function of input variables of the leaching process: composition of bauxite, composition of the sodium aluminate solution and the caustic module of the solution before and after the leaching process. As the statistical modeling tools, Multiple Linear Regression Analysis (MLRA) and Artificial Neural Networks (ANNs) were used. The fitting level, obtained by using the MLRA, was R2 = 0.463, while ANN resulted with the value of R2 = 0.723. This way, the model, defined by using the ANN methodology, can be used for the efficient prediction of the Al2O3 degree of recovery as a function of the process inputs, under the industrial conditions of the alumina factory Birac, Zvornik. The proposed model also has got a universal character and, as such, is applicable in other factories practicing the Bayer technology for alumina production.