Estimation device, learning model, and estimation method

Abstract

This invention provides a highly versatile estimation device, learning model, and estimation method. [Solution] The estimation device 1 is a device for estimating the results of flotation, which separates particles containing the substance to be separated from a plurality of particles obtained from mined ore by attaching them to bubbles and allowing them to float. The estimation device 1 includes an acquisition unit 11 that acquires a plurality of hydrophobicity information indicating a first hydrophobicity, a plurality of first proportion information, a plurality of second proportion information, and a plurality of particle information; a generation unit 12 that generates a plurality of probability information by inputting the hydrophobicity information and particle information into a learning model M and performing an estimation process for each of the plurality of particles to estimate probability information indicating the probability that a predetermined particle will float in flotation; and a calculation unit 13 that calculates at least one of the recovery rate of the substance to be separated and the grade of the substance to be separated based on the plurality of first proportion information, a plurality of second proportion information, and a plurality of probability information.

Description

[Technical Field]

[0001] One aspect of the embodiment relates to an estimation device, a learning model, and an estimation method. [Background technology]

[0002] Patent Document 1 describes a recovery rate prediction device for predicting the recovery rate of a metal to be sorted in flotation, which separates a metal to be sorted from an ore. The recovery rate prediction device of Patent Document 1 calculates the recovery rate of the metal to be sorted using a first-order rate equation, the flotation velocity equation, based on the soluble metal ratio, which is the proportion of soluble metals in the ore, and information showing the relationship between the soluble metal ratio and the mineral content. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2024-13650 [Overview of the project] [Problems that the invention aims to solve]

[0004] In estimation devices like the one described above, when accurately calculating the recovery rate of the material to be sorted using a first-order rate equation, the parameters used in the calculation (for example, information showing the relationship between the soluble metal ratio and the mineral content) are determined experimentally. In this case, for example, if the type of ore used in flotation is changed, it is necessary to conduct experiments using the new ore and re-determine the above parameters. Such estimation devices lack versatility.

[0005] Therefore, one aspect of the embodiment was made in view of the above issues, and aims to provide a highly versatile estimation device, learning model, and estimation method. [Means for solving the problem]

[0006] The estimation device according to the first aspect of the embodiment is an estimation device for estimating at least one of the recovery rate of the substance to be sorted and the grade of the substance to be sorted in flotation ore, in which particles containing the substance to be sorted are sorted from a plurality of particles obtained from mined ore by attaching them to bubbles and letting them float, and comprises a plurality of hydrophobicity information that each indicates a first hydrophobicity which is the degree of low affinity to water for the plurality of particles, a plurality of first proportion information that each indicates the proportion of the mass of one particle among the plurality of particles to the total mass of the plurality of particles, and a plurality of second proportion information that each indicates the proportion of the mass of the substance to be sorted to the mass of one particle among the plurality of particles The system includes: an acquisition unit that acquires information and multiple particle information, each representing a feature quantity for multiple particles; a generation unit that generates multiple probability information by inputting the hydrophobicity information and particle information acquired by the acquisition unit into a learning model and performing an estimation process for each of the multiple particles to estimate probability information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in flotation; and a calculation unit that calculates at least one of the recovery rate of the material to be sorted and the grade of the material to be sorted based on the multiple first proportion information and multiple second proportion information acquired by the acquisition unit and the multiple probability information generated by the generation unit.

[0007] The estimation method relating to the second aspect of the embodiment is an estimation method for estimating at least one of the recovery rate of the substance to be sorted and the grade of the substance to be sorted in flotation, in which particles containing the substance to be sorted are sorted from a plurality of particles obtained from mined ore by attaching them to bubbles and allowing them to float, comprising: a plurality of hydrophobicity information indicating the degree of low affinity to water for each of the plurality of particles; a plurality of first proportion information indicating the proportion of the mass of one of the plurality of particles to the total mass of the plurality of particles; a plurality of second proportion information indicating the proportion of the mass of the substance to be sorted to the mass of one of the plurality of particles to the mass of one of the plurality of particles; and a plurality of particles The method comprises the steps of: acquiring multiple particle information that each represents a feature for each particle; inputting the hydrophobicity information and particle information acquired in the acquisition step into a learning model to perform an estimation process for each of the multiple particles to estimate probability information that indicates the probability that the particles corresponding to the hydrophobicity information and particle information will float in flotation, thereby generating multiple probability information; and calculating at least one of the recovery rate of the material to be sorted and the grade of the material to be sorted based on the multiple first proportion information and multiple second proportion information acquired in the acquisition step, and the multiple probability information generated in the generation step.

[0008] According to the above first aspect or the above second aspect, by inputting a plurality of hydrophobicity information items each indicating a first hydrophobicity degree, which is a degree of low affinity for water, for a plurality of particles, and a plurality of particle information items each indicating a feature amount for a plurality of particles into a learning model, probability information indicating the probability that the particles corresponding to the hydrophobicity information and the particle information float in flotation separation is estimated. Here, according to the above configuration, since the hydrophobicity information corresponding to the type of ore used in flotation separation is input into the learning model, even if the type of ore used in flotation separation is changed, the probability information can be accurately estimated without reconstructing the learning model. Therefore, based on the plurality of first ratio information items, the plurality of second ratio information items, and the plurality of probability information items, at least one of the recovery rate of the substance to be separated and the grade of the substance to be separated can be accurately calculated. Thereby, an estimation device with high versatility can be provided.

[0009] The learning model according to the third aspect of the embodiment is a learning model used for flotation separation in which particles containing a substance to be separated among a plurality of particles obtained from mined ore are attached to bubbles and floated for separation. The learning model takes, as input, hydrophobicity information indicating a first hydrophobicity degree, which is a degree of low affinity for water, for the particles obtained from the mined ore, and particle information indicating a feature amount for the particles, and outputs probability information indicating the probability that the particles float in flotation separation.

