Study of NIR Spectra, Particle Size Distributions and Chemical Parameters of Wheat Flours: A Multi-Way Approach

Near infrared (NIR) reflectance spectra contain information about both physical and chemical characteristics of flour samples and have great potential for on-line/at-line quality control in a flour mill. The addition of physical characteristics such as particle size distribution data to the NIR spectra and chemical composition data of wheat flour samples was anticipated to provide a better understanding and translation of multivariate measurements into the operational routines and experiences of mill operators. This was studied using a multi-way model called “Analysis of Common Dimensions and Specific Weights” (COMDIM). By this method the underlying dimensions across several data tables with different numbers of variables are defined and the scores and loadings are interpretable in the same way as in a classical Principal Component Analysis. The method was applied on raw NIR spectra as well as after correcting the NIR spectra using the Standard Normal Variate (SNV). The model output in terms of weights, scores and loadings were highly interpretable and in agreement with common characteristics of wheat flour samples. Four underlying dimensions explained 99.4% of the total variation, both when analysing raw and SNV-corrected spectra. A comparison of the two analyses clearly shows that correcting the spectra puts more emphasis on the chemical information in the spectra. However, even corrected NIR spectra contain considerable information about the particle size properties of the flour samples. It is suggested that the COMDIM model can be a useful tool in the process control in a flour mill and it can be used on a wide range of multi-way data problems to assure a high degree of interpretability.