WO2023242752A1

WO2023242752A1 - System and method for continuous optimization of mineral processing operations

Info

Publication number: WO2023242752A1
Application number: PCT/IB2023/056128
Authority: WO
Inventors: Hugh MCNAMARA; Boris WALTER; Simon Streicher; Grant Kopec; Ronit Ganguly; Denis RUBSTOV
Original assignee: Intellisense.Io
Priority date: 2022-06-15
Filing date: 2023-06-14
Publication date: 2023-12-21
Also published as: GB202208791D0; GB2619744A

Abstract

A method for continuously optimizing an ore grinding operation in mining is provided. The method provides an integrated approach utilizing real-time and historical operational data, information generated using virtual sensors, simulation models of process parameters to arrive at an optimized plan and set points for the ore grinding operation. The method improves key performance indicators such as reducing energy consumption in comminution circuit, reducing grinding media waste, increasing grinding efficiency and increases recovery of minerals.

Description

SYSTEM AND METHOD FOR CONTINUOUS OPTIMIZATION OF MINERAL PROCESSING OPERATIONS

TECHNICAL FIELD

The present disclosure relates generally to digital optimization, more specifically, to a system and method for real-time continuous optimization of mining and mineral processing operations.

BACKGROUND

Mining and mineral processing involves a plurality of operations. It is essential to monitor and analyze the entire mineral processing operations to achieve optimized performance. The parameters that control the performances of the plurality of operations in the mining and mineral processing need to be optimized to improve key performance indicators such as cost of mining operations, specific energy consumption in comminution circuit, maximizes the yield of desired particle size, maximizes grade and recovery of mineral of interest. Hence, optimization is a critical factor for safety and profitability in the mining and mineral processing industry.

The mineral processing technology includes grinding, crushing and separation that are carried out to liberate the appropriate minerals from the ore, so that they can be classified from the gangue minerals in the downstream process. Ore grinding is typically done by milling. There are several common techniques for ore grinding, but ball milling is the most common technique. In milling, the key problem of efficiency is that the mill stops for ball charge inspections, as the physical state of the balls, number of balls, etc. are key for the process to occur efficiently (i.e.) for the ore to grind to the correct particle size. Currently, the ore grinding process is dependent on manual or human inspections and semi-empirical models. This results in higher use of grinding media (e.g., grinding balls), more waste and higher energy use per batch to achieve a satisfactory fraction of particles within the correct/desired particle size.

Online optimization is considered a feasible and economical alternative to costly or impractical physical measurements as it uses information available from other measurements and process parameters to calculate an estimate of the quantity of interest. Though existing systems provide online optimization at various levels including design level optimization and operating level optimization, they may not provide continuous optimized operation of the entire chain of operations in real-time.

Therefore, there is a need to address the aforementioned technical drawbacks in existing technologies to provide a method for continuous optimization of mineral processing operations to enhance efficiency and productivity.

SUMMARY

The present disclosure seeks to provide a system and method for continuous optimization of ore grinding operations in real-time. An aim of the present disclosure is to provide a solution that overcomes at least partially the problems encountered in prior art and provide improved methods and systems for continuously optimizing the operations of ore grinding in real-time by predicting the physical changes in the at least one process parameter associated with the ore grinding operation to improve key performance indicators such as reducing energy consumption in comminution circuit, reducing grinding media waste, increasing grinding efficiency and increasing recovery of minerals. The object of the present disclosure is achieved by the solutions provided in the enclosed independent claim. Advantageous implementations of the present disclosure are further defined in the dependent claims.

According to a first aspect, the present disclosure provides a method for continuous optimization of ore grinding operation in mining, wherein the method comprises:

- receiving input data corresponding to a set of variables for at least one process parameter associated with the ore grinding operation;

- providing the received input data to at least one predictive domain model wherein the at least one predictive domain model is representative of one or more changes in the at least one process parameter associated with the ore grinding operation;

- generating predictive output data by executing the at least one predictive domain model that simulates the ore grinding operation, wherein the predictive output data is representative of one or more physical changes in the at least one process parameter associated with the ore grinding operation; and

- optimizing the grinding ore operation by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data.

The method according to the present disclosure enables continuous optimization of ore grinding operation in real-time by predicting the physical changes in the at least one process parameter associated with the ore grinding operation; to reduce the number of mill stops for inspections; more precise control over ore grinding operations to improve efficiency; to reduce energy consumption in comminution circuit; to reduce grinding media waste; to increase grinding efficiency. Optionally, the input data comprises at least one of: historical data, real-time operational data from the ore grinding operation, intermittent data, laboratory information data, test data, design parameters associated with the ore grinding operation.

Optionally, the method further comprises pre-processing of the input data before receiving the input data.

Optionally, the pre-processing of the input data comprises at least one of: imputation of missing data, data frequency modulation, data aggregation, data mapping, and unit conversion.

Optionally, the at least one predictive domain model comprises at least one of: a predictive material model, a predictive ball charge model, a predictive liner wear model, a predictive dynamic charge model, a predictive inferential trajectory model, a predictive new ball addition model and a predictive mill optimizer model.

