Ore sorter as part of Mining 4.0 from mechatronics approach

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/391687571 Ore Sorter as Part of Mining 4.0 from Mechatronics Approach Conference Paper · February 2025 DOI: 10.1109/ICMRE64970.2025.10976272 CITATIONS 0 READS 10 5 authors, including: Nasia Balakera Institute of Communication and Computer Systems 4 PUBLICATIONS26 CITATIONS SEE PROFILE Fotios K. Konstantinidis Institute of Communication and Computer Systems 53 PUBLICATIONS633 CITATIONS SEE PROFILE Savvas Sifnaios Institute of Communication and Computer Systems 5 PUBLICATIONS68 CITATIONS SEE PROFILE All content following this page was uploaded by Nasia Balakera on 29 September 2025. The user has requested enhancement of the downloaded file. Ore sorter as part of Mining 4.0 from mechatronics approach Nasia Balakera* Institute of Communication and Computer Systems National Technical University of Athens Athens, Greece 0000-0001-5063-1212 nasia.balak[email protected] *Corresponding author Fotios K. Konstantinidis Institute of Communication and Computer Systems National Technical University of Athens Athens, Greece 0000-0002-1826-6582 f.kon[email protected] Savvas Sifnaios Institute of Communication and Computer Systems National Technical University of Athens Athens, Greece 0009-0007-5838-4317 savvas.sifnaio[email protected] George Tsimiklis Institute of Communication and Computer Systems National Technical University of Athens Athens, Greece 0000-0002-2431-8529 georg[email protected] Angelos Amditis Institute of Communication and Computer Systems National Technical University of Athens Athens, Greece 0000-0002-4089-1990 [email protected] Abstract—While the Industry 4.0 principles are taking over more industrial sectors, artificial intelligence and cutting-edge sensors and actuators are constantly developed. However, this technological rise causes higher requirements in raw material extractions. To achieve this demand, the mining industry is trying to adopt the guidelines of Mine 4.0 elevating their procedures of extraction, control and monitoring, while also disengage the human intervention from certain mining operations, such as the sorting procedure of extracted minerals. To accomplish this scope, this research enters the Cyber-Physical Sorting System (CPSS) in mining industry, which provides adaptable features for implementation in various value chains. In particular, the CPSS described in this paper will be applied in the mining field, enhancing the magnesite sorting process using RGB and hyperspectral imaging inline and X-Ray Fluorescence (XRF) analysis and RAMAN spectroscopy offline ensembles with deep convolutional neural networks for classification and an automated air nozzle system for the final separation. Keywords—Industry 4.0, Industry 5.0, Mine 4.0, Cyber-physical Sorting System, Critical Raw Material (CRM), Machine Vision, Hyperspectral Imaging, X-Ray Fluorescence (XRF), Deep Learning, Machine Learning, RAMAN spectroscopy I. INTRODUCTION In light of the Industry 4.0 (I4.0) guidelines' intervention in many technoeconomic domains and the introduction of Industry 5.0 (I5.0) in industrial hubs across the European Union, the digitalization and the automation of industrial processes consist a necessity [1]. This technological rise creates new needs for raw materials extraction. However, conventional mining methods frequently lack operating efficiency and environmentally friendly practices causing high levels of CO2 emissions and energy waste [2]. In advance, the increasingly need of Critical Raw Materials (CRMs) extraction obliged EU to import CRMs from non-european mining industries creating resilience and sustainability gaps [3, 4]. To tackle this issue, mines focus not only on aggregating CRMs from any extracted mineral sample, but also on enhancing ore waste management [5]. As a result, the term “Mine 4.0” is widely spread in the mining sector aiming to introduce I4.0 principles in mines’ procedures [6]. Mine 4.0 is based on four main axes: i) the virtualization of physical technologies, ii) the real-time monitoring and control of mine processes by accessing data instantly upon request, iii) utilization of cloud-based services, and iv) implementation of AI-based solutions and neural networks [7]. Focusing on Mine 4.0 principles and ore waste management, this research aims to implement state-of-the-art technologies in ore streams for enhancing the sorting process and achieving not only high purity levels in the sorted output samples, but also less raw material extraction needs. The suggested adaptable ore sorting