A new graphically oriented local modeling procedure called interval partial least-squares ( iPLS) is presented for use on spectral data. The iPLS method is compared to full-spectrum partial least-squares and the variable selection methods principal variables (PV), forward stepwise selection (FSS), and recursively weighted regression (RWR). The methods are tested on a near-infrared (NIR) spectral data set recorded on 60 beer samples correlated to original extract concentration. The error of the full-spectrum correlation model between NIR and original extract concentration was reduced by a factor of 4 with the use of iPLS ( r = 0.998, and root mean square error of prediction equal to 0.17% plato), and the graphic output contributed to the interpretation of the chemical system under observation. The other methods tested gave a comparable reduction in the prediction error but suffered from the interpretation advantage of the graphic interface. The intervals chosen by iPLS cover both the variables found by FSS and all possible combinations as well as the variables found by PV and RWR, and iPLS is still able to utilize the first-order advantage.

The extended multiplicative signal correction (EMSC) preprocessing method allows a separation of physical light-scattering effects from chemical (vibrational) light absorbance effects in spectra from, for example, powders or turbid solutions. It is here applied to diffuse near infrared transmission (NIT) spectra of mixtures of wheat gluten (protein) and starch (carbohydrate) powders, linearized by conventional log(1/T). Without any correction for uncontrolled light scattering variation between the powder samples, these absorbance spectra could give reasonable predictions of the analyte (gluten), but only when using multivariate calibration with a much more complex model than expected. Standard MSC preprocessing did not work for these data at all; it removed too much analyte information. However, the EMSC preprocessing yielded powder spectra that obeyed Beer's Law more or less as if they had been obtained from transparent liquid solutions, apparently by isolating the chemical light absorption from additive, multiplicative, and wavelength-dependent effects of uncontrolled light-scattering variations. The model-based EMSC and its converse, the extended inverted signal correction (EISC), gave rather complete descriptions of the diffuse absorbance spectra and virtually indistinguishable performance in the calibration set and the test set of samples.

A new extended method for separating, e.g., scattering from absorbance in spectroscopic measurements, extended inverted signal correction (EISC), is presented and compared to multiplicative signal correction (MSC) and existing modifications of this. EISC preprocessing is applied to near-infrared transmittance (NIT) spectra of single wheat kernels with the aim of improving the multivariate calibration for protein content by partial least-squares regression (PLSR). The primary justification of the EISC method is to facilitate removal of spectral artifacts and interferences that are uncorrelated to target analyte concentration. In this study, EISC is applied in a general form, including additive terms, multiplicative terms, wavelength dependency of the light scatter coefficient, and simple polynomial terms. It is compared to conventional MSC and derivative methods for spectral preprocessing. Performance of the EISC was found to be comparable to a more complex dual-transformation model obtained by first calculating the second derivative NIT spectra followed by MSC. The calibration model based on EISC preprocessing performed better than models based on the raw data, second derivatives, MSC, and MSC followed by second derivatives.

No abstract

Inline near-infrared (NIR) spectroscopy has been used to monitor a continuous synthesis of an active pharmaceutical ingredient (API) intermediate by a Grignard alkylation reaction. The reaction between a ketone substrate and allylmagnesium chloride may form significant impurities with excess feeding of the Grignard reagent beyond the stoichiometric ratio. On the other hand, limiting this reagent would imply a loss in yield. Therefore, accurate dosing of the two reactants is essential. A feedforward−feedback control loop was conceived in order to maintain the reaction as closely as possible to the stoichiometric ratio, leading the path to full process automation. The feedback control loop relies on NIR transmission measurements performed in a flow cell where, in contrast to labor-intensive offline HPLC analytical methods, the whole reaction product can be scanned in real time without sample dilution. A robust PLS (projection to latent structures) model was developed with a satisfactory standard error of prediction, providing quantification of the ketone substrate in solutions with a high variability of the major solution componentthe alkoxide product. In addition, model performance supervision tools such as the spectral residuals or simple plots of pretreated spectra can assist in the identification of spectral outliers, which in this case could be related to Grignard reagent excess. If the sampling time of the NIR instrument is short enough, manipulating the inputs to the reactor may be used to obtain information about its dynamic behavior. This information is very useful for process control design, assessment of analytical tools and definition of sampling times. In this work, a systematic procedure for chemometric model building is followed, after which a discussion is made on some of the potential applications that can be found when exploiting the fast and rich information provided by NIR spectroscopy.