[0010] According to the above third aspect, a plurality of hydrophobicity information items each indicating a first hydrophobicity degree, which is a degree of low affinity for water, for a plurality of particles, and a plurality of particle information items each indicating a feature amount for a plurality of particles are used as input, and probability information indicating the probability that the particles corresponding to the hydrophobicity information and the particle information float in flotation separation is used as output. Thereby, as described above, the probability information can be accurately estimated without reconstructing the learning model.

[0011] Also, in one aspect, one of the plurality of particles contains a plurality of minerals, and the first hydrophobicity corresponding to the one particle is the sum of the products of each of a plurality of second hydrophobicities, which are degrees of low affinity of each of the plurality of minerals for water, and each of a plurality of surface exposure ratios, which are the ratios occupied by each of the plurality of minerals on the surface of the particle, which is preferable. Here, conventionally, when chemical conditions were changed in flotation, it was necessary to reconstruct the learning model. However, according to the above configuration, since the change in chemical conditions in flotation is reflected in the first hydrophobicity using the second hydrophobicity, when the chemical conditions are changed in flotation, probability information can be accurately estimated without reconstructing the learning model. Further, since the first hydrophobicity can be calculated more accurately based on both the plurality of second hydrophobicities and the plurality of surface exposure ratios, probability information can be calculated more accurately.

[0012] Also, in one aspect, the second hydrophobicity corresponding to one of the plurality of minerals is a numerical value corresponding to the ratio of the number of first particles that float by flotation to the total number of first particles used in flotation when first particles, which are particles with a surface exposure ratio of 95% or more of the one mineral, are used in flotation, which is preferable. In this case, the first hydrophobicity can be calculated more accurately, so probability information can be calculated more accurately.

[0013] Also, in one aspect, the plurality of surface exposure ratios are preferably calculated by mineral particle analysis. In this case, the surface exposure ratio of each of the plurality of minerals can be easily and accurately calculated.

[0014] Also, in one aspect, the particle information preferably has at least one of size information indicating the size of the particle, specific gravity information indicating the specific gravity of the particle, and shape information indicating the shape of the particle. In this case, probability information can be easily and more accurately estimated based on the hydrophobicity information and the particle information.

[0015] The estimation device of the embodiment is [1] "An estimation device for estimating at least one of the recovery rate of the substance to be sorted and the grade of the substance to be sorted in flotation, in which particles containing the substance to be sorted are sorted by attaching them to bubbles and causing them to float, among a plurality of particles obtained from mined ore, An acquisition unit that acquires multiple hydrophobicity information indicating the degree of low affinity for water for each of the multiple particles, multiple first proportion information indicating the proportion of the mass of one of the multiple particles to the total mass of the multiple particles, multiple second proportion information indicating the proportion of the mass of the substance to be sorted to the mass of one of the multiple particles, and multiple particle information indicating characteristic quantities for each of the multiple particles. A generation unit generates multiple sets of probability information by inputting the hydrophobicity information and particle information acquired by the acquisition unit into a learning model, and performing an estimation process for each of the multiple particles to estimate probability information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in the flotation process. The estimation device comprises a calculation unit that calculates at least one of the recovery rate of the substance to be sorted and the quality of the substance to be sorted, based on the plurality of first proportion information and the plurality of second proportion information acquired by the acquisition unit, and the plurality of probability information generated by the generation unit.

[0016] The estimation apparatus of the embodiment states that [2] "one of the plurality of particles contains a plurality of minerals, The estimation apparatus described in [1] may be the sum of the products of a plurality of second hydrophobicities, which are the degree of low affinity of each of the plurality of minerals to water, and a plurality of surface exposure ratios, which are the proportions that each of the plurality of minerals occupies on the surface of the particle.

[0017] The estimation apparatus of the embodiment may be [3] "the estimation apparatus described in [2], wherein the second hydrophobicity corresponding to one of the plurality of minerals is a numerical value corresponding to the ratio of the number of first particles suspended by flotation to the total number of first particles used in flotation when first particles, which are particles in which the surface exposure ratio of the one mineral is 95% or more, are used in the flotation."

[0018] The estimation apparatus of the embodiment may be [4] "the estimation apparatus described in [2] or [3] in which the plurality of surface exposure ratios are calculated by mineral particle analysis."

[0019] The estimation apparatus of the embodiment may be [5] "the estimation apparatus according to any one of [1] to [4], wherein the particle information has at least one of size information indicating the size of the particle, specific gravity information indicating the specific gravity of the particle, and shape information indicating the shape of the particle."

[0020] The learning model of the embodiment is [6] "a learning model used for flotation, which separates particles containing the substance to be separated from a plurality of particles obtained from mined ore by attaching them to bubbles and causing them to float, This may also be a learning model that takes hydrophobicity information, which indicates the degree of low affinity for water in particles obtained from mined ore, and particle information, which indicates characteristic quantities of the particles, as input, and outputs probability information, which indicates the probability that the particles will float in flotation.

