Processing of X-Ray fluorescence data with neural networks for express assessment of iron ore concentrate quality

DOI: https://doi.org/10.30686/1609-9192-2026-3-217-225

Читать на русскоя языке Yu.S. Buzykova
Financial University under the Government of the Russian Federation, Moscow, Russian Federation
 Russian Mining Industry №3/ 2026 p. 217-225

Abstract: The iron mass fraction together with silica, alumina, phosphorus and sulphur contents govern both the metallurgical value and the market price of iron ore concentrates, while the requirements of the blast-furnace-grade feed with Fe > 69.5% and SiO2 < 3% tightens the tolerances on analytical accuracy and response time. The laboratory X-ray fluorescence (XRF) analysis based on fundamental parameters and certified reference material calibration delivers adequate precision, however its 15 to 40 minute turnaround time per sample can no longer keep up with modern processing plants with the output from 3.5 to 4.8 thousand tonnes of concentrate per hour. This study addresses neural network processing of raw XRF spectra for express quality prediction at the frequency compatible with the flotation and magnetic separation control loops. A set of 2140 paired observations matching the raw XRF spectra with the certified wet chemical testing was collected between September 2024 and October 2025 at three processing plants at the Kursk Magnetic Anomaly using mineralogically matching samples. Five models are benchmarked: a baseline PLS regression, gradient boosting, a one-dimensional convolutional neural network, a 1D residual network and a hybrid architecture with channel-wise attention over the spectrum. The hybrid model is the most accurate in terms of the Fe mass fraction, reaching RMSE = 0.17%, MAE = 0.12% and R2 = 0.987; SiO2 yields RMSE = 0.24%; inference time is 14 ms per spectrum using GPU and 92 ms using CPU, two orders of magnitude faster than the laboratory workflow. A calculation of the annual benefit for a plant producing 15 million tonnes of concentrate gives the operating cost savings of 38 to 52 million roubles and an additional revenue contribution of 310 to 470 million roubles through tighter Fe stabilization within ±0.2%, in 2025 prices. The results confirm that express quality assessment using neural network is a viable element in the digital control envelope of a modern iron ore processing plant.

Keywords: X-ray fluorescence, neural network, iron ore concentrate, express quality assessment, convolutional networks, beneficiation digitalization, attention mechanism, processing plant

For citation: Buzykova Yu.S. Processing of X-Ray fluorescence data with neural networks for express assessment of iron ore concentrate quality. Russian Mining Industry. 2026;(3):217–225. https://doi.org/10.30686/1609-9192-2026-3-217-225

Article info

Received: 29.03.2026

Revised: 20.04.2026

Accepted: 24.04.2026

Information about the author

Yulia S. Buzykova – Cand. Sci. (Educ.), Associate Professor, Department of Information Technology, Financial University under the Government of the Russian Federation, Moscow, Russian Federation; e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

References

1. Najafabadipour A., Hassanzadeh F., Kordestani M. Advanced deep learning models for predicting elemental concentrations in iron ore mine using XRF data: a cost-effective alternative to ICP-MS methods. Environmental Geochemistry and Health. 2025;47(4):104. https://doi.org/10.1007/s10653-025-02419-7

2. Xiao Z., Li F., Lyu S., Wang Q. A deep learning approach for improving ore classification accuracy by double-stream heterogeneous data fusion of XRF. Microchemical Journal. 2026;220:116506. https://doi.org/10.1016/j.microc.2025.116506

3. Andric V., Kvascev G., Cvetanovic M., Stojanovic S., Bacanin N., Gajic-Kvascev M. Deep learning assisted XRF spectra classification. Scientific Reports. 2024;14:3666. https://doi.org/10.1038/s41598-024-53988-z

4. Xiao D., Le B.T., Ha T.T.L. Iron ore identification method using reflectance spectrometer and a deep neural network framework. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy. 2021;248;119168. https://doi.org/10.1016/j.saa.2020.119168

5. Chen T., Sun L., Yu H., Zeng P., Qi L. Online Fe grade monitoring of iron ore slurry by Morse wavelet transform and lightweight convolutional neural network based on LIBS. Spectrochimica Acta Part B: Atomic Spectroscopy. 2023;210:106821. https://doi.org/10.1016/j.sab.2023.106821

6. Zhou S., Yuan Z., Cheng Q., Weindorf D.C., Zhang Z., Yang J. et al. Quantitative analysis of iron and silicon concentrations in iron ore concentrate using portable X-ray Fluorescence (XRF). Applied Spectroscopy. 2020;74(1):55–62. https://doi.org/10.1177/0003702819871627

