Intelligent auxiliary control system and method for phosphate ore flotation process based on LIBS online detection technology

Abstract

The present application relates to the technical field of mineral processing, in particular to a phosphate rock flotation process intelligent auxiliary control system and method based on LIBS online detection technology, the system comprises a real-time sensing layer, a digital twin layer, an intelligent decision-making layer and an automatic execution layer; the real-time sensing layer realizes real-time online detection of multiple elements such as P2O5 and MgO of raw ore, concentrate and the like through a multi-node knife-edge sampler, a multi-path splitter and an online laser spectrum analyzer; the digital twin layer adopts a mechanism and data-driven fusion model to complete grade soft measurement and ultra-real-time prediction; the intelligent decision-making layer generates operation variable optimization instructions such as reagent addition amount and liquid level in combination with expert rules and optimization algorithms; and the automatic execution layer issues the instructions to the field execution mechanism to form a closed-loop control. The present application realizes digital and intelligent operation of the flotation process, improves product quality stability, reduces reagent consumption and production cost, and improves resource comprehensive utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of mineral processing technology, specifically to an intelligent auxiliary control system and method for phosphate rock flotation process based on LIBS online detection technology. Background Technology

[0002] Phosphate rock is an important strategic mineral resource in my country, but about 80% of it is low- to medium-grade ore, requiring enrichment through flotation. Currently, the phosphate rock flotation production process faces the following serious challenges: 1. Severely Delayed Process Indicator Testing: Under the current production model, the grade analysis of key stages such as raw ore, concentrate, middlings, and tailings in flotation roughing, cleaning, and scavenging operations relies on manual, timed sampling, sample preparation, and laboratory analysis. This process takes several hours, resulting in severe delays in production data and making it impossible to adjust and control key parameters such as the addition of flotation reagents in real time and effectively.

[0003] 2. The "black box" nature of the flotation production process: The flotation process is a complex physicochemical process, and its effectiveness is affected by various factors such as the properties of the raw ore, the reagent formulation, and the operating conditions of the equipment. Due to the lack of online monitoring and analysis data on the real-time grade of key process parameters at each stage, operators mainly rely on personal experience to adjust production, resulting in large fluctuations in the quality of the concentrate product, such as the P2O5 grade and the content of impurities such as MgO, unstable recovery rates, and high production costs.

[0004] 3. Limitations of existing online detection technologies: (1) Online X-ray fluorescence (XRF) analyzer: The sensitivity for detecting light elements (such as Mg, Al, Si) is insufficient or impossible, and these are impurity elements that need to be controlled in the phosphate rock flotation process. At the same time, XRF analysis usually requires pretreatment such as pressing the sample into tablets, making it difficult to achieve real-time online analysis.

[0005] (2) Traditional LIBS technology applications: still in the laboratory stage, or only used for offline analysis. Applying it to industrial sites with complex composition and harsh environment faces challenges in signal stability caused by slurry particle size, concentration, bubbles, etc., as well as complex engineering integration problems.

[0006] Therefore, the flotation process requires a technology that can perform real-time, multi-element, and high-precision online detection of the entire phosphate rock flotation process, and can deeply integrate the detection data with the production process to ultimately form a closed-loop intelligent control solution. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent auxiliary control system and method for the phosphate rock flotation process based on LIBS online detection technology, so as to solve the problems mentioned in the background art, such as the lag in existing phosphate rock flotation detection, reliance on manual experience, and large fluctuations in product quality.

[0008] To achieve the above objectives, the present invention provides the following technical solution: An intelligent auxiliary control system for the phosphate rock flotation process based on LIBS online detection technology includes: The real-time sensing layer is used to automatically sample and perform multi-element online laser spectral analysis of the slurry at multiple process nodes in the phosphate rock flotation process to obtain data on grade changes in phosphate rock flotation operations. The digital twin layer, based on the spectral data and production process data output by the real-time sensing layer, constructs and runs a process index prediction model to achieve soft measurement and trend prediction of grade. The intelligent decision-making layer, based on the output of the digital twin layer, combines expert rules and optimization algorithms to generate optimized control instructions for the parameters of each operational variable in the flotation process; The automatic execution layer connects to the underlying distributed control system via an industrial communication protocol, and sends the optimized control commands to the field actuators to complete the closed-loop optimized control of the flotation process.

[0009] Preferably, the real-time sensing layer includes: Multiple knife-edge samplers are deployed at key process nodes such as raw ore, rougher concentrate, scavenger concentrate, total concentrate, rougher tailings, scavenger tailings, and total tailings to directly extract representative slurry samples from the original process pipeline. The slurry conveying pipeline is used to connect the knife-edge sampler and the multi-channel divider to achieve stable delivery of slurry samples. The multi-channel reducer is connected to the slurry conveying pipeline and has a built-in buffer tank, slag screen and tuning fork level switch. It is controlled by PLC program to realize the automatic switching of multiple slurry samples according to a predetermined time sequence, so as to ensure that the slurry flow entering the online laser spectrometer is stable, free of large particulate impurities and constant liquid level. An online laser spectrometer, connected to the multi-channel reducer, uses dual-pulse laser-induced breakdown spectroscopy to quantitatively analyze the elements P2O5, MgO, Fe2O3, Al2O3, SiO2, and CaO in the slurry.