[0021] The estimation method of the embodiment is described as follows: [7] "In flotation ore separation, in which particles containing a substance to be separated are separated from a plurality of particles obtained from mined ore by attaching them to bubbles and causing them to float, the estimation method is for estimating at least one of the recovery rate of the substance to be separated and the grade of the substance to be separated, The steps include obtaining a plurality of hydrophobicity information, each representing a first hydrophobicity, which is the degree of low affinity for water for the plurality of particles; a plurality of first proportion information, each representing the proportion of the mass of one of the plurality of particles to the total mass of the plurality of particles; a plurality of second proportion information, each representing the proportion of the mass of the substance to be sorted to the mass of one of the plurality of particles; and a plurality of particle information, each representing a characteristic quantity for the plurality of particles. The steps include: inputting the hydrophobicity information and particle information acquired in the acquisition step into a learning model, and performing an estimation process for each of the plurality of particles to estimate probability information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in the flotation process, thereby generating a plurality of the aforementioned probability information; The estimation method may also include a calculation step of calculating at least one of the recovery rate of the substance to be sorted and the quality of the substance to be sorted based on the plurality of first proportion information and the plurality of second proportion information obtained in the acquisition step, and the plurality of probability information generated in the generation step. [Effects of the Invention]

[0022] According to one aspect of the present invention, it is possible to provide a highly versatile estimation device, learning model, and estimation method. [Brief explanation of the drawing]

[0023] [Figure 1] Figure 1 is a block diagram showing the functional configuration of the estimation device 1 according to an embodiment. [Figure 2] Figure 2 is a block diagram showing an example of the hardware configuration of the estimation device 1 shown in Figure 1. [Figure 3] Figure 3 is a schematic diagram showing an example of one of several particles obtained from ore. [Figure 4] Figure 4 shows an example of a learning model M according to the embodiment. [Figure 5]Figure 5 is a flowchart showing the procedure of the estimation method according to the embodiment. [Figure 6] Figure 6 is a table illustrating the effects of the estimation device 1 according to this embodiment. [Figure 7] Figures 7(a), 7(b), and 7(c) are graphs illustrating the effects of the estimation device 1 according to the embodiment. [Modes for carrying out the invention]

[0024] Embodiments of the present invention will be described in detail below with reference to the attached drawings. In this description, the same reference numerals will be used for the same element or element having the same function, and redundant explanations will be omitted.

[0025] Figure 1 is a schematic diagram of the estimation device 1 according to an embodiment. The estimation device 1 is a device for estimating the sorting results in flotation, which sorts particles containing a target substance from a plurality of particles obtained from ore. The particles are obtained, for example, by crushing the ore. The particles contain at least one type of mineral. The target substance is an element or a mineral. The target element is, for example, copper, lead, and zinc. In the embodiments described below, the sorting of particles containing the target element will be described in flotation.

[0026] Flotation is a method of physically separating particles containing a target element from a group of particles obtained by crushing ore or minerals. For example, in flotation, a feed containing multiple particles is suspended in water to obtain a suspension, and then bubbles are introduced into the suspension. This allows for the separation of floss products (hereinafter referred to as floss) by causing water-repellent particles (hydrophobic particles) to adhere to the bubbles and float, while tail products (hereinafter referred to as tail) are separated by allowing water-attracting particles (hydrophilic particles) to remain in the suspension without adhering to the water. Here, by making the surface of particles containing the target element hydrophobic through chemical treatment using agents such as surfactants, it becomes possible for these particles to adhere to bubbles during flotation and be separated as floss.

[0027] Estimation device 1 estimates the recovery rate and grade of the elements to be separated in the flotation process as a result of the flotation. The recovery rate of the elements to be separated is the value obtained by dividing the mass of the elements to be separated recovered in the floss by the mass of the elements to be separated contained in the floss and tail. The grade of the elements to be separated is the value obtained by dividing the mass of the elements to be separated contained in the floss by the mass of all particles contained in the floss. The recovery rate and grade of the elements to be separated will be described in detail later.

[0028] Figure 2 is a block diagram showing an example of the hardware configuration of the estimation device 1 in Figure 1. As shown in Figure 2, the estimation device 1 is physically a computer including a processor (CPU: Central Processing Unit) 101A and a GPU (Graphics Processing Unit) 101B, a recording medium (RAM: Random Access Memory) 102 or ROM (Read Only Memory) 103, a communication module 104, and an input / output module 106, etc., each of which is electrically connected. The estimation device 1 may also include input / output devices such as a display, keyboard, mouse, touch panel display, etc., or data recording devices such as a hard disk drive and semiconductor memory. Furthermore, the estimation device 1 may be composed of multiple computers.

[0029] As shown in Figure 1, the estimation device 1 is configured to include, as functional components, an acquisition unit 11, a generation unit 12, a calculation unit 13, and a construction unit 14.

[0030] The acquisition unit 11 acquires multiple first proportion information, multiple second proportion information, and multiple particle information. Multiple first proportion information is information that shows the proportion of the mass of one particle among multiple particles to the total mass of multiple particles. For example, the first proportion information shows the mass of a predetermined particle when the total mass of multiple particles is set to 100. Multiple second proportion information is information that shows the proportion of the mass of the element to be sorted to the mass of one particle among multiple particles. For example, the second proportion information is the proportion of the mass of the element to be sorted to the mass of a predetermined particle. Multiple particle information is information that shows the characteristic quantities for each of the multiple particles. The particle information includes size information that shows the size of a predetermined particle, specific gravity information that shows the specific gravity of the particle, and shape information that shows the shape of the particle. For example, the information that shows the size of a particle is the diameter of a circle having the same area as the particle, and the information that shows the shape of a particle is a value obtained by dividing the circumference of the circle by the circumference of the particle.

[0031] The acquisition unit 11 further acquires multiple hydrophobicity information. The multiple hydrophobicity information is information indicating the degree of low affinity for water for each of the multiple particles. The first hydrophobicity is a numerical value that relatively indicates the degree of low affinity for water for each particle among the multiple particles. For example, the first hydrophobicity may be a numerical value between 0 and 1, a numerical value between 0 and 100, or any other numerical value.

[0032] If a particle contains only one mineral, the first hydrophobicity corresponding to that particle is the second hydrophobicity of that single mineral. If a particle contains multiple minerals, the first hydrophobicity corresponding to that particle is the sum of the products of each of the multiple second hydrophobicities and each of the multiple surface exposure ratios. The multiple second hydrophobicities represent the degree of low affinity for water of each of the multiple minerals contained in a given particle. The second hydrophobicity is a numerical value that relatively indicates the degree of low affinity for water of each mineral among the multiple minerals. For example, the second hydrophobicity may be a numerical value between 0 and 1, a numerical value between 0 and 100, or a numerical value that falls within any other range not mentioned above. The method for obtaining information indicating the second hydrophobicity will be described later.

