Abstract: A simulation procedure for the optimization of enzymatic reactors used in sandwich flow injection systems is evaluated. The system is modeled as a plug-flow reactor with axial dispersion. To calibrate it, dispersion coefficients can be evaluated using residence time distribution techniques; meanwhile, enzymatic kinetics must be determined for the system considered, according to the values of the substrate conversion attained. The model has been linked to an optimization routine based on the Powell algorithm. The proposed approach has been evaluated in a system performing simultaneous determinations of glucose and glycerol, considered the common carbon sources in a fermentation process.

Abstract: The present study shows and gives evidence of the applicability of natural computation techniques in the modeling and optimization of a sequential injection flow system of anal. for colorimetric iron(III) determination in water samples. The reaction with thiocyanate is used as reagent color. A neural network consisting of two hidden layers, each one formed by eight neurons, was used to model the system. Optimization of the system in terms of sensitivity, linearity and sampling rate was carried out by using jointly the neural network and genetic algorithms. The latter were used with a set of 50 crossed and mutated chromosomes over 100 generations. In the system thus developed, 140 µL of sample and 70 µL of reagent were sequentially introduced into the holding coil and propelled toward the detector at a flow of 5 mL/min. The system gave a sampling rate of 110 samples per h. A comparison of the results obtained in the anal. of six samples with those obtained using the reference method (atomic absorption spectrophotometry) showed the high quality of results provided.