Parallel factor analysis (PARAFAC) is applied to three calibrations of a field‐portable, cuvette‐based, single‐measurement, excitation–emission matrix fluorometer. In the first example the fluorometer is calibrated based on interactions between a non‐fluorescent DDT‐type pesticide and a fluorescent dye. PARAFAC is employed to deconvolve the fluorescence profiles of dissociated and complexed dye states. Calibration is performed based on the intensity of dye–pesticide fluorescence. In the second example, weighted PARAFAC (W‐PARAFAC) is applied to determination of three polynuclear aromatic hydrocarbons (PAHs). The weighted algorithm is required to incorporate saturated channels of the CCD detector into the calibration model. In the third example, W‐PARAFAC is applied to calibration of two carbamate pesticides. The weighted algorithm is required to account for Rayleigh and Raman scattering overlapping with the fluorescence spectra. For theses three applications, parts‐per‐trillion to parts‐per‐billion detection limits are observed in aqueous solutions. Copyright © 2000 John Wiley & Sons, Ltd.

No abstract

No abstract

Parallel factor analysis (PARAFAC) is applied to three calibrations of a field-portable, cuvette-based, singlemeasurement, excitation-emission matrix fluorometer. In the first example the fluorometer is calibrated based on interactions between a non-fluorescent DDT-type pesticide and a fluorescent dye. PARAFAC is employed to deconvolve the fluorescence profiles of dissociated and complexed dye states. Calibration is performed based on the intensity of dye-pesticide fluorescence. In the second example, weighted PARAFAC (W-PARAFAC) is applied to determination of three polynuclear aromatic hydrocarbons (PAHs). The weighted algorithm is required to incorporate saturated channels of the CCD detector into the calibration model. In the third example, W-PARAFAC is applied to calibration of two carbamate pesticides. The weighted algorithm is required to account for Rayleigh and Raman scattering overlapping with the fluorescence spectra. For theses three applications, parts-per-trillion to parts-per-billion detection limits are observed in aqueous solutions. Copyright

SummaryCritical parameters for the separation of enkephalin related peptides by micellar electrokinetic chromatography were identified by using statistical experimental design. Nine experimental variables: micelle concentration, pH of the background electrolyte, addition of organic modifier to the background electrolyte, injection length, temperature of the capillary, applied voltage, ionic strength of the buffer, the composition of the injected solution and stacking were investigated in a twolevel fractional factorial design. The pH was found to be of fundamental importance but a great variation in separation performance was observed, subsequent modeling revealed two subgroups with respect to pH. For further screening of significant factors, the pH was kept constant and by using D-optimal design it was possible to study six interaction terms together with eight main effects in only 20 experiments. The effect of the interaction terms were often found to be as high as the main effects, thus exemplifying the complexity of the system. The effect of sodium dodecyl sulfate (SDS) concentration on efficiency and resolution was highly dependent on the amount of organic modifier, temperature and ionic strength. The migration factor calculated from obtained migration times was found to be a non-linear function of the factors involved. A complete separation within reasonable analysis time was obtained at the centre point.

In this work a methodology is presented for the transformation of non-linear response data via a neural network and subsequent standard linear PLS regression. The superb transparency of linear PLS is retained with respect to the diagnostic capabilities via residual analysis and leverage, thus making this method an excellent candidate for process modelling and control.The approach developed performs an initial linear PLS to elucidate the relationship between predicted and observed values, to determine the initial parameters for the neural network and to determine the optimai number of PLS components. The parameters of the neurai network are optimized via a modified simplex optimization, with a linear PLS regression at the predetemined number of components beiig the objective function, minimizing the mean squared error of cross-vaiidation. The optimai neurai network was defined as the one giving the lowest mean squared error of cross-vaiidation.The applicability of this approach was demonstrated using three real-life industrial data sets, which gave reductions in the estimates of mean squared error in the range of 64%-98% of the original error. O 1996 by John Wiley & Sons, Ltd.

No abstract

No abstract