H.J. Luinge

Determination of the fat, protein and lactose content of milk using Fourier transform infrared spectrometry

Abstract In this paper the quantitative in-line determination of the composition of milk using Fourier transform infrared spectrometry is described. The fat, protein and lactose content are predicted of which the former two are compared with Rose-Gottlieb and Kjeldahl reference values respectively. Fat, protein and lactose are also determined using a MultiSpec infrared filter instrument. Calibration techniques such as classical and inverse least squares regression, principal component regression and partial least squares regression are applied and the results are compared. All methods appear to perform comparably with respect to prediction error and are equivalent to the conventional filter based method.

Automated interpretation of vibrational spectra

Abstract Automated interpretation of spectra has been a subject of research for many years. In this review attention is focused on computerized systems capable of deducing structural information from vibrational spectra. Different techniques used are compared and some future trends are addressed.

Partial least squares regression as a multivariate tool for the interpretation of infrared spectra

Abstract The potentials of partial least squares regression (PLS) as a tool for the automated interpretation of infrared spectra have been investigated. Class probabilities are obtained by applying a sigmoid function to the PLS predictions. Results are compared with those obtained from artificial neural networks (ANN) applied to full spectra and to the most relevant principal component scores. The predictions obtained from all approaches appear to be comparable, but the time required for training is considerably shorter for the extended PLS approach.

Recognition of visual characteristics of infrared spectra by artificial neural networks and partial least squares regression

Abstract The potentials of artificial neural networks and partial least squares regression for computerized interpretation of infrared spectra are studied. Experiments are carried out to establish the capabilities of these methods to recognize characteristic band shapes and spectral patterns, commonly used by experienced spectroscopists for interpretation. Classification is performed on (i) organic, inorganic and polyaromatic compounds using the entire spectral profiles, (ii) organophosphorus and non-organophosphorus compounds using specific absorption patterns of the OP and PS bands, and (iii) alcohols, carbamates and terminal alkynes using the shape of the individual OH, NH and CH bands. Results are compared with the information obtained from classification using frequency/intensity-structure correlation tables, and with interpretation as performed by experts. Classification by skilled interpreters is found to be superior in all cases. The multivariate methods give a significant improvement of the results compared to the predictions obtained from frequency/intensity data. Differences between artificial neural networks and partial least squares regression are small when full spectra or spectral regions are considered. Networks score better in recognising individual bands. The band width and the absorption frequency play an important role in the recognition process. The results prove to be practically insensitive to reduction of the number of spectral data points by a factor 16.

Trace-level identity confirmation from infrared spectra by library searching and artificial neural networks

Abstract Confirmation of the identity of analytes at trace-level concentrations from infrared spectra is described. Both library search and artificial neural networks (ANNs) are used for this purpose. The effect of pre-processing, library composition and search algorithm on the results are investigated. Best separation of positives and negatives is obtained with ANN. Library search methods, however, are superior when prevention of false positives is essential. The correlation coefficient is scale-invariant and gives excellent results.

Artificial intelligence for the interpretation of combined spectral data Design and development of a spectrum interpreter

Abstract The design of a knowledge-based system for the interpretation of combined spectral data for structure elucidation (EXSPEC) is described. Some basic design features are discussed and the functioning of the knowledge base, inference mechanism and user-interface is outlined. Attention is focussed on the development of a spectrum interpreter for infrared and mass spectral data. Interpretation of spectra for 120 liquid alcohols used for rule generation was successful. The system can be run on a Macintosh II or, more slowly, on a Macintosh Plus.

Influence and correction of peak shift and band broadening observed by rank analysis on vibrational bands from variable-temperature measurements

Abstract Variable-temperature infrared (IR) spectra of cyclohexane and IR and Raman spectra of chlorocyclohexane have been investigated by graphic eigenvalue analysis. Thermal effects known as peak shift and band broadening combined with heteroscedastic noise in vibrational bands are found to have severe influence on the interpretation of the outcome of rank analysis. Methods for correction of frequency shifts and band broadening in the spectral profiles due to temperature variation are developed and tested.

Artificial intelligence used for the interpretation of combined spectral data *1 : Part II. PEGASUS: a PROLOG program for the generation of acyclic molecular structures

A computer program, PEGASUS (PROLOG-based EXSPEC Generator for Acyclic StrUctureS), has been developed which can be used to generate exhaustively and non-redundantly all possible acyclic isomers that satisfy a given molecular weight or formula PEGASUS was written in PROLOG and implemented on an inexpensive personal computer (Apple Macintosh Plus). The program is described and the scope for its application is surveyed.

Polarised FT-IR and Raman spectra of β-d-fructopyranose single crystals

Abstract Polarised FT-IR and Raman spectra have been recorded for β- d -fructopyranose single-crystal samples at room temperature. The assignment of absorption bands to the stretching (νOH) and out-of-plane bending (γOH) vibrations of the hydrogen-bonded OH groups is proposed on the basis of the ‘oriented gas’ model approximation. For the very weak H-bond the transition dipole moment is located in the direction of the OH bond whereas it lies in the O⋯O direction for the stronger H-bond, O(2)H⋯O(1). Different couplings between the νOH vibrations are discussed and proposed based on polarised spectra. Assignment is made of the stretching vibrations of the CH2 and CH groups. Other vibrations appear to be mixed and very complex. Only a tentative assignment for the CO stretching vibrations is proposed.

Analysis of Carbon Black-Filled Rubber Materials by External Reflection FT-IR Spectrometry:

External reflection Fourier transform infrared (FT-IR) spectrometry in combination with partial least-squares regression allows qualitative and quantitative analysis of rubber materials with a high carbon black content (up to 35% w/w) on a routine level. The method developed proves to be reproducible and yields reliable qualitative information. The prediction potential of the method is shown to be sufficient for (semi)-quantitative analysis. Proper selection of spectral windows greatly improves the results. It is shown that natural, butadiene, and styrenebutadiene rubber can be determined with a precision of 9, 8, and 6% w/w, respectively.