Comparison of cluster analysis methods for ore grade recognition on a conveyor before grinding

Introduction. The article is devoted to modern aspects of controlling ore crushing processes, including methods and means for controlling its chemical composition. The main attention is paid to problems related to unstable physical and mechanical properties of processed materials. Materials and methods. To recognize ore grades, it was proposed k-means methods, fuzzy clustering, Kohonen’s artificial neural network. It consisted in dividing the sample into groups that were homogeneous according to a given indicator and corresponding to a certain variety.To test the performance of the clustering algorithm, samples of experimental data on the total iron content, hard-to-grind varieties in the ore, and its average size were used. To demonstrate an example of ore grade recognition used for this, the Matlab. Testing of the algorithms took place in the “sliding” window mode. Results. The simulation results showed that the clustering process is performed most efficiently using the k-means method, since the silhouette coefficient and the Kalinski–Kharabas index have high values, while the Davis–Buldin index have low values. The worst result was obtained when modeling the process using the fuzzy clustering method. Discussion. For ore grade recognition, the k-means method ranks first in terms of quality, Kohonen’s neural network ranks second, and the fuzzy clustering method ranks last. Conclusion. The correct choice of the recognition algorithm will make it possible to implement competent control of the grinding process and maintain its operating modes at a given level. Resume. The proposed k-means algorithm will make it possible to recognize the ore grade, depending on which the mill control modes can be changed, maintaining the grinding parameters. Suggestions for practical applications and directions for future research. The directions of further research are the development of a cluster analysis algorithm with the ability to recognize ore grades in real time by using special measuring instruments.