Classification algorithms for quality control of industrially manufactured mineral fertilizers

A comparison of different classification algorithms for dynamic description of the quality of industrially produced mineral fertilizers is presented. Five brands of fertilizers (NPK 16-16-8, NP(S) 12-40 (10), NPK(S) 4-30-15 (16), NPK(S) 0-20-20 (5), MAP (NP) 12-52) were analyzed using optical and X-ray fluorescence methods of control. A method of identifying the characteristic features of industrially produced mineral fertilizers using constructed “objects-characteristics” data array is described. Estimates of the quality of determination of physicochemical parameters of fertilizers (the content of N, P, К and total content of S, types of the particle size of the press-fitted granules and the pre-drying conditions) are given. The procedure of choosing optimal parameters for each considered classification algorithm is shown. The algorithms were estimated using the F-measure (harmonic mean of “precision” and “recall” of classifier). The training of the classification algorithms and assessment of the quality of their “work” were carried out according to the cross-validation strategy on 10 data sets (folds) with a delayed test (30% of the total data volume). The total value of the quality metric was calculated as the average for all classes. The use of the principal component analysis for informative representation of the studied objects on the plane is described. Finally, a combined method of analysis for automated and informative study of industrial products was proposed.

1. Remes A., Saloheimo K., Jamsa-Jounela S. L. Effect of speed and accuracy of on-line elemental analysis on flotation control performance / Miner. Eng. 2007. Vol. 20. N 11. E 1055 -1066.

2. Yunovidov D. V, Sokolov V V, Bakhvalov A. S., Don-skikh V. A, Develop of reference sample of apatite concentrate. The effective control of homogeneity with X-ray fluorescence / GIAB. 2016. Vol. 7. E 131 - 144 [in Russian],

3. Offroy M., Duponchel L. Topological data analysis: a promising big data exploration tool in biology, analytical chemistry and physical chemistry / Anal. Chim. Acta. 2016. Vol. 910. N 3. E 1 - 11.

4. Sharma A. et al. Characterizing soils via portable X-ray fluorescence spectrometer: 3. Soil reaction (pH) / Geoderma. 2014. Vol. 232 - 234. E 141 - 147.

5. Yunovidov D. V, Sokolov V V., Bakhvalov A. S. The Use of the Sample Spectrum for Assessing the Impact of Different Stages of the NEKS Fertilizer Ereparation on the Results of X-Ray Fluorescence Analysis / Zavod. Lab. Diagn. Mater. 2017. Vol. 83. N 9. E 15 - 21 [in Russian],

6. Angeyo К, H. et al. Feasibility for direct rapid energy dispersive X-ray fluorescence (EDXRF) and scattering analysis of complex matrix liquids by partial least squares / Appl. Radiat. Isot. 2012. Vol. 70. N 11. Ё 2596 - 2601.

7. Bueno Guerra M. B. et al. Comparison of analytical performance of benchtop and handheld energy dispersive X-ray fluorescence systems for the direct analysis of plant materials / J. Anal. At. Spectrom. 2014. Vol. 29. N 9. E 1667 - 1674.

8. Moran K. Big Data in Chemical Toxicity Research: The Use of High-Throughput Screening Assays To Identify Fotential Toxicants / Chem. Res. Toxical. 2014. Vol. 27. N 10. E 1643 - 1651.

9. Kaniu M. I. et al. Energy dispersive X-ray fluorescence and scattering assessment of soil quality via partial least squares and artificial neural networks analytical modeling approaches / Talanta. 2012. Vol. 98. E 236 - 240.

10. Kunz M. R., Kalivas J. H., Andries E. Model updating for spectral calibration maintenance and transfer using 1-norm variants of tikhonov regularization / Anal. Chem. 2010. Vol. 82. N 9. E 3642 - 3649.

11. https://github.com/DimYun/DSpectra (accessed June 13, 2018).