The Integration of Artificial Intelligence (AI) and Machine Learning (ML) in Mineral Processing: A Mini Review

The mineral processing industry is a critical component of global economic development and sustainable resource management, particularly in the era of electrified mobility and renewable-energy technologies. However, the sector faces twin challenges: declining ore grades and increasing pressure to minimize its environmental footprint, including water and energy use [1,2]. These challenges expose the limitations of traditional process-control strategies, which often struggle to manage the inherent complexity, nonlinearity, and variability of mineral processing circuits [3].

The emergence of the Fourth Industrial Revolution (Industry 4.0) has created the technological foundation needed to address such long-standing inefficiencies [4]. The proliferation of advanced sensor networks, low-cost connectivity, and scalable computing infrastructure has resulted in an unprecedented flow of data from the plant floor. Modern mineral processing facilities are equipped with vision systems, online analysers (e.g., prompt gamma neutron activation analysis, laser-induced breakdown spectroscopy), vibration sensors, and other smart instruments that collectively generate large volumes of multidimensional, high-frequency data [1-5]. Yet this data deluge introduces its own challenge: the information complexity far exceeds human analytical capacity, making it difficult to translate raw data into actionable insights [6].

Within this context, Artificial Intelligence (AI) and Machine Learning (ML) have emerged as pivotal technologies, marking a paradigm shift from reactive control to predictive and prescriptive analytics [7]. These methods can identify complex, non-linear correlations within multivariate process data that are typically imperceptible to human operators [8]. Early applications of AI in mining and metallurgy-primarily in the form of rule-based expert systems-date back several decades [9], but progress was limited by computational constraints and data scarcity. Today, the convergence of big data, high-performance computing, and advanced algorithms has transformed AI from a theoretical concept into a practical enabler of operational excellence [10,11].

This mini-review synthesizes current progress in integrating AI and ML into mineral processing operations. It explores transformative applications in ore characterization, process optimization, and predictive maintenance, evaluates barriers to large-scale implementation, and identifies future research directions that may accelerate the transition toward intelligent, autonomous mineral processing plants.

The application of AI and ML in mineral processing is diverse, extending beyond proof-of-concept studies to deliver measurable improvements in process performance, efficiency, and safety. This section highlights three critical domains in which AI/ML integration has demonstrated substantial industrial impact-from ore characterization and sorting to process optimization and predictive maintenance.

Ore Characterization and Sorting

The heterogeneity of run-of-mine ore is a primary source of process variability, causing downstream fluctuations in grade, throughput, and recovery. Conventional ore characterization methods, such as laboratory assays, provide delayed and discontinuous feedback, limiting real-time control. AI, particularly computer vision and deep learning, is transforming this space.

Convolutional Neural Networks (CNNs) can now analyse highresolution images from conveyor-belt cameras or drone-based inspections to perform near-real-time lithological classification and grade estimation [12,13]. These networks identify and quantify mineralogical features based on visual cues such as colour, texture, and reflectance [14]. Such capabilities underpin modern sensorbased sorting systems, which use AI-driven insights to trigger pneumatic or mechanical separation mechanisms, rejecting barren material prior to energy-intensive comminution [15]. This pre-concentration step reduces energy and water consumption per unit of valuable metal produced while stabilizing feed quality for downstream processes [16]. Furthermore, integrating hyperspectral and laser-induced breakdown spectroscopy (LIBS) data with ML algorithms-such as principal component analysis (PCA) and clustering-enables a richer geochemical understanding of ore variability in real time, enhancing sorting selectivity and efficiency [17].

Process Modelling and Optimization

Core mineral processing operations, particularly comminution and flotation, exhibit strong non-linearities, multivariable dependencies, and time delays, making them challenging to control through traditional Proportional–Integral–Derivative (PID) or Linear Model-Predictive Control (MPC) systems. ML techniques excel in such environments by learning complex input–output relationships directly from operational data, providing predictive models that capture interactions among feed characteristics, reagent dosage, and key Performance Indicators (KPIs) such as grade, recovery, and particle size distribution [18,19].