Optionally, the method further comprises the predictive material model receiving the input data to generate predictive ore hardness data as the output data.

Optionally, the method further comprises the predictive ball charge model receiving the input data to generate predictive ball charge data as output data.

Optionally, the method further comprises the predictive liner wear model receiving the input data to generate predictive liner wear data as output data.

Optionally, the method further comprises the predictive dynamic charge model receiving the input data to generate predictive composition data as output data, wherein the predictive composition data comprises at least one of: a percentage of water, a percentage of solids, and a percentage of rocks in the ore grinding operation.

Optionally, the method further comprises the predictive inferential trajectory model receiving the input data to generate at least one of: predictive toe angle data, predictive shoulder angle, and predictive trajectory data as output data in the ore grinding operation.

Optionally, the method further comprises the predictive new ball addition model receiving the input data to generate ball addition recommendations as output data.

Optionally, the method further comprises employing machine learning techniques for training the at least one predictive domain model to generate the predictive output data simulating the ore grinding operation.

Optionally, the method further comprises storing the predictive output data generated by the at least one predictive domain model.

Optionally, the method further comprises using the stored predictive output data as the input data for the at least one predictive domain model.

Optionally, the at least one process parameter associated with the ore grinding operation comprises at least one of: an ore hardness, a ball charge, a liner wear, a composition of mill charge, a toe angle, a shoulder angle, a trajectory, a ball addition, an ore feed, a process water and a grinding mill speed.

Optionally, the method further comprises generating at least one control variable for controlling the at least one process parameter and providing the at least one control variable to automatic controllers as set points. Optionally, the at least one control variable is provided to the automatic controllers continuously in real-time for at least one of equipment level optimization and/or circuit level optimization.

According to a second aspect, the present disclosure provides a system for continuous optimization of ore grinding operation in mining, wherein the system comprises:

- a database arrangement configured to store at least one predictive domain model;

- a data processing arrangement configured to:

- receive input data corresponding to a set of variables for at least one process parameter associated with the ore grinding operation;

- provide the received input data to at least one predictive domain model wherein the at least one predictive domain model is representative of one or more changes in the at least one process parameter associated with the ore grinding operation;

- generate predictive output data by executing the at least one predictive domain model that simulates the ore grinding operation, wherein the predictive output data is representative of one or more physical changes in the at least one process parameter associated with the ore grinding operation; and

- optimize the grinding ore operation by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data. Optionally, the data processing arrangement is further configured to pre-process the input data before receiving the input data.

Optionally, the data processing arrangement is further configured to employ machine learning techniques for training the at least one predictive domain model to generate the predictive output data simulating the ore grinding operation.

Optionally, the database arrangement is further configured to store the predictive output data generated by the at least one predictive domain model.

Optionally, the data processing arrangement is further configured to use the stored predictive output data as the input data for the at least one predictive domain model.

Optionally, the data processing arrangement is further configured to generate at least one control variable for controlling the at least one process parameter and providing the at least one control variable to automatic controllers as set points.

Optionally, the at least one control variable is provided to the automatic controllers continuously in real-time for at least one of equipment level optimization and/or circuit level optimization.

Embodiments of the present disclosure eliminate the aforementioned drawbacks in existing known approaches for optimizing ore grinding operations. The advantage of the embodiments according to the present disclosure is that the embodiments enable continuous optimization of ore grinding operations in real-time. The present embodiments help to improve key performance indicators of mining operations such as reducing energy consumption in comminution circuit, reducing grinding media waste, increasing grinding efficiency and increasing recovery of minerals. Additional aspects, advantages, features and objects of the present disclosure are made apparent from the drawings and the detailed description of the illustrative embodiments construed in conjunction with the appended claims that follow. It will be appreciated that features of the present disclosure are susceptible to being combined in various combinations without departing from the scope of the present disclosure as defined by the appended claims.

It will be appreciated that the aforesaid present method is not merely a "method of doing a mental act" as such, but has a technical effect in that the method employs measurement data and functions as a form of technical control using machine learning of a technical artificially intelligent system; the method involves building an artificially intelligent machine learning model and/or using the machine learning model to address, for example, to solve, the technical problem of continuously optimizing mining operations, for example, ore grinding operations.

Additional aspects, advantages, features and objects of the present disclosure are made apparent from the drawings and the detailed description of the illustrative embodiments construed in conjunction with the appended claims that follow.