[0033] The surface exposure ratio is the proportion of each mineral that occupies on the surface of a given particle. For example, the surface exposure ratio is the ratio of the exposed length of the mineral to the total circumference of the particle, or the proportion of the length occupied by the mineral relative to the total circumference of the particle. Note that the surface exposure ratio may also be other indicators that show the proportion of each mineral that occupies on the surface of the particle.

[0034] Figure 3 is a schematic diagram showing an example of one particle among several particles obtained from ore. Figure 3 shows particle R containing minerals X and Y. The first hydrophobicity will be explained in detail with reference to Figure 3. As an example, mineral X is a mineral with very low affinity for water, and the second hydrophobicity of mineral X is set to 1. Mineral Y is a mineral with very high affinity for water, and the second hydrophobicity of mineral Y is set to 0. The surface exposure ratio of mineral X is set to 0.3, and the surface exposure ratio of mineral Y is set to 0.7. In such an example, when particle R is used in flotation, if the portion of mineral X exposed on the surface of particle R comes into contact with a bubble, particle R will adhere to the bubble, and if the portion of mineral Y exposed on the surface of particle R comes into contact with a bubble, particle R will not adhere to the bubble. Therefore, the probability that particle R will adhere to a bubble when it comes into contact with a bubble is the probability that the portion of mineral X exposed on the surface of particle R comes into contact with the bubble, which is 0.3, the surface exposure ratio of mineral X. Based on the above, the first hydrophobicity of particle R is set to φ R Therefore, the first degree of hydrophobicity φ R This can be calculated as shown in equation (1).

number

number

[0035] Next, the method for generating each piece of information acquired by the acquisition unit 11 will be explained. First, multiple first proportion information, multiple second proportion information, multiple particle information, and information indicating multiple surface exposure ratios are calculated by mineral Liberation Analyzer (MLA). For example, MLA data is generated by performing electron microscope-fluorescence X-ray analysis (SEM-EDX) on multiple particles obtained from ore mined from a mine, and then performing mineral particle analysis on the analysis results. The MLA data includes various feature quantities for each of the multiple particles and an MLA image showing the shape of the multiple particles. As an example, the MLA data includes first proportion information, second proportion information, third proportion information, and particle information for each particle. The third proportion information is information indicating multiple surface exposure ratios. The acquisition unit 11 acquires the MLA data and acquires multiple first proportion information, multiple second proportion information, multiple third proportion information, and multiple particle information based on the acquired MLA data. Then, the acquisition unit 11 acquires multiple surface exposure ratios related to a predetermined particle based on the third proportion information for that predetermined particle.

[0036] Next, the second hydrophobicity is calculated as the probability that the mineral corresponding to that second hydrophobicity will float in the flotation process. For example, by performing flotation on multiple particles, floss and tail data are obtained for multiple particles. The above-described mineral particle analysis is performed on the floss and tail, and floss data, which is MLA data for the floss, and tail data, which is MLA data for the tail, are generated. In the floss and tail data, particles with a surface exposure ratio of a given mineral that is equal to or greater than a predetermined value are extracted as single-separated particles. The ratio of the number of single-separated particles contained in the floss to the total number of extracted single-separated particles is calculated as the second hydrophobicity of the given mineral. The predetermined value is, for example, 95% or more. For example, if flotation is performed using 100 single-separated particles and 20 single-separated particles are contained in the floss, the second hydrophobicity will be 0.2. Based on the above, the second hydrophobicity corresponding to one of several minerals is a numerical value that corresponds to the ratio of the number of single-element separated particles that are suspended by flotation to the total number of single-element separated particles used in flotation, when single-element separated particles (first particles) in which the surface exposure ratio of the said single-element separated particle is 95% or more are used in flotation.

[0037] As described above, when calculating the second degree of hydrophobicity, the value of the second degree of hydrophobicity changes if the surface of multiple particles used in flotation is made hydrophobic by chemical treatment using agents such as surfactants. Therefore, the chemical conditions in flotation are reflected in the second degree of hydrophobicity.

[0038] Finally, the product of the second hydrophobicity calculated as described above and the surface exposure ratio is calculated for each mineral contained in the particle. Then, the first hydrophobicity is calculated by taking the sum of the products of the second hydrophobicity and the surface exposure ratio for each mineral. In this way, hydrophobicity information is generated by calculating the first hydrophobicity through mineral particle analysis.

[0039] The generation unit 12 generates multiple probability information, each indicating the probability that multiple particles will float in the flotation process. Specifically, the generation unit 12 performs an estimation process for each of the multiple particles to estimate probability information indicating the probability that a predetermined particle will float in the flotation process, based on hydrophobicity information and particle information corresponding to that predetermined particle. In the estimation process, the generation unit 12 inputs the hydrophobicity information and particle information acquired by the acquisition unit 11 into the learning model M to estimate probability information indicating the probability that the particle corresponding to the hydrophobicity information and particle information will float in the flotation process. The probability information may be a numerical value between 0 and 1, a numerical value between 0 and 100, or any other numerical value.

[0040] Figure 4 shows an example of a learning model M. As shown in Figure 4, the learning model M is a neural network comprising an input layer M1, a hidden layer M2, and an output layer M3. The input layer M1 and the output layer M3 each contain, for example, one node. The hidden layer M2 has three fully connected layers M21, M22, and M23. Each of the fully connected layers M21, M22, and M23 contains 128 nodes. The input layer M1 receives an input signal representing a vector of hydrophobicity information and particle information. Each node in the fully connected layers M21, M22, and M23 outputs an output signal as the result of substituting the input signal, after applying an affine transformation, into an activation function (ReLU function). The output layer M3 normalizes the result of applying an affine transformation to the input signal at one node to a range of 0 to 1 using a sigmoid function or the like, and outputs it. The output of the output layer M3 is considered to be the probability (distribution probability (recovery rate)) that a particle is included in the froth.

