Applications of Near-infra-red Spectroscopy (NIRS) without calibration
[Introduction]
The paper describes three useful types of application of NIRS that can be accomplished without the need for calibration. Examples are given for all three. Excellent spectral precision is essential to successful application of NIRS. The efficiency of the application depends heavily on the quality of the spectra. This can be affected by minute differences among instruments of the same make and type. The second source of variance lies in differences among sample cells used to present the sample. The paper describes the methods of determination of these two sources of variance. Efficient blending in operations such as animal feed manufacture or flour milling is an important part of quality control. The third application is the determination of blending efficiency, using spectral data, again with no calibration.
[Materials and Methods]
The standard deviation (SD) of the spectral data from different instruments of the same type was measured by passing liquid manure through four instruments mounted in-line. The SD among sample cells involved scanning the same sample of air-dried soil, which had been ground to pass through a 2 mm screen,10 times with re-loading the sample between scans, and computing the SD of the unprocessed (“raw”) spectral data at up to 4 wavelengths. Suggested wavelengths for this process are 960, 1210, 1680, 2230 nm (10415, 8265, 5950, and 4485 cm-1). The precision is reported as the standard deviation (SD) of the spectral data recorded at up to 4 wavelengths.
For determination of blending-efficiency, the ingredients of a commercial feed mix were placed into an industrial blender. The blender was activated, and after 5 seconds 10 samples were taken. The blender was then activated for the full blending-time and a second set of 10 samples taken. Both sets of samples were scanned, and the SD of the spectra data computed for both series of samples.
[Results]
The SD of the absorbance of manure among the four instruments had an average CV of less than 1.5%. That of 20 individual sample cells averaged 0.7%, but differences of up to over 0.016 were observed in the absorbance data among the 20 cells. This was sufficient to incur small biases in prediction of loss on ignition. For the blending efficiency study, the SD of the absorbance of the 5-second-blended samples was up to 7 times higher than that of the fully blended samples. A little experience would enable operators to recognize when mixes are fully-blended. The same procedure can be used in a flour mill to determine the efficiency of blending mill-streams. Similar differences were observed in blending a set of 5 flour streams, using the same procedure.