IMPROVEMENTS OF THE NEAR-INFRARED DIFFUSE-REFLECTANCE TECHNIQUE (SPECTROSCOPY, CHEMOMETRICS, CORRELATION)

Near-infrared reflectance analysis (NIRA) is an analytical technique that uses the diffuse reflectance of a sample to determine spectrophotometrically chemical concentrations or physical properties such as hardness.; Because NIRA uses a series of multiple linear regressions to deduce automatically corrections for background or sample-matrix interferences, it can analyze many sample types that would be difficult by another spectroscopic technique. An automatic correction of interferences allows NIRA to operate independently of fully understood chemistry or spectroscopy.; The present work lists and defines the theoretical and practical considerations involved in developing and utilizing NIRA. The theoretical definitions are related to other well-developed spectrometric techniques such as spectral subtraction, curve fitting, and conventional ultraviolet-visible or mid-infrared absorption measurements. Among the practical considerations addressed are selecting the analyte, choosing "training" samples, and creating a calibration. Examples of the NIRA determination of the physical properties of hydrocarbon mixtures and the carbonate concentration of geological samples are presented.; A large part of NIRA methods development involves time-consuming, trial-and-error procedures. Specifically, training-sample selection and the process of creating a calibration are currently suboptimal. In the present study new methods for sample selection and calibration are presented, both of which are based on the use of linear-algebra techniques. These new methods speed and simplify the development of NIRA analytical methods.; As a spectrophotometric technique, NIRA is subject to photometric errors. These errors can be generated by the thermal noise of the detector, changes in cell pathlength, or variations in particle size of the sample. Each kind of error is addressed in the present study. Furthermore, methods which minimize the effect of these errors on the creation of a NIRA method are presented.; Understanding the operation of NIRA and interpreting its output is difficult because near-infrared spectra are essentially featureless and the analytical-wavelength selection is made by a computer algorithm. It would be much easier to understand the NIRA technique if spectra of the various sample constituents were available. To this end, a rapid, convenient method for reconstructing near infrared-component spectra is introduced. Based on a mathematical cross-correlation procedure, the technique requires only a set of near-infrared spectra of samples in which the concentration of the component of interest is known.