Process Modelling and Optimization

Core mineral processing operations, particularly comminution and flotation, exhibit strong non-linearities, multivariable dependencies, and time delays, making them challenging to control through traditional Proportional–Integral–Derivative (PID) or Linear Model-Predictive Control (MPC) systems. ML techniques excel in such environments by learning complex input–output relationships directly from operational data, providing predictive models that capture interactions among feed characteristics, reagent dosage, and key Performance Indicators (KPIs) such as grade, recovery, and particle size distribution [18,19].

Comminution:In crushing and grinding circuits, ML models such as Gradient Boosting Machines (GBMs) and Artificial Neural Networks (ANNs) are used to predict product size and optimize mill load and power consumption. By integrating sensor data from acoustic emissions, bearing pressure, and motor power, these models can recommend setpoints for feed rate and mill speed to maximize throughput while minimizing overgrinding and mechanical wear [20,21].

Flotation:Froth flotation has become one of the most prominent application areas for AI. ML algorithms trained on froth image features-such as bubble size distribution, froth velocity, and texture stability-use CNNs to infer process performance parameters like concentrate grade and recovery [22,23]. Supervised learning models then relate these image features to control variables (air flow, froth depth, reagent dosage), allowing for dynamic, closedloop control [24].

Beyond supervised learning, Reinforcement Learning (RL) has emerged as a transformative technology enabling autonomous process control. RL agents interact continuously with a simulated or real process environment to learn optimal control policies that maximize a long-term reward, such as cumulative metal recovery or energy efficiency [25,26]. Successful implementations often adopt hybrid architectures that combine deep learning for feature extraction with physics-informed or regression models for interpretable prediction [11].

Predictive and Prescriptive Maintenance

Unplanned downtime of critical assets-such as SAG mills, crushers, and slurry pumps-remains a major contributor to production losses and maintenance costs. The transition from preventive (time-based) to predictive (condition-based) maintenance represents a major shift toward data-driven reliability management.

By analysing time-series data from vibration, temperature, acoustic, and motor-current sensors, ML models can detect early signs of mechanical degradation long before catastrophic failure occurs [27,28]. This approach allows maintenance activities to be scheduled based on asset condition rather than fixed intervals, improving equipment availability and operational safety [29]. Advanced anomaly-detection algorithms (e.g., autoencoders, isolation forests, one-class SVMs) identify deviations from normal behaviour, while regression and survival models forecast remaining useful life (RUL) of components such as mill liners or pump impellers [30].

The next stage-prescriptive maintenance-extends predictive capabilities by recommending specific actions to prevent or mitigate failures. This is supported by digital twin technology, which creates dynamic, virtual replicas of physical assets. When coupled with real-time sensor feeds, digital twins enable simulation of alternative maintenance strategies and performance outcomes, guiding operators toward cost-optimal decisions [31,32]. Industrial implementations have shown that AI-driven maintenance frameworks can improve overall equipment effectiveness (OEE) by up to 20 % and reduce unplanned downtime by over 50 % [33].

The implementation of AI in mineral processing has advanced from conceptual demonstrations to robust, plant-scale applications through an array of specialized software, hardware, and datamanagement tools. Table 1 summarizes the principal AI subfields, key technologies, and their relevance to specific mineral-processing functions. The classification is organized to reflect the data flow across a typical plant-from ore feed and processing control to asset management and system-wide integration.

Table 1: Representative AI Tools and Technologies in Mineral Processing.

Illustrative Case Examples

(a) Intelligent Ore Sorter using CNNs: A U-Net convolutional architecture processes high-resolution belt images in real time. Each rock fragment is segmented and classified as “ore” or “waste” and the output mask triggers pneumatic actuators within milliseconds. This results in accurate rejection of barren material, improving energy efficiency and feed consistency.

(b) Froth Flotation Optimizer with XG Boost and Deep Q-Learning: An XG Boost model predicts concentrate grade several minutes ahead based on froth image features, chemical assays, and air-flow data. Its output informs a Deep Q-Network (DQN) agent trained to maximize a composite reward functione. g., (Grade × Recovery) – Reagent Cost-thereby enabling autonomous reagent-dosage control.