[0041] The calculation unit 13 calculates the recovery rate and grade of the elements to be sorted based on multiple first proportion information, multiple second proportion information, and multiple probability information. The recovery rate E of the elements to be sorted is calculated as shown in the following formula (3). The grade D of the elements to be sorted is calculated as shown in the following formula (4).

number

[0042] The calculation of the mass of the element to be sorted contained in the froth in formulas (3) and (4) will be described. First, the mass of a predetermined particle is calculated by multiplying the mass F of the feed by the ratio w of the mass of the predetermined particle to the mass F of the feed i Subsequently, the mass of the element to be sorted in the predetermined particle is calculated by multiplying the mass of the predetermined particle by [X], which is the ratio that the mass of the element to be sorted occupies in the mass of the predetermined particle i Subsequently, the expected value of the mass contained in the froth among the mass of the element to be sorted in the predetermined particle is calculated by multiplying the mass of the element to be sorted in the predetermined particle by R, which is the probability of being distributed from the feed to the froth i Finally, the sum of the expected values calculated for each of all the particles is taken as the mass of the element to be sorted contained in the froth.

[0043] The calculation of the mass of the element to be sorted contained in the feed in formula (3) will be described. First, the mass of a predetermined particle is calculated by multiplying the mass F of the feed by the ratio w of the mass of the predetermined particle to the mass F of the feed i Subsequently, the mass of the element to be sorted in the predetermined particle is calculated by multiplying the mass of the predetermined particle by [X], which is the ratio that the mass of the element to be sorted occupies in the mass of the predetermined particle iBy multiplying by a certain factor, the mass of the element to be sorted in a given particle is calculated. Finally, the sum of the masses calculated for each particle is considered to be the total mass of the element to be sorted in the feed.

[0044] Let's explain the total mass of particles contained in the floss in equation (4). First, the mass of the feed F is given by the ratio w of the mass of a given particle to the mass of the feed F. i The mass of a given particle is calculated by multiplying it by R. Next, the mass of the given particle is multiplied by the probability that it is distributed from the feed to the floss. i By multiplying by a factor, the expected value of the mass of a given particle that is contained in the floss is calculated. Finally, the sum of these expected values ​​calculated for each particle is considered to be the total mass of all particles contained in the floss.

[0045] The construction unit 14 uses training data generated based on froth and tail data obtained by flotation to construct a learning model M using machine learning. This model takes hydrophobicity information and particle information of a predetermined particle as input and outputs probability information of that predetermined particle in a manner that approximates the flotation results of that predetermined particle. The learning model M is a learning model that takes hydrophobicity information and particle information as input and outputs probability information. The construction unit 14 stores the constructed learning model M in the built-in memory or storage medium of the estimation device 1. Machine learning includes supervised learning, unsupervised learning, and reinforcement learning, and among these learning methods are deep learning and neural network learning. In this embodiment, the neural network shown in Figure 4 is employed. In addition to being constructed by the construction unit 14, the learning model M may also be generated by an external computer or the like and downloaded to the estimation device 1.

[0046] The construction unit 14 generates training data based on floss data and tail data obtained by flotation. Specifically, the construction unit 14 extracts a portion of the floss data and a portion of the tail data to match the experimental results of the allocation ratio to floss and tail for multiple particles. The construction unit 14 merges the portion of the floss data and a portion of the tail data to create simulated feed data. At this time, the construction unit 14 sets the probability information of particles contained in the floss to 1 and the probability information of particles contained in the tail to 0. Furthermore, the construction unit 14 acquires hydrophobicity information and particle information based on the simulated feed data, as described above. The construction unit 14 uses the combination of the acquired hydrophobicity information and particle information and the simulated feed data as training data.

[0047] The construction unit 14 constructs a learning model M by repeatedly performing batch learning using the generated training data. At this time, the construction unit 14 optimizes the parameters within the learning model M using stochastic gradient descent.

[0048] Next, the flow of the estimation method according to the embodiment will be described. Figure 5 is a flowchart showing the procedure of the estimation process according to the embodiment. In the estimation process, the recovery rate of the element to be sorted and the grade of the element to be sorted are estimated in flotation, in which particles containing the element to be sorted are sorted by attaching them to bubbles and floating them among a plurality of particles obtained from mined ore.

[0049] First, training data is generated by the estimation device 1 (step S1). Next, a learning model M is constructed by the estimation device 1 based on the training data (step S2). Subsequently, multiple hydrophobicity information, multiple first proportion information, multiple second proportion information, and multiple particle information are acquired by the estimation device 1 (step S3: acquisition step).

[0050] Next, estimation device 1 performs estimation processing for each of the multiple particles to generate multiple probabilistic information (step S4: generation step). In this estimation process, hydrophobicity information and particle information are input to the learning model M, and probabilistic information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in flotation is estimated.

[0051] Finally, the estimation device 1 calculates the recovery rate and grade of the elements to be sorted based on multiple first proportion information, multiple second proportion information, and multiple probability information (Step S5: Calculation step).

[0052] The effects and benefits of the estimation device 1, learning model M, and estimation method according to the above embodiment will be explained.