It will be appreciated that features of the present disclosure are susceptible to being combined in various combinations without departing from the scope of the present disclosure as defined by the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

The summary above, as well as the following detailed description of illustrative embodiments, is better understood when read in conjunction with the appended drawings. To illustrate the present disclosure, exemplary constructions of the disclosure are shown in the drawings. However, the present disclosure is not limited to specific methods and instrumentalities disclosed herein. Moreover, those in the art will understand that the drawings are not to scale. Wherever possible, the same elements have been indicated by identical numbers. Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams wherein:

FIG. 1 is a flow diagrams illustrating steps of a method for continuous optimization of ore grinding operation in mining, in accordance with an embodiment of the present disclosure;

FIG. 2 is a schematic illustration of system for continuous optimization of ore grinding operation in mining, in accordance with an embodiment of the present disclosure;

FIG. 3 is a flowchart illustrating a machine learning workflow to predict an operational output of the ore grinding operation of the system of FIG.l in accordance with an embodiment of the present disclosure;

FIG. 4 is a flow-diagram chart illustrating a SAG mill optimizer with at least one domain model in accordance with an embodiment of the present disclosure; and

FIG. 5 is a schematic illustration showing an example result of a virtual densitometer to achieve target in-mill density in accordance with an embodiment of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS

The following detailed description illustrates embodiments of the present disclosure and ways in which they can be implemented. Although some modes of carrying out the present disclosure have been disclosed, those skilled in the art would recognize that other embodiments for carrying out or practicing the present disclosure are also possible. According to a first aspect, the present disclosure provides a method for continuous optimization of ore grinding operation in mining, wherein the method comprises: receiving input data corresponding to a set of variables for at least one process parameter associated with the ore grinding operation; providing the received input data to at least one predictive domain model wherein the at least one predictive domain model is representative of one or more changes in the at least one process parameter associated with the ore grinding operation; generating predictive output data by executing the at least one predictive domain model that simulates the ore grinding operation, wherein the predictive output data is representative of one or more physical changes in the at least one process parameter associated with the ore grinding operation; and optimizing the grinding ore operation by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data. Beneficially, the predictive output data defines a range of operation of the at least one process parameter. Beneficially, the at least one process parameter is dynamically modified in response to the change in the generated predictive output data.

It will be appreciated that in a process of ore grinding operation in mining, a change one or more process parameters result in physical changes of the output ore, for example average ore particle size. A desired physical change such as a change in the average ore particle size is referred throughout this text as an optimized output, r, with respect to the process, an optimization of the output. Furthermore, such physical changes of the output ore, are in turn, a result of changes in input conditions, for example, a change in an amount of energy (or power) input for grinding the ore. Therefore, an advantage of the present invention is that it enables continuous optimization of ore grinding operation in real-time by predicting the physical changes in the at least one process parameter associated with the ore grinding operation; to reduce the number of mill stops for inspections; more precise control over ore grinding operations to improve efficiency; to reduce energy consumption in comminution circuit; to reduce grinding media waste; to increase grinding efficiency. Advantageously, the method employs multiple predictive domain models that are interrelated and interconnected to each other in terms of receiving the input data and providing the predictive output data to each other. However, each of the predictive domain model works independently and does not require the execution of other domain models to generate their respective predictive output data. Furthermore, the present method advantageously employs the predictive approach for controlling the at least one process parameter associated with the ore grinding operation. In other words, each of the predictive domain model acts as a "virtual sensor" allowing the ore grinding operation to be adjusted at once in response to the predictive output data received from each of the predictive domain model, optimizing all the parameters needed at the same time. The present method does not merely act as controller controlling a parameter in response to data from a physical sensor. The present method advantageously analyzes all the physical changes in the at least one process parameter associated with the ore grinding operation by using the predictive output data as input data and provides further optimized predictive output data at a given time. Additionally, the method can be employed in any mining and mineral processing operation to improve efficiency. The method of the present disclosure balances operational and financial considerations by proactively providing control variables required for continuous optimization. According to a second aspect, the present disclosure provides a system for continuous optimization of ore grinding operation in mining. The system comprises a database arrangement and a data processing arrangement. The database arrangement is configured to store at least one predictive domain model. The data processing arrangement configured to: receive input data corresponding to a set of variables for at least one process parameter associated with the ore grinding operation; provide the received input data to at least one predictive domain model wherein the at least one predictive domain model is representative of one or more changes in the at least one process parameter associated with the ore grinding operation; generate predictive output data by executing the at least one predictive domain model that simulates the ore grinding operation, wherein the predictive output data is representative of one or more physical changes in the at least one process parameter associated with the ore grinding operation; and optimize the grinding ore operation by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data.

The method and system of the present disclosure provides an integrated approach utilizing operational data and models of ore grinding operation to provide continuous optimization to enhance performance. The method and system of the present disclosure minimizes production cost, maximizes recovery of desired mineral at optimized grade. Further, the method provides simulation models to predict the key outputs of individual ore grinding operation units such as average particle size in the cyclone overflow, by utilizing both realtime and historical data from various data sources associated with the mineral processing operations. The term "continuous optimization" refers to a strategy or process for improving the efficiency and effectiveness of the ore grinding process, leading to increased productivity, reduced costs, and improved mineral recovery. Moreover, continuous optimization refers to the ongoing process of improving the efficiency and effectiveness of the grinding process over time. Beneficially, continuous optimization involves the use of advanced data analytics and machine learning techniques to identify patterns and correlations in the input data, which is used to further refine and optimize the grinding process. Specifically, in the present invention such advanced data analytics and machine learning techniques are present in the predictive domain models, which effectively functions as process sensors, in this case virtual sensors, which in turn, enable the continuous optimization.

