Keywords: Fault diagnosis, process monitoring, Mining operations, mineral processing
Abstract: Fault Tolerant Control (FTC) plays a crucial role in ensuring the reliable operation of complex systems, particularly in industries like mining, where equipment failure can lead to significant operational and financial consequences. This talk explores the challenges associated with implementing FTC in the mining industry. It outlines the fundamental concepts of FTC, which are designed to maintain system performance despite faults or component failures. Through examples drawn from the mining sector, the talk highlights real-world challenges such as system complexity, uncertainty, and the need for effective fault detection and isolation strategies. The talk provides a comprehensive overview of the current state of FTC research and its practical applications, emphasizing the importance of ongoing innovation to improve the efficiency, safety, and reliability of mining operations.

Keywords: Artificial intelligence, machine learning systems, and human machine systems, Automation, instrumentation, Mining operations, mineral processing
Abstract: This study proposes a method to quantify uncertainty in soft sensors' measurements for estimating ore mass flow rate on conveyor belts in mining. A linear regression model previously implemented in a PLC is extended with Conformal Prediction (CP) and a sliding window to generate adaptive prediction intervals. Residuals are updated incrementally for efficiency and adaptability. One-way ANOVA and Tukey’s HSD showed that window size significantly affects interval coverage and width. Larger windows (W100) yielded wider intervals (286.62 t/h) and higher coverage (95.3%), while smaller windows (W40) were narrower (243.97 t/h) with lower coverage (84.9%) but greater responsiveness. Processing time stayed under 0.1 seconds across all configurations, confirming suitability for real-time PLC use. The approach balances robustness and responsiveness, offering a lightweight, interpretable solution for uncertainty-aware control in industrial environments.

Keywords: Fault diagnosis, process monitoring, Ore preparation, flotation, Mining operations, mineral processing
Abstract: Since several decades, most measurements in mineral processing plants have been reconciled to improve their accuracy and validity. However, existing methods cannot simultaneously reconcile mineral grades, liberation distributions, and elemental assays. This work proposes a multilinear framework for reconciling measurements from an industrial flotation cell, including particles flowrates, solids fractions, particle size distributions, elemental and mineralogical assays, and mineral liberation distributions for complex lithologies. Results demonstrate improved reliability of performance indicators, such as recovery, by enforcing mass balance equations. A limited amount of minerals are considered by iterating on the gangue elemental composition, thus greatly reducing the problem size.

Keywords: Artificial intelligence, machine learning systems, and human machine systems, Fault diagnosis, process monitoring, Mining operations, mineral processing
Abstract: Mining operations are increasingly confronted with a multitude of challenges, including price volatility, declining ore grades, and escalating energy costs. These challenges are exacerbated by variations in mineral hardness, which contribute to accelerated wear on critical equipment. Among this machinery, ball mills are particularly susceptible to wear and component failures, leading to unplanned maintenance, costly downtime, and disruptions in production. This research seeks to enhance the capabilities of predictive maintenance (PdM), with a concentrated emphasis on Anomalous Behavior Detection (ABD) and Digital Twin (DT) technologies, specifically tailored for ball mill applications within a multi-agent AI system (MAS) framework. We present a novel architectural design that synergizes DT and ABD through a semi-autonomous multi-agent AI system comprising two primary agents: the PdM agent and the Quality Assurance Agent. The primary function of the PdM agent is to identify anomalous behavior, while the Quality Assurance Agent is tasked with assessing the implications of parameter modifications on mill efficiency. Furthermore, we described the principal challenges related to data quality, system integration, and real-time responsiveness that must be systematically addressed to facilitate successful implementations in the future.

Keywords: Fault diagnosis, process monitoring, Artificial intelligence, machine learning systems, and human machine systems
Abstract: In industrial processes, foaming detection is typically reactive, relying on fixed thresholds applied to individual process variables. This study presents a methodology for the early prediction of foaming events in the CO2 removal section of an ammonia production plant. The proposed approach first identifies key variables and then uses Long Short-Term Memory (LSTM) Autoencoder, which are well-suited for learning complex temporal dependencies in multivariate time series data, to achieve early prediction of foaming events. The method is evaluated using data from real foaming events. The results demonstrate its effectiveness in providing early warnings, offering an advantage over traditional reactive detection methods.

