In this study, the simultaneous conductometric titration method for determination of mixtures of 4-nitrophenol, 2,4-dinitrophenol, and 2,4,6-trinitrophenol based on principal component artificial neural network (ANN) calibration model was proposed. The three-layered feed-forward ANN trained by back-propagation learning was used to model the complex nonlinear relationship between the concentration of 4-nitrophenol, 2,4-dinitrophenol, and 2,4,6-trinitrophenol in their ternary mixtures and the conductance of the solutions at different volumes of titrant. The principal components of the conductance matrix were used as the input of the network. The network architecture and parameters were optimized to give low prediction error. The optimized networks predicted the concentrations of nitrophenols in synthetic mixtures. The results showed that the used ANN can proceed the titration data with low relative prediction errors (5.53%, 4.03%, and 4.71% for 4-nitrophenol, 2,4-dinitrophenol, and 2,4,6-trinitrophenol, respectively) and satisfactory recoveries.
' INTRODUCTIONAmong the transduction methods developed, conductometric transducer is quite simple and is easily fabricated because it has no reference electrode. 1 Nowadays, almost in all industrial plants and clearly in all research analytical laboratories, there is a digital conductomer that is computer controlled. Conductometry is a relatively inexpensive, simple, and accurate method. Conductometric titrations have a more selective character when acidÀbase, 2 complexometric, 3,4 or precipitation 5,6 reactions are explored, thus widening the range of analytical applications.In binary or ternary mixtures of acids or bases, if the differences between acidity constants of individual acids are less than four logarithmic units, we cannot observe all of the titration end points. So, it is impossible to have an accurate determination in these types of mixtures.Introducing the multivariate statistical methods in analytical chemistry creates a suitable and easy to use device to tackle and remove such problems. That these methods used a whole data set in the course of titration (first-order method) instead of a single or scalar datum (zero-order methods such as end point in conventional titrations) gave good capability to these approaches to determine the concentrations of all constituents of a mixture.The application of multivariate calibration to potentiometric titration data was introduced by Lindberg and Kowalski 7 in 1988 for the simultaneous determination of acid mixtures using acidÀbase titration and partial least-squares (PLS) regression. After that, this PLS calibration method has been applied to complexometric titration, 8 pH metric titration, 9 potentiometric precipitation titration, 10 and conductometric titration 11 by different researchers.In these methods, the authors assumed a linear relationship between the volume of titrant added and analytes concentrations.Coelho and Gutz introduced a chemometric method based on multiparametric nonlinear regression for simultaneo...