[0053] In the estimation device 1 and estimation method according to the embodiment, multiple hydrophobicity information, each indicating the degree of low affinity to water for multiple particles, and multiple particle information, each indicating feature quantities for multiple particles, are input to the learning model M. As a result, probability information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in flotation is estimated. With the above configuration, since hydrophobicity information corresponding to the type of ore used in flotation is input to the learning model M, even if the type of ore used in flotation is changed, the probability information can be estimated accurately without reconstructing the learning model M. Therefore, the recovery rate of the elements to be sorted and the grade of the elements to be sorted can be calculated accurately based on multiple first proportion information, multiple second proportion information, and multiple probability information. This makes it possible to provide a highly versatile estimation device 1.

[0054] Furthermore, in the estimation device relating to the comparative example, it is conceivable to construct a learning model relating to the comparative example using the results of flotation using a specific ore as training data, and then calculate the recovery rate of the elements to be sorted using the learning model relating to the comparative example. In the learning model relating to the comparative example, for example, information indicating the elemental composition of ore particles and particle information are input, and probability information is output. In such a learning model, if the type of ore used for flotation is changed, it may be necessary to retrain the learning model relating to the comparative example using training data corresponding to the changed type of ore. In contrast, according to the estimation device 1 of the embodiment, even if the type of ore used for flotation is changed, probability information can be estimated with high accuracy without rebuilding the learning model. As a result, an estimation device 1 with higher versatility can be provided compared to the estimation device relating to the comparative example.

[0055] The above effects will be explained in detail with reference to Figures 6 and 7. Specifically, the explanation will be given using experimental values ​​of the recovery rate and quality of the element to be sorted, estimated values ​​of the recovery rate and quality of the element to be sorted in the estimation device 1 according to the embodiment, and estimated values ​​of the recovery rate and quality of the element to be sorted in the estimation device according to the comparative example. In the following explanation, the element to be sorted is copper (Cu).

[0056] First, as a prerequisite, we will explain the T ore, A ore, and B ore used in calculating experimental and estimated values. Figure 6 is a table showing an example of the ore used for generating training data and the ore used for flotation. As shown in Figure 6, the T ore is the sample used to obtain the training data used to construct the learning model M. The T ore contains 0.94 wt% copper, 16.4 wt% iron, and 2.49 wt% sulfur (S). The main Cu mineral contained in the T ore is chalcopyrite (CuFeS2), and the main gangue minerals contained in the T ore are silicate minerals and iron oxide minerals. The A ore and B ore are samples used to verify the versatility of the learning model M. The A ore contains 0.98 wt% copper, 1.27 wt% iron, and 0.55 wt% sulfur. Ore B contains 1.01 wt% copper, 1.13 wt% iron, and 0.54 wt% sulfur. The main Cu minerals contained in ores A and B are chalcocite (Cu2S), bornite (Cu5FeS4), and chalcopyrite (CuFeS2), while the main gangue minerals contained in ores A and B are silicate minerals and calcite.

[0057] Next, we will describe an example of flotation performed on ore T, ore A, and ore B as a pretreatment for calculating experimental and estimated values. First, a flotation machine (Denver D-12 manufactured by 911METALLURGY CORP.) equipped with a 4L cell is prepared, and multiple particles obtained from the ore and water are introduced into the cell. Next, sodium hydroxide (NaOH) is introduced into the cell to adjust the pH of the solution in the cell to 10. Subsequently, a capture agent (Aero Float 3894) is introduced into the cell at a rate of 50g / t-ore (amount of chemical used in flotation per ton of ore) for 2 minutes (conditioning is performed using the capture agent), thereby adjusting the hydrophobicity of the surface of minerals containing the elements to be separated (e.g., chalcopyrite CuFeS2, chalcocite Cu2S, and bornite Cu5FeS4, etc.) so that they adhere reliably to the bubbles. Next, a foaming agent (MIBC) is added to the cell at a rate of 25 g / t-ore for 1 minute (conditioning is performed using the foaming agent) to prepare the cell for bubble generation. Finally, the solution in the cell is stirred at 750 rpm for 10 minutes, generating bubbles within the cell. Particles that adhere to the bubbles and become suspended are collected as floss, while particles that do not adhere to the bubbles and remain in the solvent are collected as tails.

[0058] Next, based on the flotation results for ore T, ore A, and ore B, mineral particle analysis is performed on multiple particles obtained from the ore, and feed data, which is MLA data related to the feed, is acquired. Then, mineral particle analysis is performed on the above flotation results, and floss data and tail data are acquired. For example, in the floss and tail data for ore T, the allocation rate to floss of ore T is 4.1 wt%, and the allocation rate to tail is 95.9 wt%. In the floss and tail data for ore A, the allocation rate to floss of ore A is 1.1 wt%, and the allocation rate to tail is 98.9 wt%. In the floss and tail data for ore B, the allocation rate to floss of ore B is 1.5 wt%, and the allocation rate to tail is 98.5 wt%.

[0059] Finally, based on the floss and tail data for ore A and ore B, experimental values ​​for the recovery rate and grade of Cu, the element to be sorted, are calculated. The experimental value E0 for Cu recovery rate is calculated using the following formula (5), and the experimental value D0 for Cu grade is calculated using the following formula (6).

number

[0060] The estimation device 1 according to this embodiment generates training data based on froth data and tail data obtained by flotation using T ore. The estimation device 1 constructs a learning model M based on this training data. Using the constructed learning model M, the estimation device 1 calculates the recovery rate and grade of Cu, the element to be separated, in flotation using T ore, A ore, and B ore.

[0061] In the estimation device of the comparative example, the recovery rate and grade of Cu, the element to be sorted, are calculated using a learning model that takes information indicating the elemental composition of ore T and particle information as input and outputs probabilistic information. Specifically, the estimation device of the comparative example generates training data based on froth data and tail data obtained by flotation using ore T. The estimation device of the comparative example constructs a learning model of the comparative example based on this training data. Using the constructed learning model, the estimation device calculates the recovery rate and grade of Cu, the element to be sorted, in flotation using ore T, ore A, and ore B.