(c) Mill-Liner Wear Prognostics with a Digital Twin: Sensor data (vibration and acoustic) feed a Random Survival Forest model that estimates liner failure probability. These predictions are visualized in a digital twin of the SAG mill, allowing maintenance planners to simulate the trade-off between continued operation and immediate liner replacement, thereby supporting prescriptive, cost-optimal decisions.

Discussion

The ecosystem of AI tools in mineral processing demonstrates a clear trend toward integration-across data collection, model development, and real-time decision support. Cloud and edge computing platforms provide the scalability necessary for industrial deployment, while MLOps frameworks ensure reproducibility and continuous model improvement. Equally critical are explainability techniques such as SHAP and LIME, which help translate opaque model outputs into interpretable insights for engineers and operators, fostering confidence and compliance in AI-assisted control environments [35,36].

Despite remarkable progress, the widespread adoption of AI and ML in mineral processing remains constrained by several technical, organizational, and human-centric challenges. These barriers must be systematically addressed to unlock the full potential of intelligent, autonomous mineral processing systems.

High-quality, representative data are the foundation of any AI system. However, mineral processing data are often heterogeneous, sparse, and noisy, reflecting diverse sensors, process configurations, and measurement frequencies. The lack of standardized data architectures and ontologies hampers interoperability between sites and vendors [34]. Moreover, labelled datasets-essential for supervised learning-remain scarce because of limited historical records and confidentiality restrictions within mining companies.

Future work should prioritize the creation of open, anonymized benchmark datasets and the adoption of Industrial Data Space (IDS) principles to enable secure data sharing and model transferability.

The “black-box” nature of advanced AI models-especially deep neural networks and reinforcement-learning agents-often undermines user confidence. In critical process industries such as mining, operators must understand why an algorithm recommends a certain action before implementing it in production. Explainable AI (XAI) methods, including Shapley Additive explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), have emerged as effective tools to improve transparency [35,36]. Developing domain-specific XAI frameworks that can relate model behaviour to physical and chemical process parameters will be key to integrating AI into day-to-day operations.

Bridging the skills gap is another critical requirement for successful AI integration. Most metallurgical and processengineering curricula still emphasize traditional control and design methods rather than data-centric thinking. Consequently, there is a growing need for interdisciplinary training programs that combine mineral-processing expertise with data science, machine learning, and automation [37]. Equally important is cultivating a culture of collaboration between domain experts and data scientists. Embedding AI specialists within operational teams can accelerate adoption and ensure that algorithms address real industrial challenges rather than abstract optimization goals.

Future research is expected to focus on hybrid modelling, where first-principles (mechanistic) equations are coupled with datadriven approaches to exploit their complementary strengths [38]. These physics-informed models promise improved generalizability across ore types and plant configurations. Furthermore, deep reinforcement learning (DRL) presents a compelling path toward plant-wide optimization and fully autonomous control loops [39]. Transfer learning and domain adaptation techniques will also play a major role in scaling ML solutions across geographically and geologically diverse operations [40]. Finally, the ongoing evolution of IoT, cloud computing, and edge analytics will expand the realtime data infrastructure required to support these advanced AI frameworks, enhancing connectivity and decision-making across the entire mining value chain [41].

Artificial Intelligence (AI) and Machine Learning (ML) are reshaping the mineral processing industry, shifting it from empirical and experience-based operation toward data-driven, adaptive, and increasingly autonomous systems. Their integration across key domains-ore characterization, process optimization, and maintenance-has demonstrated tangible benefits in productivity, resource efficiency, and safety. AI-driven approaches enable a deeper understanding of process dynamics, facilitate real-time optimization, and support predictive and prescriptive decisionmaking. These advancements not only enhance recovery and reduce energy and water consumption but also contribute to sustainability objectives by lowering environmental impact and operational risk.

Despite the substantial progress, the transition from isolated pilot applications to fully autonomous, AI-empowered plants is still in its early stages. Persistent barriers-including limited data availability, model interpretability, and workforce readinessrequire targeted research and strategic organizational adaptation. In this context, explainable AI, hybrid modelling, and reinforcement learning are expected to play pivotal roles in bridging the gap between theoretical promise and industrial practice.