Moreover, the term "optimize" in the present disclosure refers to the modification of the at least one process parameter in response to the generated predictive output data for efficient ore grinding operation in mining.

The data processing arrangement is configured to fetch and execute computer-readable instructions stored in the database arrangement. In an embodiment, the data processing arrangement may be a hardware processor. The data processing arrangement processor may be implemented as microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitries, and any other devices that manipulate signals based on operational instructions. The database arrangement may include any computer- readable medium, for example, volatile memory, such as static random-access memory (SRAM) and dynamic random-access memory (DRAM), non-volatile memory, such as read only memory (ROM), erasable programmable ROM (EPROM), flash drives, hard disks and optical disks. The database arrangement may include cloud-based data storage arrangements including but not limited to Google Cloud, Microsoft Azure, Amazon Web Services, IBM Cloud and so on.

The operational data includes a wide variety of data associated with the ore grinding operation. The operational data may be obtained through sensors that are associated with instrumentation installed for the ore grinding operation. The sensors may also be associated with quality parameters including raw material, intermediate and final product. The sensors also include data corresponding to source and quantity of different raw material processed and data corresponding to equipment failure and maintenance history. Further, the operational data may include historical process data, manufacturing operations management data and ERP (Enterprise Resource Planning) data.

Optionally, variables required to generate the at least one domain model may be determined using unsupervised feature selection techniques such as Principal Component Analysis (PCA), association rule mining or using model-based feature selection techniques such as LASSO, Random Forest (RF), and the like.

In an embodiment, the input data comprises at least one of: historical data, real-time operational data from the ore grinding operation, intermittent data, laboratory information data, test data, design parameters associated with the ore grinding operation. Beneficially, each type of the input data may add respective characteristics to the at least one domain model. Furthermore, more variety of the data sets in the input data may increase the accuracy and optimization of the predictive output data.

In another embodiment, the method further comprises pre-processing of the input data before receiving the input data. Beneficially, the preprocessing of the input data may increase the accuracy of the input data, thus, increasing the chances of obtaining more accurate and optimized predictive output data.

In yet another embodiment, the pre-processing of the input data comprises at least one of: imputation of missing data, data frequency modulation, data aggregation, data mapping, and unit conversion. The term "imputation of missing data" refers to incomplete or missing data that can lead to inaccuracies in predictive models. Beneficially, imputation techniques are used to estimate missing data based on the available data, which can improve the accuracy of the predictive model. Input data may be collected at irregular time intervals, leading to inconsistencies in the data. Beneficially "data frequency modulation" involves resampling the data at a regular time interval to ensure consistency. Input data may be collected from multiple sources. Beneficially, "data aggregation" techniques can be used to combine the data into a more meaningful and consistent format. Input data may be in different formats or from different sources. Beneficially, "Data mapping" involves transforming the data into a common format that can be used for analysis and modeling. Input data may be collected in different units, which can lead to errors in predictive models. "Unit conversion" involves converting the data into a common unit of measurement to ensure consistency and accuracy.

In yet another embodiment, the at least one predictive domain model comprises at least one of: a predictive material model, a predictive ball charge model, a predictive liner wear model, a predictive dynamic charge model, a predictive inferential trajectory model, a predictive new ball addition model and a predictive mill optimizer model. Beneficially, each of the predictive domain model generates predictive output data corresponding to at least one operating physical changes in the at least one process parameter associated with the ore grinding operation.

In an embodiment, the method further comprises storing the predictive output data generated by the at least one predictive domain model. Optionally, the predictive output data is stored in the database arrangement.

In yet another embodiment, the method further comprises using the stored predictive output data as the input data for the at least one predictive domain model. Beneficially, the use of predictive output data as input data for the at least one predictive domain model enables continuous optimization of the at least one process parameter associated with the ore grinding operation. A time period at which the stored predictive output data is used as the input data for the at least one predictive domain model, the predictive model to be employed may be decided according to dynamics of a specific circuit and the availability of operational data. For example, predictive liner wear data may be used by at least one of: the predictive ball charge model, the predictive dynamic charge model and the predictive inferential trajectory model as the input data to generate their respective predictive output data. Therefore, depending on the dynamics of a specific circuit and the availability of operational data, the predictive ball charge model may be used, and/or predictive dynamic charge model and/or the predictive inferential trajectory model.

In an embodiment, the at least one process parameter associated with the ore grinding operation comprises at least one of: an ore hardness, a ball charge, a liner wear, a composition of mill charge, a toe angle, a shoulder angle, a trajectory, a ball addition, an ore feed, a process water and a grinding mill speed. In yet another embodiment, the predictive material model receives the input data to generate predictive ore hardness data as the output data. Beneficially, the predictive ore hardness data may be used by the at least one of: the predictive ball charge model, the predictive liner wear model and the predictive dynamic charge model as the input data to generate their respective predictive output data.

