It could be challenging to consider the exact mathematical derivation for the input-output relationships because of the complexity in the design procedure. Thus, the dynamic performance modeling of a filter is an essential consideration for the filter design process. In the present paper, the ability and accuracy of a neural network system (ANNs) have been investigated for dynamic modeling of filtration. For the implementation of the current technique, the MATLAB codes and related instructions are efficiently used. The main objective of this research is to predict filter performance as a function of the pressure loss and particle numbers at the inlet and outlet of the cellulose filter. To show the best fitting algorithm, an extensive comparison test was applied on the ANNs with RBF (radial basis function) and FFN (feed forwards network). Resulting from the comprehensive evaluation test, the ANN procedure yields very accurate results. The results show that there is an excellent agreement between the testing data (not used in training) and estimated data, with average errors minimal. The optimum values for an architecture of ANN obtained three, two, one, and 2/3E-13 for hidden layers, inputs neurons, output neurons, and MSE error, respectively. Also, ANN in RBF type and the trained FFN method is able to accurately capture the efficiency of the non-linear dynamic of the filter even for a new condition that has not been used in the training process (testing data). For the sake of comparison, the results of the proposed RBF model are compared with those of the FFN model.

Keywords