NIRS calibrations to estimate quality of argentine soybean extrusion-press expeller

There are in Argentina more than 300 small-scale plants which process soybeans using extrusion-press methods. The product obtained after oil extraction is a protein concentrate called expeller, which is lower in protein and higher in residual oil than soybean meal obtained by using solvents. Quality control in these plants is not a common practice, and laboratory results are usually available days after the product has been sold. Due to its speed, low operation costs and lack of laboratory personnel needs, NIRS technology could be the best alternative to carry on quality control programs in extrusion plants.

The objective of this work was to evaluate the performance of NIRS technology to determine quality parameters of argentine soybean expellers.

Between 2012 and 2014 a total of 650 samples were collected from 62 plants located in five provinces. At each plant, samples were taken in ten different days which, along with the wide geographical distribution, provided high representativity and variability. Samples were milled at 1 mm, and NIRS spectra were taken using a FOSS NIRSystems 6500 monochromator, reflectance mode, 400 to 2500 nm range, using rectangular 27 cm2-area cells. Each sample demanded about 5 min to be read in duplicates. Using WINISI III software (PCA and Select, preliminary calibrations), representative samples were chosen to be part of the calibration set. The following parameters were analyzed using conventional laboratory techniques: %moisture (MOIST, oven 105 °C), %crude protein (CP, Kjeldahl), %oil (OIL, Soxhlet), %ash (ASH, Calcination at 900 °C), %CP Solubility in OHK (CPsol), and ureasic activity (UA, pH difference). Calibrations were developed using MPLS, after mathematical treatments were applied to the spectra (derivatives, smooth, scatter correction). Spectral and compositional outliers were discarded. Data was converted from as-is to dry-matter basis using the simultaneous prediction of %MOIST of each sample.

Number of samples, mean, standard deviation, minimum and maximum values of the calibration set, standard errors of calibration (SEC) and cross validation (SECV), coefficient of determination (R2), RPD (SD/SEC) and Figure of Merit (FM, range/SEC) were for MOIST: 294, 6.95, 1.95, 3.97, 13.03, 0.29, 0.31, 0.97, 6.72 and 31.2, for CP: 204, 40,3, 2.50, 32.5, 46.8, 0.48, 0.51, 0.96, 5.2 and 29.8; for OIL: 194, 8.25,m 2.57, 3.43, 15.33, 0.23, 0.31, 0.94, 11.17 and 51.7; for ASH: 251, 5.76, 0.31, 5.03, 6.58, 0.175, 0.203, 0.68, 1.77 and 8.85, and for CPsol: 199, 75.33, 7.58, 54.03, 89.43, 2.41, 2.79, 0.90, 3.14 and 14.7, respectively. No satisfactory calibration was obtained for UA.

We conclude that NIRS technology enables to estimate in a rapid, precise and cheap way the main quality parameters of soybean extrusion-press expellers produced in Argentina. NIRS would be the best technology to carry on quality control programs in small-scale plants.