Prediction of filtration characteristics by means of multivariate data analysis

The behavior of solid-liquid suspensions during filtration processes is influenced strongly by the properties of the particles and the liquid phase. Although the prediction of common filtration characteristics, such as cake resistance, cake porosity and compressibility, has been studied extensively, general theoretical models that could be applied for complex real-life suspensions do not exist. Prediction of filtration characteristics has proved to be very difficult even in those cases where an extensive set of experimentally obtained material data is available. This is mainly caused by the complexity of the cake formation process which means that the number of influential variables that should be simultaneously considered in the models is large. Traditionally applied calculation and modelling techniques have been incapable for processing large sets of input variables which has consequently restricted the complexity of the models. This paper introduces an alternative procedure for predicting the filtration characteristics of solid-liquid suspensions from the measured material properties. Multilinear partial least squares regression (N-PLS) was applied for creating empirical models between the experimentally determined pressure filtration parameters and the particle size and shape data obtained by an automated image analyzer. The density and dynamic viscosity of the liquid phase were also included in the models as input variables. The filtration characteristics of the test suspensions were described by four different parameters and separate models were derived for each of those. All four models were tested with an independent set of samples in order to validate the created models. The results presented in this paper clearly show that the introduced procedure can be applied for creating models that enable the filtration characteristics to be reliably correlated with the particle size and shape distributions.