Ultimately, the future of mineral processing will be defined by the ability to integrate AI technologies seamlessly with existing process-control frameworks and domain expertise. By fostering collaboration between metallurgists, control engineers, and data scientists, the industry can accelerate the realization of intelligent, sustainable, and resilient mineral processing operations capable of meeting the demands of a rapidly evolving global economy.

Both authors equally contributed in every part of this work.

The authors declare no conflicts of interest.

1. E Hay, M Nehring, P. Knights, M S Kizil (2019) Determining the optimal orientation of ultimate pits for mines using fully mobile in-pit crushing and conveying systems. Journal of the Southern African Institute of Mining and Metallurgy 119(11): 949-956.
2. S A Northey, G M Mudda, T T Werner, S M Jowitt, N Haque, M, et al. (2017) The exposure of global base metal resources to water criticality, scarcity and climate change. Global Environmental Change (44): 109-124.
3. N O Lotter Modern (2011) Process Mineralogy. An integrated multi-disciplined approach to flowsheeting 24(12): 1229-1237.
4. H Lasi, P Fettke, H G Kemper, T Feld, M Hoffmann, et al. (2014) Industry 4.0. Business & Information Systems Engineering (6): 239-242.
5. K Anvari, J Benndorf (2025) Real Time Mining–A Review of Developments Within the Last Decade. Mining 5(3): 38-40.
6. B A Wills, J A Finch (2015) Wills' Mineral Processing Technology. An Introduction to the Practical Aspects of Ore Treatment and Mineral Recovery. Butterworth-Heinemann, Wills' Mineral Processing Technology - 8th Edition | Elsevier Shop
7. S Russell, P Norvig (2021) “Artificial Intelligence: A Modern Approach.” Pearson, (2021). Artificial Intelligence: A Modern Approach
8. M. Bishop (2006) Pattern Recognition and Machine Learning. Springer.
9. S L Jamsa-Jounela, A J Niemi (1992) Expert systems in mineral and metal processing. Pergamon Press, Expert systems in mineral and metal processing - Aalto University's research portal
10. D Hodouin, S L Jaams-Jounela, M T Carvalho, L Bergh (2001) State of the art and challenges in mineral processing control. Control Engineering Practice 9(9): 995-1005.
11. Z Liu, Z Zhang, W Zhang (2025) A Hybrid Framework Integrating Traditional Models and Deep Learning for Multi-Scale Time Series Forecasting. Entropy 27(7) 695-700.
12. C Robben, H. Wotruba (2019) Sensor-Based Ore Sorting Technology in Mining-Past, Present and Future. Minerals 9(9) 523.
13. Y Fu, C Aldrich (2020) Deep Learning in Mining and Mineral Processing Operations: A Review. IFAC Papers Online 53(2) 11920-11925.
14. L Perez-Barnuevoa, E Pirard, R Castroviejo (2013) Automated characterization of intergrowth textures in mineral particles. A case study. Minerals Engineering (52): 136-142.
15. S Luukkanen A, Tanhua Z, Zhang, R Mollehuara, Canales I, et al. (2022) Towards waterless operations from mine to mill. Minerals Engineering (187): 107793-107799.
16. C Robben H Wotruba (2019) Sensor-Based Ore Sorting Technology in Mining-Past, Present and Future. Minerals 9(9): 523-530.
17. L Nagli, M Gaft, Y Raichlin, I Gornushkin (2018) Cascade generation in Al laser induced plasma. Optics Communications 415(15): 127-129.
18. M Martinovic, K Dokic, D Pudic (2025) Comparative Analysis of Machine Learning Models for Predicting Innovation Outcomes: An Applied AI Approach. Applied Sciences, 15(7): 3636-3640.
19. J Henriques, PM Castro, R Dias, B Magalhaes, M Estrela, et al. (2023) Potential Industrial Synergies in the Steelmaking and Metal-Processing Industry: By-Products Valorization and Associated Technological Processes. Sustainability 15(21): 15323-15330.
20. S Streicher J A du Preez (2021) Strengthening Probabilistic Graphical Models: The Purge-and-merge Algorithm. IEEE Access (9): 149423-149432.