In yet another embodiment, the predictive ball charge model receives the input data to generate predictive ball charge data as output data. Beneficially, the predictive ball charge data may be used by the at least one of: the predictive dynamic charge model, the predictive inferential trajectory model, the predictive new ball addition model and the predictive mill optimizer model as the input data to generate their respective predictive output data.

The predictive ball charge model helps to estimate ball charge (Jb) in real-time, which allows for more precise ball charge control. The realtime ball charge estimates help to reduce mill stops for ball charge inspections. The real-time ball charge estimates help in modeling other grinding parameters. The predictive ball charge model tracks the number and weight of balls in the mill. It models the ball wear as a linear and uniform decrease in diameter (i.e.) purely abrasive wear. The wear rate is a constant, mm per minute. It is calculated from the manual measurements and historical ball consumption rate as detailed by LA. VERMEULEN and D.D. HOWAT et al. (1986). The predictive ball charge model uses the operational data including mill power, mill weight, and physical ball parameters including weight of the ball and size of the ball and discharge grate parameters to estimate the Ball charge level (Jb) and the Ball mass.

In an embodiment, the at least one domain model includes a ball wear model. Ball wear will be variable, and it is estimated considering ore hardness, ball hardness, mill operating regime, and updated with Machine Learning extension. The ball wear model allows for the prediction of the wear and consumption of grinding balls, which can have a significant impact on the overall cost and efficiency of the grinding process. By accurately modeling the ball wear and consumption, the grinding process can be optimized to achieve maximum throughput and minimize operating costs.

In an embodiment, combining models for power and mill weight to estimate ball charge uses data fusion algorithms for more accurate ball charge estimate. The term "data fusion algorithm" refers to the joint analysis of multiple inter-related datasets in order to gain insights and make informed decisions about the mining process. The process of correlating and fusing information from multiple sources generally allows more accurate inferences than those that the analysis of a single dataset can yield. The significance of data fusion algorithms in ore mining lies in their ability to provide a comprehensive view of the mining process by combining data from various sources, such as multiple domain models, sensor data and operational data. Data fusion algorithms can help mining companies improve safety, increase productivity, and reduce costs by improving accuracy, real-time monitoring, predictive maintenance, and optimization. Ball charge may be estimated with a combination of mill weight data and mill motor power data. Data fusion is employed to combine the ball charge estimates taken from the three sources: Ball charge wear model, weight and Mill power. Methods of nonlinear Bayesian filtering are employed, among others, in order to accurately estimate uncertainties of each data source and provide a combined estimate of the ball charge in real time. Using Bayesian filtering, it is possible to update the estimated ball charge parameters in real-time, which allows for more precise control of the grinding process. In yet another embodiment, the at least one domain model includes a predictive ball breakage model. The Ball breakage is modelled as a Bernoulli trial with a constant probability of a ball to break. The predictive ball breakage model updates its state every minute including the numbers of balls of each batch, estimated number of broken and ejected balls. The Bernoulli trial allows for the calculation of the rate of breakage of particles as a function of the operating conditions of the mill and the properties of the ore. By accurately modeling the probability of ball breaking a particle, the ball breakage model optimizes the performance of grinding mills and improves the efficiency of the ore grinding process.

In yet another embodiment, the at least one domain model includes a predictive new ball ejection model. Ball ejection through discharge grate is modelled as a probability of the ball of a size less than a certain aperture of the discharge grate to leave the mill. The ejection probability equation is calculated as detailed by T.A.APELT and N. F. THORNHILL et al. (2009). The predictive new ball ejection model is updated to have multiple aperture sizes and fractional opening areas in the discharge rate. Ball ejections will be modelled more precisely with a new proprietary physical model of ejection.

In yet another embodiment, the predictive liner wear model receives the input data to generate predictive liner wear data as output data. The liners in a grinding mill are used to protect the shell of the mill and to improve the efficiency of the grinding process by providing a suitable surface for the grinding balls to impact and grind against. However, due to the high-energy impact and abrasive nature of the grinding process, the liners are subject to wear over time, which can lead to reduced grinding efficiency, increased operating costs. The significance of the liner wear model in ore grinding is that it allows for the optimization of the liner design and replacement schedule to achieve maximum grinding efficiency and minimize operating costs. By accurately modeling the wear of the liners, the grinding process can be optimized to achieve optimal performance. Beneficially, the predictive liner wear data may be used by the at least one of: the predictive ball charge model, the predictive dynamic charge model and the predictive inferential trajectory model as the input data to generate their respective predictive output data.

In yet another embodiment, the predictive dynamic charge model receives the input data to generate predictive composition data as output data, wherein the predictive composition data comprises at least one of: a percentage of water, a percentage of solids, and a percentage of rocks in the ore grinding operation. Beneficially, the predictive composition data may be used by the at least one of: the predictive inferential trajectory model and the predictive mill optimizer model as the input data to generate their respective predictive output data.