[0062] Figures 7(a), 7(b), and 7(c) are graphs showing experimental values ​​of the recovery rate and grade of the elements to be sorted, and estimated values ​​of the recovery rate and grade of the elements to be sorted using the estimation device 1 according to the embodiment and the estimation device according to the comparative example, for each type of ore. Figure 7(a) is a graph for ore T, Figure 7(b) is a graph for ore A, and Figure 7(c) is a graph for ore B. In Figures 7(a), 7(b), and 7(c), the horizontal axis represents the recovery rate of Cu into the floss, and the vertical axis represents the grade of Cu into the floss. In Figures 7(a), 7(b), and 7(c), plots showing combinations of numerical values ​​for the recovery rate of the elements to be sorted and numerical values ​​for the grade of the elements to be sorted are placed on the graphs.

[0063] As described above, the learning model of the estimation device in the comparative example and the learning model M of the estimation device 1 in the embodiment are constructed using training data based on froth data and tail data obtained by flotation using T ore. Therefore, as shown in Figure 7(a), for T ore containing chalcopyrite as the main Cu mineral, the plot Q of the estimated value of the estimation device in the comparative example T and plot R of the estimated value of the estimation device 1 according to the embodiment T However, the plot of experimental values ​​P T It is almost identical.

[0064] However, as shown in Figures 7(b) and 7(c), ore A and ore B contain Cu minerals other than the main Cu minerals in ore T, so plot Q of the estimated values ​​of the estimation device for the comparative example. A ,Q B This is a plot of experimental values ​​P A ,P B The results deviate significantly from the experimental values. For example, the error in the estimated recovery rate of the estimation device in the comparative example is about 20% of the experimental value, and the error in the estimated grade of the estimation device in the comparative example is also about 20% of the experimental value. This result indicates that the estimation device in the comparative example (an estimation device that uses a model that uses the elemental composition of particles as input features) cannot accurately estimate (predict) the results of flotation of a different type of ore (an ore with a different mineral or chemical composition than the ore used to create the training data).

[0065] In contrast, as shown in Figures 7(b) and 7(c), in the estimation device 1 according to the embodiment, the plot of the estimated value R is obtained in both the case of ore A and ore B. A and R B Each of these is a plot of the estimated value of the estimation device related to the comparative example Q A and Q B Plot P of experimental values ​​is better than each of the others. A and P B It is located in a place close to each of the experimental values ​​P A and P B These results are in close agreement with each of the above. For example, the error in the estimated recovery rate of the estimation device 1 according to the embodiment is 5% or less compared to the experimental value, and the error in the estimated grade of the estimation device 1 according to the embodiment is 5% or less compared to the experimental value. This result shows that the estimation device 1 according to the embodiment (the estimation device 1 that uses a learning model M used as a feature quantity to input the first hydrophobicity of particles) can estimate the results of flotation of a different type of ore (an ore with a different mineral or chemical composition than the ore used to create the training data) with high accuracy, indicating that it is a highly versatile estimation device.

[0066] Next, other effects and advantages of the estimation device 1 according to this embodiment will be described. For example, in the initial stages of a project to develop a new mine, it is difficult to obtain a sufficient amount of ore samples and generate training data to build a learning model. However, with the estimation device 1 of this embodiment, flotation can be performed using a small amount of ore samples, and a learning model M can be constructed using training data generated based on the results of the flotation. Furthermore, the first hydrophobicity can be defined not only for specific minerals and elements, but also for particles containing any minerals and elements. Therefore, with the estimation device 1 of this embodiment, it is possible to estimate the results of flotation using particles containing any minerals and elements, thus providing a highly versatile estimation device 1.

[0067] Furthermore, in this embodiment, one of the multiple particles contains multiple minerals, and the first hydrophobicity corresponding to this one particle is the sum of the products of each of the multiple second hydrophobicities, which represent the degree of low affinity of each of the multiple minerals to water, and each of the multiple surface exposure ratios, which represent the proportion of each of the multiple minerals that occupy on the surface of the particle. With this configuration, changes in chemical conditions in flotation are reflected in the first hydrophobicity using the second hydrophobicity, so that when chemical conditions in flotation are changed, probabilistic information can be estimated with high accuracy without reconstructing the learning model M. In addition, since the first hydrophobicity can be calculated with higher accuracy based on both the multiple second hydrophobicities and the multiple surface exposure ratios, probabilistic information can be calculated with higher accuracy.

[0068] Furthermore, in this embodiment, the second hydrophobicity corresponding to one of the multiple minerals is a numerical value that corresponds to the ratio of the number of first particles suspended by flotation to the total number of first particles used in flotation, when first particles, which are particles with a surface exposure ratio of 95% or more of the said one mineral, are used in flotation. This allows for a more accurate calculation of the first hydrophobicity, and thus allows for a more accurate calculation of probability information.

[0069] Furthermore, in this embodiment, the surface exposure ratios of multiple minerals are calculated by mineral particle analysis. In this case, the surface exposure ratio of each of the multiple minerals can be calculated easily and accurately.

[0070] Furthermore, in this embodiment, the particle information includes size information indicating the size of the particle, specific gravity information indicating the specific gravity of the particle, and shape information indicating the shape of the particle. In this case, probability information can be easily and accurately estimated based on the hydrophobicity information and particle information.

[0071] In the learning model M according to this embodiment, multiple hydrophobicity information, each indicating the degree of low affinity for water for multiple particles, and multiple particle information, each indicating the feature quantities for multiple particles, are taken as input, and probability information, which indicates the probability that the particles corresponding to the hydrophobicity information and particle information will float in flotation, is output. As a result, as described above, the probability information can be estimated with high accuracy without rebuilding the learning model M.

[0072] Although various embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and may be modified or applied to other things without changing the gist of each claim.

