ARTIFICIAL NEURAL NETWORK MODELING FOR DENSITY OF SOME BINARY SYSTEMS
LISA GABRIELA 1, CURTEANU SILVIA 2 TECHNICAL UNIVERSITY OF IAĆI, FACULTY OF CHEMICAL ENGINEERING (1)DEPARTMENT OF PHYSICAL CHEMISTRY, (2)DEPARTMENT OF CHEMICAL ENGINEERING BD. D. MANGERON, 71A, RO-700050 IAĆI, ROMANIA E-MAIL: GAPREOT@CH.TUIAĆI.RO
This paper presents experimental results of density for three binary systems: toluene - n-propanol, toluene - iso-propanol and toluene - propanoic acid. In order to obtain density values at different temperature and concentrations, empirical models were developed using experimental data. Two types of models were built and compared: feedforward neural networks and empirical equations which give the dependence of density on temperature and concentration. Accurate results were obtained in training and validation phases, using neural networks with simple topologies and short training time. In addition, the trend of the predicted densities was qualitatively consistent. The empirical equations also provide good concordance between simulation results and experimental data.