In yet another embodiment, the predictive inferential trajectory model receives the input data to generate at least one of: predictive toe angle data, predictive shoulder angle, and predictive trajectory data as output data in the ore grinding operation. Beneficially, the predictive output data of the predictive inferential trajectory model may be used by the at least one of: the predictive liner wear model and the predictive mill optimizer model as the input data to generate their respective predictive output data. The inferential trajectory model may define, at a given time period, or at an approximate instant, a prediction of particle size distribution, for example, defining out-of- specification coarse ore and in-specification fine ore fractions. In turn, out-of-specification coarse ore and in-specification fine ore fractions refer, respectively, to ore particles that too large to be discharged from the mill or particles that are small enough to leave the mill. The output of the predictive inferential trajectory model therefore may be used an input to other predictive models which will simulate conditions of the overall process and define changes of other process parameters for process optimization. In particular, such changes of process parameters may be applied automatically applied to the process in realtime by the automatic controllers.

In yet another embodiment, the predictive new ball addition model receives the input data to generate ball addition recommendations as output data. The addition of new grinding balls to the mill prevents the balls from becoming too small to grind the ore with the same efficiency. In this case, for a given fixed number of grinding balls, a smaller griding ball will have less surface contact with the ore than a larger ball, and thus, the amount of material ground at a given time decreases, in turn decreasing the grinding efficiency. Therefore, the addition of new grinding balls to the mill is important for maintaining the grinding efficiency. However, adding too many balls can cause overloading of the mill and reduce the grinding efficiency, while adding too few balls can cause the existing balls to wear down faster and increase the operating costs. Therefore, the new ball addition model is significant because it allows for the optimization of the ball addition rate and the ball size distribution to achieve maximum grinding efficiency and minimize the wear of the grinding balls and the overall operating costs.

In yet another embodiment, the predictive mill optimizer model receives input data to generate at least one of: a predictive ball addition, a predictive ore feed, a predictive process water and a predictive grinding mill speed data as output data. Beneficially, the output data generated by the predictive mill optimizer model may be used to generate at least one control variable for controlling the at least one process parameter. The term "control variable" refers to a variable that is used to control one or more process parameters in the ore grinding process. By controlling the process parameters in real time based on the output data generated by the predictive mill optimizer model, the ore grinding process can be optimized to improve efficiency, reduce energy consumption, and improve product quality. Optionally, a SAG mill optimizer provides control variables for continuous optimization in real-time based on the at least one domain model. The control variables may be provided with respect to mill volume, ore feed, process water and mill speed for optimization.

In yet another embodiment, the at least one domain model includes a predictive material transport and influence model. The predictive material transport & influence model tracks material from geological models through transportation and blending. Financial metrics are correlated to determine optimal particle size and distribution for the recovery.

In an embodiment, the method further comprises employing machine learning techniques for training the at least one predictive domain model to generate the predictive output data simulating the ore grinding operation. Beneficially, the supervised or unsupervised machine learning techniques may be employed for training the at least one predictive domain model. The trained predictive model is used to generate predictive output data that can be used to simulate the ore grinding operation and optimize process parameters in real-time. By leveraging machine learning algorithms, it is possible to gain insights, leading to better decision-making and improved outcomes for ore griding operations.

The at least one domain model are machine learning models. The system utilizes virtual sensors developed using modeling and simulation that can provide real-time prediction of important process parameters of the mining and mineral processing operations, which are either infrequently or not measured at all. Thus, virtual sensors may be developed for the process parameters such as specific fracture energy, particle size distribution, slurry density of the intermediate stream, solids hold up in the ore grinding mills and the like.

The at least one domain model includes a dynamic overload threshold model. Dynamic overload threshold model utilizes a physics-informed model of the relationship between tumbling mill's weight and power drawn by the motor and also dynamically updates this model using machine learning and Bayesian approaches. Thus, a virtual sensor may be developed for the mill weight or bearing pressure at which an overload event may occur.

In an embodiment, the method further comprises generating at least one control variable for controlling the at least one process parameter and providing the at least one control variable to automatic controllers as set points.

In another embodiment, the at least one control variable is provided to the automatic controllers continuously in real-time for at least one of equipment level optimization and/or circuit level optimization.

Optionally, the set points include three levels of set points comprising at least one of: circuit level set points, equipment level set points, and equipment level alerts and recommendations. The circuit level set points may include balance of energy between mills and circulation of loads.

The control variables provide equipment level set points that may include Bearing pressure set points, cyclone pressure set points, percentage of mill feed solids, percentage of cyclone feed solids, pressure (P80) and product Particle Size Distribution. The control variables provide circuit level set points that may include balance of energy between mills, circulating load. The circuit level optimization balances throughput and grinding duty based on ore characteristics and downstream requirements. The optimization includes equipment level alerts and recommendations including ball addition recommendations, liner replacement recommendations and overload alerts. The automatic controllers are configured to receive the at least one control variable as set points for controlling the least one process parameter.

