Discriminant Analysis Applied to Moisture Determination in Raw Materials for Animal Feed by Pulsed NMR

Abstract

Animal feeds are complex mixtures of several different raw materials. Unfortunately, a good determination of moisture of these raw materials by low resolution pulsed nuclear magnetic resonance (NMR) requires the use of a specific prediction equation for each type of raw material. In this way, the moisture contents of nine raw materials, corn, wheat, cassava, peas, soft wheat brans, alfalfa, meals of sunflower seed, rapeseed, and soybean were determined in the range of 7-15% with standard errors of 0.18-0.47%. Standard errors of moisture prediction fluctuated between 0.4-1.75% when moisture content was determined by a single equation common to all types. Classificatory discriminant analysis was used to automatically identify the type of each sample on the basis of the NMR signals, without human intervention. Discrimination performances of subsets from two to five parameters derived from NMR signals of the tranverse relaxation (free induction decay and CPMG spin-echo sequences) and of the longitudinal relaxation (saturation-recovery sequence) were evaluated. By using a subset of five NMR parameters, 93.5% of the samples in an independent set of unknown materials were correctly classified and their moisture contents were predicted with standard errors of 0.18-0.41%.