Measurement of IWTO-19 Ash Content by Near Infrared Reflectance (NIR) Analysis

Abstract

The prediction of ash content of laboratory-scoured core samples utilising Near Infrared Reflectance Analysis (NIRA) has been investigated. Modified Partial Least Squares (MPLS) Regression was found to underestimate ash content when the sample being tested contained significant quantities of dag. The underestimation was not a consequence of saturation of the NIRA detector but rather appeared to be due to an inability of the MPLS technique to adequately account for dag which was present in the sample but masked by wool.

Application of Artificial Neural Networks (ANN) Regression to the calibration data set produced improved results. The underestimation at higher ash levels was not as evident, indicating that ANN is better able to utilise the spectral information to predict total ash content.

High levels of dag were found to adversely affect the repeatability of the IWTO-19 method for determining ash content. Uneven distribution of dag within samples was believed to be responsible. This finding has implications for NIRA, as any method of prediction can only be as good as the reference method to which it is calibrated. 

Citation

"Measurement of IWTO-19 Ash Content by Near Infrared Reflectance (NIR) Analysis", D. J. Petrie, J. J. Lidgard, J. W. Marler and A. H. M. Ireland, Report 02, Raw Wool Group, IWTO Buenos Aries Meeting, May 2003