Application of the system modified by piezosensors and artificial neural networks for quantitative analysis of a ternary mixture of diethyl ether-methyl acetate-ethyl acetate

Abstract

The aim was to develop a method for quantitative determination of diethyl ether, methyl a cetate and ethyl a cetate at their joint presence. Determination of the individual substances and mixtures carried out in static conditions in a cell with 9 modified piezoelectric sensors. Sorption was evaluated by the maximum change in the oscillation frequency of the sensor. Given the greatest sensitivity to certain substances studied modifiers 12: tris -β -cyanoethoxy propane (Tris-β-CEP), sorbitan monopalmitate ( Twin-40), dinonyl phthalate (DNF), pentaerythritol tetra benzoate (TBPE), di-2-ethylhexyl sebacate (DEGSb), triphenyl (TFF), polyethylene glycol-300 (PEG-300), polyethylene glycol-2000 (PEG-2000), polyethylene glycol succianate (PEGS), polyethylene glycol sebacate (PEGSb), polyethylene glycol phthalate (PEGF), and stearic acid. Sorption isotherms alkyl acetates and ether have a linear form that allows them to define the entive range of concentrations studied. To identify the components of a mixture of its constituent substances should form a statistically distinquishable «visual impressions» at a given confidence level. For each system of ethyl acetate-diethyl ether, methyl acetate, ethyl acetate-ether dizhtilovy calculated deviation criteria (Studen's statistic), we compare them with the value of the maximum deviation criterion (rmax(0.95; 4)=1.69). If the calculated value is greater than analytical table sensor signals are statistically in distinquishable. To reduce the number of inputs to 9 ( the number in the matrix pezosensoror) neural network has been established. Determine the significance of the input signals can be reduced the least (TBPE, DEHSb, PEG 2000). To study the three-component gas system created the model mix. Analytical signals obtained after passingthrough the cell detection model ternary mixtures, treated by artificial neural network using the standard program NeuroPro 0.25. as an activation function used logistic function ( sigmoid) method of optimization is the method of conjugate gradients. To verify that the network the task tested the neural network method was introduced-found with the use of a set of compounds not included in the sample obuchabschuyu. Test results confirm the accuracy of esters in the mixture. The error calculation does not exceed 12%.