system is able to identify and classify multiple ore samples by using a multisensing system consisting of: i) an RGB camera for space and size data of the samples on the conveyor belt, ii) a hyperspectral camera, which is providing spectral data of the samples up to 1700nm, iii) an offline XRF analysis together with iv) RAMAN spectroscopy aiming to validate the inline results and enhance the classification process by using chemical data. In parallel, an autonomous air-nozzle system is used for the final separation step of the ore samples, while the whole CPSS is orchestrated 359 2025 11th International Conference on Mechatronics and Robotics Engineering 979-8-3315-0929-3/25/$31.00 ©2025 IEEE 2025 11th International Conference on Mechatronics and Robotics Engineering (ICMRE) | 979-8-3315-0929-3/25/$31.00 ©2025 IEEE | DOI: 10.1109/ICMRE64970.2025.10976272 Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on September 29,2025 at 09:19:28 UTC from IEEE Xplore. Restrictions apply. by AI-driven algorithms [8]. The main contributions of this research are: • The quotation of a multispectral sorting system for material classification and separation in real-time. • A pipeline designed aiming to the adaptability in multiple material use-case and sensing equipment support. • An AI-driven intelligence orchastration targeting in the classification of magnesite from mixed ore stream. The paper’s structure is presented as follows. In section II, we are going to explore the related work of our technologies. The ore sorting mechanism pipeline will be examined in section III. And last but not least, we draw conclusions and present our future steps in section IV. II. RELATED WORK Mineral sorting is an essential process in the mining industry, enabling the separation of valuable minerals from waste material, which enhances processing efficiency and reduces overall costs. Advances in sorting technology have led to two main categories: traditional mechanical sorting and sensor-based sorting. Mechanical methods rely on the physical properties of minerals, such as density, size, and magnetism, to achieve separation, making them widely applicable due to their robustness and cost-effectiveness. Sensor-based sorting, on the other hand, uses advanced technologies, such as hyperspectral imaging, X-ray fluorescence (XRF), and optical imaging, to analyze the chemical and spectral properties of minerals in realtime, offering high precision in identifying ore characteristics. Each approach brings unique strengths and limitations, making the choice of sorting technology dependent on factors like ore composition, processing objectives, and the level of required accuracy [9]–[12]. This section explores recent advancements in both categories, focusing on performance metrics and practical applications in mineral processing. Traditional mechanical sorting methods include techniques such as gravity-based separation, magnetic sorting, and conveyor-based optical systems, which use the physical differences in ore and waste material to streamline processing. These methods have become staples in mineral processing, achieving high throughput at relatively low cost. For instance, conveyor-based systems that use optical or magnetic properties have been shown to reduce downstream milling costs by as much as 60%, enhancing throughput by filtering out waste at early processing stages [13], [14]. Gravity-based sorting, another widely used technique, exploits density differences and is effective in ores with significant density contrasts [15]. Meanwhile, magnetic sorting proves particularly useful for magnetically susceptible minerals, such as magnetite-rich ores [16]. Although economical and straightforward, these mechanical methods are often limited by the requirement for pronounced physical distinctions between ore and waste, reducing their applicability when such differences are minimal. In contrast, sensor-based sorting technologies provide a higher level of precision, making them suitable for ores with subtle compositional differences. This category includes X-ray fluorescence (XRF), hyperspectral imaging, and laser-induced breakdown spectroscopy (LIBS), each of which identifies minerals based on chemical or spectral characteristics. Hyperspectral imaging, for example, operates across a wide range of wavelengths (up to 2500 nm) and identifies minerals by their spectral signatures, achieving an identification accuracy of up to 90% [10], [17]. Similarly, XRF detects elements based on their fluorescent X-ray emissions, distinguishing valuable minerals from waste with an accuracy of around 