Keywords: Fault diagnosis, process monitoring, Mining operations, mineral processing, Automation, instrumentation
Abstract: The deployment of autonomous vehicles in lithium mining operations faces significant safety challenges due to the complexity and unpredictability of mining environments. Reliable hazard detection and response are critical to preventing accidents and operational disruptions. Super alarms, an advanced alarm management strategy, have shown promise in identifying critical event patterns. However, conventional methods are limited by high false alarm rates and poor adaptability in dynamic, real-time systems. This study proposes the use of V-nets, a formalism for managing discrete event sequences, as a novel approach to generating and simulating super alarms for autonomous vehicles in lithium mines. V-nets structure event sequences to enhance the accuracy and adaptability of super alarm systems, addressing the limitations of traditional rule-based or statistical models. The proposed approach enables a context-aware and predictive alarm mechanism, reducing false alarms while improving the detection of hazardous situations. A simulation framework models autonomous vehicle trajectories in a mining environment, integrating V-net-based super alarm generation. Performance is evaluated using key metrics such as detection accuracy, response time, and false alarm rates. Results demonstrate that V- nets significantly improve alarm precision, reducing unnecessary alerts by 77% while achieving 96.3% accuracy in detecting critical safety events. These findings highlight the potential of V-nets in advancing alarm management for industrial automation, offering a scalable and intelligent solution for safety-critical mining applications. This research provides a foundation for future implementations of V-nets in real-world autonomous mining operations, contributing to improved risk management, operational efficiency, and safety in the rapidly evolving mining industry.

Keywords: Robotics, mechatronics, Measurement, sensors, Energy, environment, health, safety
Abstract: This paper presents the development of a modular proof-of-concept system capable of autonomously deploying communication nodes from Unmanned Ground Vehicles (UGVs). This enables the establishment of a temporary mesh communication network for hazardous environments, such as subterranean mines and during Search-and-Rescue (SAR) deployments. A prototype system was developed to facilitate the storage and deployment of two distinct types of self-righting communication nodes, each offering unique advantages and disadvantages. The system was designed to be modular, enabling multiple storage units to be stacked to meet the required number of nodes for a particular mission. Experimental testing demonstrated the system’s potential for further development to enhance it’s robustness against the harsh environments encountered in SAR applications and subterranean mining.

Keywords: Ore preparation, flotation
Abstract: Flotation typically represents the first step in mineral processing separation, and a major contributor to recovery losses in the mine-to-metals value chain. Industrial real-time measurement of the pulp and froth conditions of flotation cells provides monitoring, control and optimization benefits.

In this talk, the various industrial approaches to measuring pulp and froth conditions (e.g., computer vision, hydrodynamic, mechanical techniques) will be reviewed and critiqued. The actual and potential application of pulp and froth measurements in industrial monitoring, control and optimization will also be reviewed and illustrated with a number of practical case studies.

Keywords: Ore preparation, flotation
Abstract: Grinding-classification produces ore slurries with proper concentration and size of particles, for flotation process, and meanwhile, a large amount of energy is consumed. Optimal control is benefit to stabilize the particle, maximize the throughput and save energies. However, due to the detection problems and the complexity of the process, there are still many difficulties to achieve the object in practice.

This presentation provides a comprehensive review and assessment of the commonly used control methods and current research progress in grinding operations. It also discusses the key challenges encountered in this field and briefly introduces our recent practices in control of a SABC grinding process.

Keywords: Artificial intelligence, machine learning systems, and human machine systems, Measurement, sensors, Mining operations, mineral processing
Abstract: Multivariate image regression has important applications in mining, mineral processing and manufacturing, where the collection of image data is becoming an integral part of process operations. Deep learning models, and convolutional neural networks in particular, can provide better models than traditional approaches, but their rapid adoption has also brought new challenges, of which the scarcity of reliable labeled data is a major bottleneck. In this study, multivariate image regression on a coal image data set with seven target values only and assays from arsenic froth image data with 30 target values were considered. It is shown that pretrained deep convolutional neural networks can directly generate features that are competitive with traditional methods. On both data sets, they could directly extract features from the images that yielded models as or more reliable than what could be obtained with engineered features sets based on grey level co-occurrence matrices, latent variables and local binary patterns. Fine-tuning yielded further significant improvement on both data sets.