The at least one predictive domain model is deployed such that the complete set of outputs can be generated at time intervals suited for real time decision making. Such time interval, or the at least one predictive domain model execution time interval, can be selected and adjusted based on the dynamics of the processes and requirements. Beneficially, the at least one predictive domain model deployment may be achieved by using web and cloud services or local on-site deployment, and containerized deployment technology. Optionally, the predictive output data processing outputs are provided in sync with the selected model execution time interval.

Beneficially, the complete set of predictive output data is stored in the database arrangement.

The present disclosure also relates to the system as described above. Various embodiments and variants disclosed above apply mutatis mutandis to the system.

Optionally, the data processing arrangement is further configured to pre-process the input data before receiving the input data.

Optionally, the data processing arrangement is further configured to employ machine learning techniques for training the at least one predictive domain model to generate the predictive output data simulating the ore grinding operation. Optionally, the database arrangement is further configured to store the predictive output data generated by the at least one predictive domain model.

The system provides a digital twin of an entire chain of ore grinding operation by modelling and simulation in real-time. The digital twin may be a primary grinding digital twin, secondary grinding digital twin, cyclone digital twin.

Dual and parallel optimization may be performed for critical processes associated with the ore grinding operation. The Dual and parallel optimization may be performed based on a method for learning to refine a rule-based fuzzy logic controller. A reinforcement learning technique is used in conjunction with a multilayer neural network model of a fuzzy controller.

DETAILED DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic illustration of a method 100 for continuous optimization of ore grinding operation in mining. At step 102, input data corresponding to a set of variables for at least one process parameter associated with the ore grinding operation is received. At step 104, the received input data is provided to at least one predictive domain model wherein the at least one predictive domain model is representative of one or more changes in the at least one process parameter associated with the ore grinding operation. At step 106, predictive output data is generated by executing the at least one predictive domain model that simulates the ore grinding operation, wherein the predictive output data is representative of one or more physical changes in the at least one process parameter associated with the ore grinding operation. At step 108, the grinding ore operation is optimized by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data.

FIG. 2 is a schematic illustration of system 200 for continuous optimization of ore grinding operation in mining in accordance with an embodiment of the present disclosure. The system 200 comprises a database arrangement 202 and a data processing arrangement 204. The database arrangement 202 is configured to store at least one predictive domain model. The data processing arrangement 204 configured to receive input data corresponding to a set of variables for at least one process parameter associated with the ore grinding operation; provide the received input data to at least one predictive domain model wherein the at least one predictive domain model is representative of one or more changes in the at least one process parameter associated with the ore grinding operation; generate predictive output data by executing the at least one predictive domain model that simulates the ore grinding operation, wherein the predictive output data is representative of one or more physical changes in the at least one process parameter associated with the ore grinding operation; and optimize the grinding ore operation by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data.

FIG. 3 is a flowchart illustrating a machine learning workflow to predict an operational output of an ore grinding operation in accordance with an embodiment of the present disclosure. At step 302 of the machine learning workflow, at least one input variable that includes raw operational data, intermittent operational data, lab operational data, test operational data and design parameters, are obtained. The at least one input variable is obtained through the REST API. At step 304 of the machine learning workflow, at least one domain model generated for the at least one process parameter associated with the ore grinding operation is obtained. The process parameters may include at least one of ball charge, liner wear, trajectory, residence time, cyclone underflow, density, particle size distribution, lithology and flow rates. At step 306 of the machine learning workflow, a model factory is configured to produce training models by performing at least one of visualization, tuning and comparing with the at least one input variables and at least one model generated for the at least one process parameter. At step 308 of the machine learning workflow, the trained models are stored in a local server. The trained models may be stored in a cloud-based server. At step 310 of the machine learning workflow, the trained models are accessed from the model service layer of the local or cloud-based server through the RESTAPI or Flink/Spark.

FIG. 4 is a flow-diagram illustrating a SAG mill optimizer 400 with at least one domain model in accordance with an embodiment of the present disclosure. The SAG mill optimizer 400 includes at least one domain model including a material model 402, a ball charge model 404, a linear wear model 406, a dynamic charge model 408, an inferential trajectory model 410 and a ball addition recommender 412. The at least one domain model provides a framework for best estimations of the SAG mill performance. The at least one domain model provides real-time insights into the operational characteristics of the SAG mill and provides visibility on unmeasured stream properties. The material model 402, for example, estimates of the ore hardness. The Ball charge model 404, for example, estimates ball charge level which is a function of the bulk fraction of the SAG mill volume (Jb) occupied by balls. The Linear Wear Model 406, for example, estimates mill volume and linear state. The dynamic charge model 408, for example, estimates mill charge including percentage of rock, solids and water. The inferential trajectory model 410, for example, estimates toe angle, shoulder angle and trajectory. The ball addition recommender 412 provides recommendations for ball additions. The SAG mill optimizer 400 provides control variables for continuous optimization in real-time based on the at least one domain model. The control variables may be provided with respect to mill volume, ore feed, process water and mill speed for optimization. Combining models for power and mill weight may be employed to estimate accurate ball charge using data fusion algorithms.