91% in certain ores [18]. Sensor fusion, or the combination of these technologies, further improves sorting precision by minimizing classification errors and optimizing low-grade ore recovery, saving costs by reducing reagent use and energy consumption [19]. While highly accurate, these sensor-based methods are more costly and complex, making them most feasible for ores that lack strong physical sorting characteristics. In summary, both mechanical and sensor-based sorting technologies offer valuable benefits depending on the type of ore and desired processing outcome. Traditional mechanical sorting methods are economical and capable of handling large volumes, though they rely on clear physical differences between ore and waste. Sensor-based methods, while offering higher accuracy through chemical and spectral analysis, require a more significant upfront investment and technical expertise to maintain. Ultimately, the choice between these approaches depends on a balance of cost, ore type, and processing requirements, as each method provides unique advantages and faces specific challenges in application [9], [13], [19]–[21]. III. SELF - ADAPTABLE ORE SORTING: ARCHITECTURE As visualized in Fig.1, the architecture of the self-adaptable ore sorting system consists of three main layers. The first layer located at the bottom of the architecture is the Sensing & Actuation Layer, which contains the inline and offline sensing systems and the actuation system together. The aim of this layer is the data collection from the samples and the mechanical systems. The following layer is called Control & Intelligence Layer and is the connection layer between the bottom layer and the upper layer. In this layer, the systems of the first layer are controlled and monitored, while the aggregated data are first processed by using cutting-edge technologies and AI-based algorithms, and then broadcasted to the upper layer. Finally, the Fig. 1. Self-adaptable ore sorting: Three-layer Architecture 360 Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on September 29,2025 at 09:19:28 UTC from IEEE Xplore. Restrictions apply. third layer is the Management & Integration Layer, where in order to maximize the system’s efficiency, it serves as a decision-making agent that connects the outcomes of the Intelligence Layer and coordinates the hardware equipment. To this end, the sorting system’s results are able to be cloudbroadcasted and shared with the stakeholders and waste management services. To sum up, the aforementioned layers provide a self-adaptable sorting system with the ability to adjust to different sorting applications, as well as different input materials. IV. SELF - ADAPTABLE ORE SORTING: COMPONENTS An adaptable and autonomous ore sorting mechanism for magnesite separation from mixed ore streams is being developed aiming to enhance the classification process of ore samples and reach high accuracy purity levels of magnesite in sorter’s output stream. As visualized in Fig. 1, the ore sorting design could be divided into three main subsystems: i) the pre-treatment unit, ii) the multi-sensing system, and iii) the air nozzle system. A pretreatment unit consisting of drummer - siever is integrated in order to feed the industrial conveyor belt in an even and proper way, while it mechanically separates the samples based on their size. In addition, a multi-sensing system is developed to achieve the object detection and classification process by utilizing a RGB camera and a Hyperspectral camera respectively [22, 23]. After the sensing components, the final separation procedure is applied by the air nozzle system, which separates the ore samples into two different buckets, one for the pure magnesite samples and one for the rejected ones. The orchastration of the whole sorting system is achieved by the deployment of AI-based models and the integration of the industrial control system [24]. Finally, an offline XRF analysis and RAMAN spectroscopy are taking place providing chemical data aggregated from the ore samples and enhancing the validating process of the developed models and algorithms for the classification. To this end, in the aforementioned subsections, the components of the pipeline will be further analysed. A. Material Feeding The input stream of the proposed system consists of a variety of extracted minerals, such as magnesite, dolomite, calcite, sepiolite, peridotite and serpentinite. Those minerals are differentiated from each other with multiple optical and chemical criteria, i.e. color, magnetic properties, texture, etc. Specifically, the magnesite