(251f) Prediction of Compression-Permeability Characteristics of Solid-Liquid Systems Using Artificial Neural Networks

Filtration characteristics such as average specific filtration resistance and porosity of a given solid-liquid system are essential in the design of filtration processes. Usually such characteristics are determined experimentally by pressure/vacuum filtration tests or using the compression-permeability (C-P) cell by fitting power law type relationships to experimental data. However, these procedures require tests with a considerable amount of the target slurry. A statistical modeling tool called artificial neural network (ANN) is used in this study to predict the filtration characteristics without the need for these tests and using very small amount of the material to be tested. The filtration database containing various flocculated and non flocculated slurry properties is developed and used to train two ANNs. The input parameters for the first ANN were the applied pressure and the particle size distribution whereas the output parameter was the porosity (ε) of the compressed cake. In the case of the second ANN, the input parameters were the porosity and the particle size distribution. The logarithmic product of the specific filtration resistance and the particle true density of the cake (logαρ_s) was chosen as the output parameter of this network. Spherical organic polymers (polymethyl methacrylate beads, polyethylene beads) and non-spherical inorganic particles (kaolin, iron oxide, zinc oxide) etc. with median diameter of 2.5-172 micrometers were dispersed in water or methanol to obtain the C-P data. The use of porosity obtained from the sediment in gravitational and centrifugal fields as one of the input parameters to the two ANNs allowed the prediction of the characteristics with very good accuracy. The ANNs trained with these data were found to be fairly effective in predicting the filtration resistance of another slurry containing same particles used during the training stage. By using the ANN method, it is shown that the predicted values of these two parameters (ε and α ) are in good agreement with the measured values. With the help of this method, after the ANN is thoroughly trained with various slurries, the filtration property of another slurry can be predicted with much less amount of the target slurry compared with conventional filtration experiment.