Keywords: Measurement, sensors, Data mining and statistical analyses, Artificial intelligence, machine learning systems, and human machine systems
Abstract: Diffuse reflectance spectroscopy is a non-destructive and low-cost technique capable of capturing a large amount of spectral information relevant to mineral analysis. This study evaluates the feasibility of using such spectral data, complemented with reference measurements obtained through QEMSCAN analysis, to develop robust models for predicting the mineralogical and elemental composition of minerals in copper concentrates. Average spectra were acquired from various types of samples mostly a mixture of pyrite and chalcopyrite originating from three different locations in order to generate mineralogical models for chalcopyrite (%CuFeS2 ) and pyrite (%FeS2 ), as well as elemental models to estimate %Cu, %Fe, and %S contents. The spectral data were preprocessed using techniques such as Savitzky–Golay smoothing (Savgol) and multiplicative scatter correction (MSC) to enhance spectral quality prior to modeling. A regression model based on partial least squares and support vector machines (PLS-SVR) was subsequently trained. The training strategy considered 60% of the data for calibration, 20% for validation, and 20% for prediction. The results yielded coefficients of determination (R2) greater than 0.8 in the prediction phase for %Cu, %Pyrite, and %Chalcopyrite, with root mean square errors (RMSE) of 1.1592 for %Cu, 1.6547 for %Pyrite, and 2.4276 for %Chalcopyrite, suggesting that this methodology holds potential for estimating different types of compositions. Future work will focus on expanding the dataset, improving implementation, and exploring data fusion strategies.

Keywords: Artificial intelligence, machine learning systems, and human machine systems, Ore preparation, flotation, Mining operations, mineral processing
Abstract: This work presents the development of an interpretable machine learning model for predicting silica content in the concentrate of an iron ore flotation process. The explainable boosting machine (EBM) algorithm was chosen for its ability to provide interpretable insights into the rationale behind each prediction, allowing operators to gain a clearer understanding of process conditions and make more informed adjustments. The model was trained using industrial data collected over 18 months from an iron ore flotation plant in Brazil. It achieved 87% recall and 73% precision in detecting the critical class of high silica level, with an overall accuracy of 76%.

Keywords: Artificial intelligence, machine learning systems, and human machine systems, Mining operations, mineral processing, Measurement, sensors
Abstract: This work investigates the structure and potential of incremental learning methods to improve the mining industry by enhancing long-term soft sensor reliability, despite the frequent and dynamic changes in operating conditions typical of this sector. Several incremental learning methods were evaluated using operational data from a case study involving mass flow rate estimation on a crushing circuit. Results showed that these models achieved better long-term accuracy; while requiring a computational cost adequate for industrial application, supporting their suitability for real-world deployment in mining environments.

Keywords: Advanced process control, Artificial intelligence, machine learning systems, and human machine systems, Mining operations, mineral processing
Abstract: This work presents an Adaptive Learning-Based Model Predictive Control (ALB-MPC) framework for a thickening process characterized by high complexity and nonlinear dynamics. The approach leverages operational data to identify an accurate process model as a Nonlinear AutoRegressive eXogenous (NARX) structure, built using a learning method known as Lazily Adaptive Constant Kinky Inference (LACKI). Additionally, a neural network is employed as an online tuning mechanism to adapt the predictive controller parameters and enhance control performance. Simulation results indicate that the proposed control framework achieves performance comparable to or better than controllers with fixed parameters.

Keywords: Artificial intelligence, machine learning systems, and human machine systems, Mining operations, mineral processing, Advanced process control
Abstract: Despite the prevalence of established tuning methods, a significant proportion of industrial control loops remain poorly tuned, often requiring manual intervention or re-tuning due to changing process dynamics. Reinforcement learning (RL) offers a promising solution by enabling automatic controller tuning through interaction with the environment. This paper investigates the use of RL to tune an inverse controller for a nonlinear 2 × 2 bulk tailings treatment (BTT) surge tank system. A proximal policy optimisation (PPO) agent is trained to select safe and effective tuning parameters. Simulation results demonstrate that the RL-tuned controller exceeds benchmark performance.