Modeling of ion exchange process using Mass Action Law and artificial neural networks

Abstract

The Mass Action Law is usually employed in modeling of ion exchange processes equilibrium. This methodology is based on the definition of the chemical equilibrium constant and considers the non ideality of solid and aqueous phases. Another alternative to chemical and phase equilibrium modeling is the use of Artificial Neural Networks. This work makes a comparison between both methodologies used on modeling of the equilibrium on ion exchange processes of the binary systems Pb2+-Na+,  Cu2+-Na+ e Na+-Pb2+, and the ternary system Cu2+-Na+- Pb2+ in the conditions of concentration corresponding to 0,005 eq/L and temperature of 303K, using the natural zeolyte clinoptilotita as an ion exchanger. The obtained data by the Mass Action Law from the binary systems were used as an input signal on the Artificial Neural Network training. The used networks had three layers (input, hidden and output layer), and as input signals there were used the concentration and the composition of the ions in solution and as output variable the composition of the ions on the ion exchanger were used. Results have shown that both methodologies were efficient on the binaries systems modeling. Both methodologies were also applied on prediction of the ternary systems behavior from binary systems data. There were made tests with Artificial Neural Networks including the ternary system data on the learning step. The obtained results from non predictive networks on the ternary system equilibrium description were better than those obtained from the Mass Action Law and from predictive networks.

Key words: Mass Action Law, artificial neural network, ion exchange.