sample is a white colored ore with dull and clay-like texture. Thus, achieving to detect and classify magnesite from the other ore is not that accurate by using only optical sensor (RGB camera). For this reason, inline hyperspectral camera and offline X-Ray Fluorescence (XRF) analysis and RAMAN spectroscopy are applied. Fig. 3. The input stream. B. Intelligent Control System 1) Multi-sensing system: The multi-sensing system of the ore sorting as already mentioned consists of two parts: the inline and the offline part. Starting with the inline part, an RGB Basler ace 2 camera is used providing spatial information (e.g. shape, volume, position on the conveyor-belt, color, etc.). These data are feeding the object detection model, which respectively broadcasts this data to the air nozzle system. After the RGB camera, the hyperspectral camera is mounted and operates in wavelengths up to 1700nm with 230 bands, providing spectral data to the intelligent system orchestrated by ML/DL models [25]. Those CNN models have initially been trained with multi-spectral dataset and then proceed with the segmentation and classification processes of the inline passing through samples [26]. The final class of each ore sample is sent to the control unit of the air-nozzle system which will open-close the proper air valves for blowing the right sample to the right bucket [27]. Regarding the offline sensing process, a random selection of ore samples coming from the input stream is scanned using a handheld XRF sensor. The result of this analysis is an amount of peaks, one peak for each sample that signifies the material. Thus, in order to calculate the percentage of the elements in each sample, firstly, the peaks of each element are identified, e.g. for magnesite is 1.25keV, and then is calculated the peak area of each peak to find which element is found in maximum amount in each sample [28]. The data collected from XRF analysis are, also, compared with the results, i.e. peaks per sample, from RAMAN spectroscopy [29]. Finally, the offline data are retrofitting the AI-driven models reassuring high classification accuracy levels, while validating the inline classification process. 2) Intelligent Control Unit: To achieve high efficiency of the sorting system, the components of the whole architecture are orchestrated by the intelligent control unit. The sensors and actuators of the system transmit their collected data to a PLC S7-1200 by using the Fig. 2. Ore sorting design 361 Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on September 29,2025 at 09:19:28 UTC from IEEE Xplore. Restrictions apply. PROFINET protocol facilitating the PLC’s controlling and monitoring function. In parallel, PLC together with an inverter manage the conveyor’s motor speed reassuring the sorting system’s adjustability on demand based on the requirements. To upload and handle online the collected data from the physical layer, the PLC is connected with an IoT2050 Gateway, which transfers the data to the cloud platform enhancing their accessibility. Finally, the described industrial system is able to be controlled and monitored by an HMI located at the industrial premises and thus providing multiple control options to the user either physical or remote [24]. C. Air – nozzle actuation system At the end of the pipeline of ore sorting system, the airnozzle actuation system is located with the main objective to proceed with the final separation of the samples into two divided buckets [30]. The air-nozzle system is an automated actuation system consisting of an air compressor, valves, the nozzles and the control unit. In particular, high-pressure air is continuously delivered by the compressor to the system, while the valves manage the air flow to the pipes digitally with an ON/OFF function regulating the passage of air. Then, the nozzles with a diameter of 1mm are connected with the air pipes and strike directed air in a beam that displaces the ore samples. The airnozzle system in order to relocate the samples falling from the conveyor-belt and blow them to the proper bucket, the sorter’s classification and object detection models send their data to the control unit of the actuation system. Thus, by knowing the coordinates on the conveyor belt of each sample and the class that it belongs, the nozzles hit with an air-beam in the proper pressure and achieve the final separation. V. DISCUSSION The design demonstrates a self-adaptable ore sorting system which is designed to be modular with the goal to be adjusted in challenging mining environments. Specifically, the design could be adapted to the