Controle de qualidade de bauxitas gibbsíticas: predição dos parâmetros AvAl2O3 e RxSiO2 a partir de dados difratométricos por reflexão e transmissão utilizando estatística multivariada

Abstract

Currently, traditional wet chemistry methods are used for quality control of bauxites. Such methods indirectly quantify the gibbsite and kaolinite content as available alumina (AvAl2O3) and reactive silica (RxSiO2), respectively, and they are very costly and time-consuming. In order to achieve a rapid and reliable method to estimate these parameters, as alternative to current wet chemistry methods, in this work it is evaluated the use of multivariate statistics – Partial Least Square Regression (PLSR) on XRD data of Brazilian bauxites. The X-ray diffractograms were collected in the reflection and transmission modes, and the data collected by each of these treatments were compared with respect to the quality of the PLSR models. The method was optimized through Principal Component Analysis (PCA) and Factorial Design of Experiments (DOE), from which it was possible to identify outliers, grouping samples with mineralogical similarities into three clusters (C1, C2 and C3), and obtain optimized parameters for the collection and pre-treatment of diffractograms. The best results were obtained using the reflection mode, reducing the 2θ range to 13º – 34º 2θ, increasing the step size from 0.026º to 0.065º, and using standardized data. These collection conditions, although not ideal for most XRD applications, provided both a better accuracy of the predictive models of AvAl2O3 and RxSiO2 and a reduction in the collection time (~ 40 seconds). The results showed that the precision obtained was within the industrially acceptable limits for the quality control of gibbisitic bauxites (AvAl2O3 = 0.49% and 0.83%, and RxSiO2 = 0.32% and 0.23%, respectively for samples of the groups C1 and C2). The prediction was not satisfactory only for marginal bauxites samples (grouped mainly in C3). XRD by transmission allowed the elimination of the preferential orientation effect, however, the accuracy of the model was acceptable only for C1 samples. Compared to the traditional wet chemistry, the proposed method is significantly faster, easier to implement and perform the analyzes, requires less space and manpower, besides no chemical reagents are needed. In addition, with the implementation of X-ray diffraction in the laboratory of the bauxite and alumina industry, it is possible to follow the mineralogy of the ore that feeds the Bayer process and, therefore, to be aware of how variations in the mineralogical composition can impact the process. It is worth noting that such information is still unknown, controlling only the chemical parameters.