21. R J Batterham (2017) The mine of the future-Even more sustainable. Minerals Engineering (107): 2-7.
22. Y Fu C Aldrich (2018) Froth image analysis by use of transfer learning and convolutional neural networks. Minerals Engineering (115): 68-78.
23. J T McCoy, L Auret (2019) Machine learning applications in minerals processing: A review. Minerals Engineering (132) 95-109.
24. A Szmigiel, D B Apel, K Skrzypkowski, L Wojtecki, Y Pu et al (2024) Advancements in Machine Learning for Optimal Performance in Flotation Processes: A Review. Minerals 14(4): 331-335.
25. B J Shean, JJ Cilliers (2011) A review of froth flotation control. International Journal of Mineral Processing 100(4): 57-71.
26. H Panetto, B Iung, D Ivanov, G Weichhart, X Wang, et al. (2019) Challenges for the cyber-physical manufacturing enterprises of the future. Annual Reviews in Control (47): 200-213.
27. A K S Jardine, D Lin, D Banjevic (2006) A review on machinery diagnostics and prognostics implementing condition-based maintenance. Mechanical Systems and Signal Processing 20(7) 1483-1510.
28. X Bampoula, G Siaterlis, N Nikolakis K (2021) Alexopoulos. A deep Learning Model for Predictive Maintenance in Cyber-Physical Production Systems Using LSTM Autoencoders. Sensors 21(3): 972-980.
29. R K MobleyAn (2002) Introduction to Predictive Maintenance. 2nd edition, Butterworth-Heinemann, an-introduction-to-predictive-maintenance.pdf
30. Z Liu, P Zhang, Y Yu, M Li, Z Zeng, et al. (2024) A novel fault diagnosis model of rolling bearing under variable working conditions based on attention mechanism and domain adversarial neural network. Journal of Mechanical Science and Technology 38 (3): 1101-1111.
31. Y Li, J Xiong, Z Yin (2019) Molten pool stability of thin-wall parts in robotic GMA-based additive manufacturing with various position depositions. Robotics and Computer-Integrated Manufacturing 56 (1-12): 1-11.
32. E Negri, L Fumagalli, M Macch (2017) A Review of the Roles of Digital Twin in CPS-based Production Systems. Procedia Manufacturing (11): 939-948.
33. B R Rakishev, AA Orynbay, A B Mussakhan, A I Tuktibayev (2022) Computer-aided design of rational parameters for the location of blasthole charges in horizontal underground development. Mining Technology 131(1) 25-37.
34. I H Sarker (2021) Machine Learning: Algorithms, Real World Applications and Research Directions. SN Computer Science (2): 160-165.
35. S M Lundberg, S I. Lee (2017) A unified approach to interpreting model predictions. NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Systems 4768-4777.
36. W Samek, K R Muller (2019) Towards Explainable Artificial Intelligence. in Explainable AI: Interpreting, Explaining and Visualizing Deep Learning. W. Samek, G. Montavon, A. Vedaldi, L.K. Hansen, K.-R. Muller eds. Lecture Notes in Artificial Intelligence. Springer.
37. T Kotsiopoulos, P Sarigiannidis, D Ioannidis, D Tzovaras (2021) Machine learning and deep learning in smart manufacturing: The smart grid paradigm. Computer Science Review (40) 100341-100345.
38. J T McCoy, S Kroon, L Auret (2018) Variational Autoencoders for Missing Data Imputation with Application to a Simulated Milling Circuit. IFAC PapersOnLine 51(21): 141-146.
39. C Vignon, J Rabault, R Vinuesa (2023) Recent advances in applying deep reinforcement learning for flow control: Perspectives and future directions. Physics of Fluids (35) 031301-031305.
40. A Milenkoski, M Vieira, S Kounev, A Avritzer, B D Payne, et al. (2015) Evaluating Computer Intrusion Detection Systems: A Survey of Common Practices. ACM Computing Surveys (CSUR) 48(1): 1-41.
41. I Ficili, M Giacobbe, G Tricomi A Puliafito (2025) From Sensors to Data Intelligence: Leveraging IoT, Cloud, and Edge Computing with AI. Sensors 25(6): 1763-1765.