Determining silica solubility in Bayer process liquor

JOM, Nov 1998 by Jamialahmadi, M, Muller-Steinhagen, H

The efficient precipitation of dissolved silica from Bayer process liquor is essential for the production of high-quality alumina and the reduction of excessive scaling in the heat exchangers in the evaporation building of Bayer processes. The accurate prediction of silica solubility in Bayer liquor is one of the key parameters in improving the design and operation of the desilication process. Previous findings, particularly with respect to the influence of temperature and concentrations of caustic soda and alumina on the solubility of silica, are inconclusive. In this article, experimental results are presented over a wide range of temperature and alumina and caustic soda concentrations. Attempts are made to utilize artificial neural networks for identifying the process variables and modeling. The radial basis function neural network architecture was used successfully to generate a nonlinear correlation for the prediction of the solubility of silica in Bayer process liquor. The resulting correlation can predict the present data and the control data of other investigators with good accuracy.

INTRODUCTION

The majority of aluminum produced today is manufactured from bauxite, which is an ore or mixture of minerals formed by the weathering of aluminum-bearing rocks. The parent rocks, which may be igneous or sedimentary, are often nepheline (3Na20K204Al2039SiO2), serpentine, granite, diorite, basalt, dolerite, crystalline schists, and limestone containing clay minerals. When these silicate rocks are subjected to a weathering process, they tend to lose such constituents as silica, magnesia, lime potash, and soda and leave behind a residue richer in alumina, iron oxide, and titania than the original rock. Consequently, clays containing 35-50% combined alumina and, subsequently, laterites consisting substantially of hydrated iron oxide are formed.

Hydrated alumina occurs inbauxite as gibbsite (A12033H20),boehmite (Al2032H20), and diaspore (A1203H20). The main impurities of bauxites are compounds of silicon, iron, and titanium. Silicon occurs as kaolinite (Al2032SiO2.2H20) and halloysite (Al,O32SiO23H20). Silica, in the form of quartz, is not perceptibly attacked during the extraction of bauxites in the Bayer process; however, silica combined as clay or other silicates (known as reactive silica) dissolves in caustic soda in the digestion units. It then reacts with soda and alumina in solution and partly precipitates as sodiumaluminum silicates (the desilication product [DSP) that are removed from the process together with the insoluble iron and titanium oxides in the red mud. It is desirable to achieve as much desilication as possible during the digestion process in order to minimize impurities in the final alumina product and reduce the potential of the solution for scale formation in the rest of the process.

Cb and C* are the bulk and saturation concentrations, respectively. Equation 1 shows that regardless of the mechanism of the desilication process, the effect of concentration is very strong. Consequently, accurate knowledge of the saturation concentration, C*, is essential for the reliable prediction of the degree of supersaturation of the liquor and the rate of DSP deposition. Several correlations have been proposed in the literature for the prediction of silica solubility in Bayer liquor. Descriptions of these correlations and the conditions for which their applications have been recommended are summarized in Table I.

These equilibrium expressions are empirical and have been obtained by curvefitting a limited number of experimental data. Therefore, it is not possible to reliably predict the effect of parameters such as temperature and composition on the solubility of silica in the Bayer liquor. The lack of reliable information on silica solubility is probably not only due to the lack of experimental work, but also to the presence of various amounts of inorganic and organic impurities accompanying the hydrated alumina (Table II).

The presence of trace impurities may have a significant effect on the solubility of silica in the Bayer process, which makes it almost impossible to develop a mathematical model from thermodynamic principles. The major complication results from the complex stoichiometry and the thermodynamics of an enormous number of possible reactions. Therefore, the solubility data have to be updated periodically as plant liquor composition changes. Modeling of such a system requires sufficient knowledge about the underlying processes, which determine the system's overall performance. Furthermore, a large number of experimental parameters need to be estimated, and sophisticated computing techniques must be employed to obtain the required precision of the model.

An alternative approach to modeling and identifying influencing parameters is the application of artificial neural networks. In recent years, artificial neural networks have emerged as powerful tools for modeling complex processes. These networks are non-algorithmic, analog, distributive, and massively parallel information processing methods that have proven to be powerful pattern-recognition tools. Since they process data and learn in a parallel and distributed fashion, they are able to discover highly complex relationships between several variables that are presented to the network. As a model-free function estimator, neural networks can map input to output no matter how complex the relationship.