[0073] In the estimation apparatus 1 according to the above embodiment, the substance to be sorted was an element, but it may also be a mineral. For example, the substance to be sorted may be a mineral with high economic value, and one example is a mineral containing elements such as copper, lead, and zinc, as well as coal, etc.

[0074] In the estimation device 1 according to the above embodiment, the first hydrophobicity was the sum of the products of each of the multiple second hydrophobicity values ​​and each of the multiple surface exposure ratios, but is not limited to this. The first hydrophobicity may be set, for example, by the flotation operator based on the operator's own experience or existing research results.

[0075] In the estimation device 1 according to the above embodiment, the second hydrophobicity was a value corresponding to the ratio of the number of individual separated particles suspended by flotation to the total number of individual separated particles used in flotation, but is not limited to this. The second hydrophobicity may be set by the flotation operator based on the operator's own experience or existing research results.

[0076] In the estimation device 1 according to the above embodiment, multiple first proportion information, multiple second proportion information, multiple particle information, and multiple surface exposure proportions were calculated by mineral particle analysis, but they may also be calculated by known methods other than mineral particle analysis.

[0077] In the estimation device 1 according to the above embodiment, the particle information included size information indicating the size of the particles, specific gravity information indicating the specific gravity of the particles, and shape information indicating the shape of the particles. However, it may also include at least one of the size information, specific gravity information, and shape information, or it may include other information not included in these.

[0078] In the estimation device 1 according to the above embodiment, both the recovery rate of the elements to be sorted and the grade of the elements to be sorted were calculated, but it is sufficient to calculate at least one of the recovery rate of the elements to be sorted and the grade of the elements to be sorted. [Explanation of symbols]

[0079] 1...Estimation device, 11...Acquisition unit, 12...Generation unit, 13...Calculation unit, 101A...CPU, 101B...GPU, 102...RAM, 103...ROM, 104...Communication module, 106...Input / Output module, D...Grade, E...Recovery rate, M...Learning model, R...Particle, X...Mineral, Y...Mineral, φ R ...1st degree of hydrophobicity, φ i ...2nd degree of hydrophobicity, θ i ...surface exposure ratio.

Claims

1. In flotation ore separation, in which particles containing a substance to be separated are separated from a plurality of particles obtained from mined ore by attaching them to bubbles and causing them to float, an estimation device for estimating at least one of the recovery rate of the substance to be separated and the grade of the substance to be separated, An acquisition unit that acquires multiple hydrophobicity information indicating the degree of low affinity for water for each of the multiple particles, multiple first proportion information indicating the proportion of the mass of one of the multiple particles to the total mass of the multiple particles, multiple second proportion information indicating the proportion of the mass of the substance to be sorted to the mass of one of the multiple particles, and multiple particle information indicating characteristic quantities for each of the multiple particles. A generation unit generates multiple sets of probability information by inputting the hydrophobicity information and particle information acquired by the acquisition unit into a learning model, and performing an estimation process for each of the multiple particles to estimate probability information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in the flotation process. Estimation device comprising: a calculation unit that calculates at least one of the recovery rate of the substance to be sorted and the quality of the substance to be sorted based on the plurality of first proportion information and the plurality of second proportion information acquired by the acquisition unit, and the plurality of probability information generated by the generation unit.

2. One of the aforementioned plurality of particles contains multiple minerals, The estimation device according to claim 1, wherein the first hydrophobicity corresponding to one particle is the sum of the products of each of a plurality of second hydrophobicities, which are the degree of low affinity of each of the plurality of minerals to water, and each of a plurality of surface exposure ratios, which are the proportion of each of the plurality of minerals that occupy on the surface of the particle.

3. The estimation apparatus according to claim 2, wherein the second hydrophobicity corresponding to one of the plurality of minerals is a numerical value corresponding to the ratio of the number of first particles suspended by flotation to the total number of first particles used in flotation when first particles, which are particles in which the surface exposure ratio of the one mineral is 95% or more, are used in the flotation.

4. The estimation apparatus according to claim 2, wherein the multiple surface exposure ratios are calculated by mineral particle analysis.

5. The estimation apparatus according to any one of claims 1 to 4, wherein the particle information includes at least one of size information indicating the size of the particle, specific gravity information indicating the specific gravity of the particle, and shape information indicating the shape of the particle.

6. A learning model used for flotation, which separates particles containing the substance to be sorted from multiple particles obtained from mined ore by attaching them to bubbles and causing them to float. A learning model that takes hydrophobicity information, which indicates the degree of low affinity for water in particles obtained from mined ore, and particle information, which indicates characteristic quantities of the particles, as input, and outputs probability information, which indicates the probability that the particles will float in flotation.

7. In flotation, a method for estimating at least one of the recovery rate of the target substance and the grade of the target substance, in which particles containing the target substance are separated from a plurality of particles obtained from mined ore by attaching them to bubbles and causing them to float. The steps include obtaining a plurality of hydrophobicity information, each representing a first hydrophobicity, which is the degree of low affinity for water for the plurality of particles; a plurality of first proportion information, each representing the proportion of the mass of one of the plurality of particles to the total mass of the plurality of particles; a plurality of second proportion information, each representing the proportion of the mass of the substance to be sorted to the mass of one of the plurality of particles; and a plurality of particle information, each representing a characteristic quantity for the plurality of particles. The steps include: inputting the hydrophobicity information and particle information acquired in the acquisition step into a learning model, and performing an estimation process for each of the plurality of particles to estimate probability information indicating the probability that the particles corresponding to the hydrophobicity information and particle information will float in the flotation process, thereby generating a plurality of the aforementioned probability information; An estimation method comprising the step of calculating at least one of the recovery rate of the substance to be sorted and the quality of the substance to be sorted based on the plurality of first proportion information and the plurality of second proportion information obtained in the acquisition step, and the plurality of probability information generated in the generation step.

Images