7. Silva A.C.P., Coimbra K.T.Z., Filho L.W.R., Pessin G., Correa-Pabón R.E. Monitoring of iron ore quality through ultra-spectral data and machine learning methods. AI. 2022;3(2):554–570. https://doi.org/10.3390/ai3020032

8. Patel A.K., Chatterjee S., Gorai A.K. Development of a machine vision system using the support vector machine regression (SVR) algorithm for the online prediction of iron ore grades. Earth Science Informatics. 2019;12(2):197–210. https://doi.org/10.1007/s12145-018-0370-6

9. Tskhadadze N.V., Kuchkovskaya N.V., Bondarenko M.P., Frolova V.B., Lazarev M.P. Designing a smart life cycle management system for a mining enterprise based on Industry 4.0 and circular economy technologies. Russian Mining Industry. 2024;(5S):12–20. (In Russ.) https://doi.org/10.30686/1609-9192-2024-5S-12-20

10. Lebedok A.V., Markworth L. Contemporary iron ore processing: challenges and solutions from ALLMINERAL. Russian Mining Industry. 2022;(3):84–88. https://doi.org/10.30686/1609-9192-2022-3-84-88

11. Chen Y., Liu Z., Zhao X., Sun S., Li X., Xu C. Soil heavy metal content prediction based on a deep belief network and random forest model. Applied Spectroscopy. 2022;76(9):1068–1079. https://doi.org/10.1177/00037028221104823

12. Viana A.Z., Månbro C., Jooshaki M., Parian M. Automated ore texture classification using μ-XRF imaging and unsupervised machine learning: correlation with surface hardness. Preprint. March 15, 2025. https://doi.org/10.2139/ssrn.5180401

13. Xiao D., Vu Q.H., Le B.T., Ha T.T.L. A method for mapping and monitoring of iron ore stopes based on hyperspectral remote sensing-ground data and a 3D deep neural network. Neural Computing and Applications. 2023;35(16):12221–12232. https://doi.org/10.1007/s00521-023-08353-y

14. Progorowicz J., Skoczylas A., Anufriiev S., Dudzik M., Stefaniak P. Estimation of final product concentration in metalic ores using convolutional neural networks. Minerals. 2022;12(12):1480. https://doi.org/10.3390/min12121480

15. Lobo A, Garcia E, Barroso G, Martí D, Fernandez-Turiel J-L, Ibáñez-Insa J. Machine learning for mineral identification and ore estimation from hyperspectral imagery in tin–tungsten deposits: simulation under indoor conditions. Remote Sensing. 2021;13(16):3258. https://doi.org/10.3390/rs13163258

16. Shkodinsky S.V., Krupnov Yu.A., Romantsova T.V. Multi-factor risk analysis of modern fintech based on multimodal analytics. Finance: Theory and Practice. 2025;29(4):112–128. https://doi.org/10.26794/2587-5671-2025-29-4-112-128

17. Borisova O.V., Dreving S.R., Loseva O.V., Fedotova M.A. State financial support measures and risk factors affecting the cost of investment projects for the introduction of industrial robotic complex. Finance: Theory and Practice. 2025;29(3):20–34. https://doi.org/10.26794/2587-5671-2025-29-3-20-34

18. Mikhaylov A.Yu., Yousif N.B., An J. How high efficiency of the Chilean stock market does impact on energy transition? Research using Deep seek AI optimization. Finance: Theory and Practice. 2025;29(1):181–194. https://doi.org/10.26794/2587-5671-2025-29-1-181-194

19. Fedorova E.A., Nevredinov A.R., Melikhov K.S., Yashchenko A.I. Import substitution impact on growth of production of mineral products and metallurgy: short-term and long-term forecasting of basic sectors of the national economy. Finance: Theory and Practice. 2023;27(6):17-30. https://doi.org/10.26794/2587-5671-2023-27-6-17-30

20. Dombrovskaya A.Yu., Brodovskaya E.V. Emigration narratives relevance assessment by large language models for social media monitoring. Vestnik Moskovskogo Universiteta. Seriya 10. Zhurnalistika. 2025;(5):209–232. (In Russ.) https://doi.org/10.55959/msu.vestnik.journ.5.2025.209232

21. Valkova J.E. Using artificial intelligence technologies to prepare and compose scientific articles. Informatics and Education. 2024;39(6):38–52. (In Russ.) https://doi.org/10.32517/0234-0453-2024-39-6-38-52

22. Shenderov V.I. The use of big data and neural networks for monitoring and forecasting environmental changes in Russia. Voprosy Ecologii. 2024;37(1):98–127. (In Russ.) https://doi.org/10.25726/u2912-9651-3244-z