ground anomalies of the mining field that will be installed, while also to the spatial availability. Moreover, the system is properly designed in order to be dust and water tolerant, characteristics that are usually found in mines. Depending on the mining industry (e.g. coal, steel, iron, tungsten, copper, etc.), new intelligent models will be developed and applied enhancing and expanding the classification accuracy of the sorting system in multiple materials and CRMs. Thus, retrofitting ability of the system is reinforced by enriching the spectral datasets of the samples. In addition, adaptable control ability of the sorting system is achieved by creating a hybrid mode either cloud-based or physical enable the data exchange and process on demand and in real-time. To this end, based on the input stream the system can change the intelligent system by applying the proper algorithms. In parallel, regarding the load of the input stream the conveyor’s speed can be rapidly adapted, as well as the sensors and actuators process payload. As far as the limitations of the presented system, on the one hand, the self-adaptable ore sorting system is designed and functionable for small sized minerals, while the actuation system (air-nozzle system) has size and weight restrictions. On the other hand, the amount of dust in the input stream may interfere in the results accuracy. However, measurements such as cleaning the ore streams or using a sieving vibrator as a pretreatment unit decrease the system’s dust intolerance. VI. CONCLUSION & FUTURE STEPS In this research, a self-adaptable ore sorting is analysed. Specifically, this system receives as input mixed ore streams aiming to sort them into pure magnesite samples and the rest minerals. To achieve this, a multi-sensing system consisting of RGB and Hyperspectral camera online and XRF and RAMAN offline is developed collecting spatial, spectral and chemical data from the samples. These data are processed by CNN models in order to classify each sample based on the material and then by implementing an air-nozzle actuation system proceed with the final separation process. In accordance with the constantly rising need for raw material extraction and EU regulations, the self-adaptable ore sorting by following Mine 4.0 principles provide a holistic solution to the ore waste management, while by reducing the ore waste and extracting pure CRMs from mixed ore stream, it reduces the need for further extractions. As future steps, we are planning not only to expand the application fields of the sorting system, but also expand the CRM dataset and enhance ore and specifically CRM waste management, while also continuing implementing state-of-the-art technologies. ACKNOWLEDGMENT This research was financially supported by the European Union’s Horizon Europe research and innovation program under the topic HORIZON-CL4-2022-RESILIENCE-01-06 called MASTERMINE with number 101091895. REFERENCES [1] Konstantindis, Fotios K., Antonios Gasteratos, and Spyridon G. Mouroutsos. "Vision-based product tracking method for cyber-physical production systems in industry 4.0." 2018 IEEE international conference on imaging systems and techniques (IST). IEEE, 2018. [2] Balakera, N., Tzelepi, V., Konstantinidis, F. K., Tsimiklis, G., & Amditis, A. (2024). Green ICT Methodology for Energy Consumption Calculation in ICT Architecture Components. Procedia Computer Science, 232, 19441952. [3] Steven E. Zhang, Julie E. Bourdeau, Glen T. Nwaila, Yousef Ghorbani. "Emerging criticality: Unraveling shifting dynamics of the EU's critical raw materials and their implications on Canada and South Africa." Resources Policy 86 (2023): 104247. [4] Padovano, A., Sammarco, C., Balakera, N., & Konstantinidis, F. (2024). Towards sustainable cognitive digital twins: A portfolio management tool for waste mitigation. Computers & Industrial Engineering, 110715. [5] https://www.mastermine-project.eu/ [6] Bartnitzki, Thomas. "Mining 4.0: Importance of Industry 4.0 for the raw materials sector." Artif. Intell 2.1 (2017): 25-31. [7] El Hiouile, Laila, Ahmed Errami, and Nawfel Azami. "Towards Mine 4.0: A Proposed Multi-Layered Architecture for Real-Time Surveillance and Anomaly Detection in an Open-Pit Phosphate Mine." Mining 4.3 (2024): 672-686. [8] Konstantinidis, F. K., Sifnaios, S., Tsimiklis, G., Mouroutsos, S. G., Amditis, A., & Gasteratos, A. (2023). Multi-sensor cyber-physical sorting system (cpss) based on industry 4.0 principles: A multi-functional approach. Procedia Computer Science, 217, 227-237. [9] A. Bolotov and V. Kondrat’ev, “Hydrometallurgical Processing of Tungsten Ores: A Review,” Minerals Engineering, 2018. [10] N. Okada, Y. Maekawa, N. Owada, K. Haga, A. Shibayama, and Y. Kawamura, “Automated Identification of Mineral Types and Grain Size Using Hyperspectral Imaging and Deep Learning for Mineral Processing,” Minerals, vol. 10, no. 9, 2020. 