FIG. 5 is a schematic illustration showing an example result of a virtual densitometer to achieve target in-mill density in accordance with an embodiment of the present disclosure; The graphical illustration elucidates, determining the amount of process water to be added to achieve target in-mill density based on the inferred secondary ball mill feed density by the virtual densitometer. The virtual densitometer is a combined output result of all the predictive domain model. The virtual densitometer effectively acts as a virtual sensor which drives the optimization of the ore grinding operation.

Modifications to embodiments of the present disclosure described in the foregoing are possible without departing from the scope of the present disclosure as defined by the accompanying claims. Expressions such as "including", "comprising", "incorporating", "have", "is" used to describe and claim the present disclosure are intended to be construed in a non-exclusive manner, namely allowing for items, components or elements not explicitly described also to be present. Reference to the singular is also to be construed to relate to the plural.

Claims

1. A method for continuous optimization of ore grinding operation in mining, wherein the method comprises:

2. The method according to claim 1, wherein the input data comprises at least one of: historical data, real-time operational data from the ore grinding operation, intermittent data, laboratory information data, test data, design parameters associated with the ore grinding operation.

3. The method according to claims 1 or 2, wherein the method further comprises pre-processing of the input data before receiving the input data.

4. The method according to claim 3, wherein the pre-processing of the input data comprises at least one of: imputation of missing data, data frequency modulation, data aggregation, data mapping, and unit conversion.

5. The method according to claim 1, wherein the at least one predictive domain model comprises at least one of: a predictive material model, a predictive ball charge model, a predictive liner wear model, a predictive dynamic charge model, a predictive inferential trajectory model, a predictive new ball addition model and a predictive mill optimizer model.

6. The method according to claim 5, wherein the method further comprises the predictive material model receiving the input data to generate predictive ore hardness data as the output data.

7. The method according to claim 5, wherein the method further comprises the predictive ball charge model receiving the input data to generate predictive ball charge data as output data.

8. The method according to claim 5, wherein the method further comprises the predictive liner wear model receiving the input data to generate predictive liner wear data as output data.

9. The method according to claim 5, wherein the method further comprises the predictive dynamic charge model receiving the input data to generate predictive composition data as output data, wherein the predictive composition data comprises at least one of: a percentage of water, a percentage of solids, and a percentage of rocks in the ore grinding operation.

10. The method according to claim 5, wherein the method further comprises the predictive inferential trajectory model receiving the input data to generate at least one of: predictive toe angle data, predictive shoulder angle, and predictive trajectory data as output data in the ore grinding operation.

11. The method according to claim 5, wherein the method further comprises the predictive new ball addition model receiving the input data to generate ball addition recommendations as output data.

12. The method according to any of the preceding claims, wherein the method further comprises employing machine learning techniques for training the at least one predictive domain model to generate the predictive output data simulating the ore grinding operation.

13. The method according to any of the preceding claims, wherein the method further comprises storing the predictive output data generated by the at least one predictive domain model.

14. The method according to claim 13, wherein the method further comprises using the predictive output data stored as the input data for the at least one predictive domain model.

15. The method according to any of the preceding claims, wherein the at least one process parameter associated with the ore grinding operation comprises at least one of: an ore hardness, a ball charge, a liner wear, a composition of mill charge, a toe angle, a shoulder angle, a trajectory, a ball addition, an ore feed, a process water and a grinding mill speed.

16. The method according to any of the preceding claims, wherein the method further comprises generating at least one control variable for controlling the at least one process parameter and providing the at least one control variable to automatic controllers as set points.

17. The method according to claim 16, wherein the at least one control variable is provided to the automatic controllers continuously in real-time for at least one of equipment level optimization and/or circuit level optimization.

18. A system for continuous optimization of ore grinding operation in mining, wherein the system comprises:

- a data processing arrangement configured to:

- optimize the grinding ore operation by controlling the at least one process parameter associated with the ore grinding operation, wherein the at least one process parameter is modified in response to the generated predictive output data.

19. The system according to claim 18, wherein the data processing arrangement is further configured to pre-process the input data before receiving the input data.

20. The system according to claims 18 or 19, wherein the data processing arrangement is further configured to employ machine learning techniques for training the at least one predictive domain model to generate the predictive output data simulating the ore grinding operation.

21. The system according to claims 18-20, wherein the database arrangement is further configured to store the predictive output data generated by the at least one predictive domain model.

22. The system according to claim 21, wherein the data processing arrangement is further configured to use the predictive output data stored as the input data for the at least one predictive domain model.

23. The system according to claims 18-22, wherein the data processing arrangement is further configured to generate at least one control variable for controlling the at least one process parameter and providing the at least one control variable to automatic controllers as set points.

24. The system according to claim 23, wherein the at least one control variable is provided to the automatic controllers continuously in real-time for at least one of equipment level optimization and/or circuit level optimization.