362 Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on September 29,2025 at 09:19:28 UTC from IEEE Xplore. Restrictions apply. [11] H. Knapp, M. Robben, M. Dehler, and H. Wotruba, “X-ray Transmission Sorting of Tungsten Ore,” in Conference Proceedings, 2013. [12] E. E. Urbano, J. Costa, L. Graça, and R. Scholz, “Ore-waste and Ore Type Classification Using Portable XRF: A Case Study of an Iron Mine from the Quadrilátero Ferrífero, Brazil,” Geologia USP - Serie Cientifica, vol. 20, pp. 3–15, June 2020. [13] H.J. Wang, X.P. Liu, G. Wang, and S. Han, “Optimization of Mobile Manipulator Sorting Path Based on Improved Genetic Algorithm,” J. Beijing Univ. Posts Telecommun., vol. 43, no. 5, p. 34, 2020. [14] X. Luo, K. He, Y. Zhang, P. He, and Y. Zhang, “A Review of Intelligent Ore Sorting Technology and Equipment Development,” International Journal of Minerals, Metallurgy and Materials, vol. 29, no. 9, pp. 1647– 1655, Sep. 2022. [15] Arvidson, “Sorting: Possibilities, Limitations and Future,” in Diva Portal, 2022. [16] Lessard and McHugh, “Development of Ore Sorting and Its Impacts on Mineral Processing Economics,” CEEC, 2020. [17] D. Peukert, C. Xu, and P. Dowd, “A Review of Sensor-Based Sorting in Mineral Processing: The Potential Benefits of Sensor Fusion,” Minerals, vol. 12, no. 11, p. 1364, 2022. [18] C. Bergmann, “Developments in Ore Sorting Technologies,” Council for Mineral Technology, Randburg, p. 75, 2009. [19] C.Y. Xu, “The Application of Photoelectric Color Separator and the Optimization in Mineral Separation Process of Some Mine,” World Nonferrous Met., no. 17, p. 31, 2016. [20] N. G. Cutmore and J. E. Eberhardt, “The Future of Ore Sorting in Sustainable Processing,” in Green Processing, AusIMM, 2001. [21] R. Crosby and M. Buxton, “Sensor-Based Ore Sorting Technology in Mining—Past, Present and Future,” MDPI, 2020. [22] Konstantinidis, F. K., Sifnaios, S., Arvanitakis, G., Tsimiklis, G., Mouroutsos, S. G., Amditis, A., & Gasteratos, A. (2023). Multi-modal sorting in plastic and wood waste streams. Resources, Conservation and Recycling, 199, 107244. [23] Evangeliou, N., Giakoumidis, N., Chaikalis, D., Tsoukalas, A., Unlu, H. U., Xing, D., & Tzes, A. (2022). Mechatronic design of a delivery octarotor drone. International Journal of Mechanical Engineering and Robotics Research, 11(5), 351-356. [24] Balakera, N., Konstantinidis, F. K., Tsimiklis, G., Latsa, E., & Amditis, A. (2023, February). Iiot network system from data collection to cyberphysical system transmission under the industry 5.0 era. In International Congress on Information and Communication Technology (pp. 929-941). Singapore: Springer Nature Singapore. [25] Sifnaios, S., Zorzos, I., Arvanitakis, G., Konstantinidis, F. K., Tsimiklis, G., & Amditis, A. (2023, October). Exploration and Mitigation of the Impact of Lighting Conditions on Multi-spectral Image Classification. In 2023 IEEE International Conference on Imaging Systems and Techniques (IST) (pp. 1-6). IEEE. [26] Sifnaios, S., Arvanitakis, G., Konstantinidis, F. K., Tsimiklis, G., Amditis, A., & Frangos, P. (2024). A Deep Learning Approach for Pixellevel Material Classification via Hyperspectral Imaging. arXiv preprint arXiv:2409.13498. [27] Zoumpoulis, P., Konstantinidis, F. K., Tsimiklis, G., & Amditis, A. (2024, May). Advancing Urban Waste Management Using Industry 5.0 Principles: A Novel Smart Bin. In 2024 IEEE International Workshop on Metrology for Industry 4.0 & IoT (MetroInd4. 0 & IoT) (pp. i-vi). IEEE. [28] Urbano, E. E. M. C., Costa, J. F. C. L., Graça, L. M., & Scholz, R. A. C. (2020). Ore-waste and ore type classification using portable XRF: a case study of an iron mine from the Quadrilátero Ferrífero, Brazil. Geologia USP. Série Científica, 20(2), 3-15. [29] Orlando, A., Franceschini, F., Muscas, C., Pidkova, S., Bartoli, M., Rovere, M., & Tagliaferro, A. (2021). A comprehensive review on Raman spectroscopy applications. Chemosensors, 9(9), 262. [30] Luo, X., He, K., Zhang, Y., He, P., & Zhang, Y. (2022). A review of intelligent ore sorting technology and equipment development. International Journal of Minerals, Metallurgy and Materials, 29(9), 16471655. 363 Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on September 29,2025 at 09:19:28 UTC from IEEE Xplore. Restrictions apply. View publication stats