[0010] Preferably, the technical parameters of the online laser spectrometer meet the following requirements: it uses two pulsed lasers, each with an energy of 100 mJ and a frequency of 10 Hz; it integrates an automatic rinsing system to clean the sample chamber and pipelines before and after each measurement; it is equipped with an online concentration meter to monitor the slurry concentration in real time for spectral data compensation and correction; it uses a high-resolution fiber optic spectrometer with a wavelength range of 207–430 nm; and the single-channel single-analysis time, including sampling, measurement, and cleaning, is ≤5 minutes.

[0011] Preferably, the process index prediction model used in the digital twin layer is a hybrid model that combines a mechanism model and a data-driven model. The data-driven model is a combination of radial basis function (RBF) neural network and partial least squares regression (PLSR). The RBF neural network is used to predict the grades of P, Mg, and Fe elements, and its output formula is: ; In the formula, Here, b is the weight value, and b is the bias term. The output values ​​are the predicted grades of phosphorus, magnesium, and iron, representing the characteristic states of the hidden layer nodes. The width of the basis functions. For feature similarity; The PLSR model is used to predict the grades of Al, Si, and Ca elements. Its regression formula is as follows: In the formula, This is the preprocessed spectral data matrix. For regression coefficients, These are the predicted grades of Al, Si, and Ca elements. This represents the regression prediction error term.

[0012] Preferably, the process performance prediction model of the digital twin layer has the following functions: The online soft measurement function outputs the current elemental grades of P2O5, MgO, Fe2O3, Al2O3, SiO2, and CaO at each process node in real time, as feedback signals for the control system. The ultra-real-time prediction function predicts the trend of changes in concentrate grade and tailings grade over a period of time based on current operating conditions. The model self-update function periodically compares laboratory test data with online prediction data. When the deviation exceeds a threshold, the incremental learning mechanism is automatically triggered to update the model parameters.

[0013] Preferably, the intelligent decision-making layer includes: An expert rule base stores IF-THEN control rules constructed based on process knowledge. These rules include: If the predicted MgO grade in the concentrate is greater than the set upper limit, and the MgO grade in the raw ore is stable, then increase the amount of inhibitor added. Or, if the predicted P2O5 grade in the IF tailings exceeds the set upper limit, THEN adjust the collector dosage and flotation machine level. The multi-objective optimization unit takes maximizing the P2O5 grade of the concentrate, minimizing the MgO content of impurities, and minimizing reagent consumption as optimization objectives. It calculates the optimal setpoint sequence of operating variables by combining the prediction results of the digital twin layer.

[0014] Preferably, the intelligent decision-making layer also includes a working condition identification unit, which identifies abnormal working conditions such as overflow and settling by analyzing the flotation machine power, stirring tank speed, aeration volume, and spectral signal stability parameters of the online laser spectrometer in real time; when an abnormality is identified, it automatically switches to a preset safety control mode or adjusts the control parameters.

[0015] Preferably, the operating variables controlled by the automatic execution layer include the amount of collector added, the amount of inhibitor added, the flotation machine level, and the aeration rate; the industrial communication protocols include the OPC UA protocol and the Modbus TCP protocol; and the real-time sensing layer constructs a closed-loop sampling channel from the original process pipeline to the online laser spectrometer through the slurry conveying pipeline.

[0016] On the other hand, the present invention also provides an intelligent auxiliary control method for the phosphate rock flotation process based on the above system, comprising the following steps: S1: Data is collected through the real-time sensing layer. Specifically, the slurry samples from each process node are intercepted from the original process pipeline using a knife-edge sampler. The samples are then transported to a multi-channel reducer via the slurry conveying pipeline for flow stabilization, impurity removal, and timing switching. Finally, the samples are sent to an online laser spectrometer to complete the acquisition of multi-element laser spectral data, while simultaneously acquiring slurry concentration data. S2: Through the digital twin layer, based on the spectral data, slurry concentration data, and real-time production process data output by the online laser spectrometer, the current and future element grades are predicted using a process index prediction model; the digital twin layer periodically uses laboratory test data to perform online calibration and updates to the process index prediction model; S3: Through the intelligent decision-making layer, the predicted value of grade is compared with the target value, and the adjustment instructions of the operation variables are generated by using expert rules and multi-objective optimization algorithms. S4: The adjustment command is sent to the actuator through the automatic execution layer. Combined with the real-time data fed back by the sampling and detection link composed of the slurry conveying pipeline and the multi-channel divider through the real-time perception layer, the closed-loop optimization control of the flotation process is realized.

[0017] Preferably, in step S3, the intelligent decision-making layer adopts a hierarchical decoupling control strategy for the coarse selection, fine selection, and sweeping operation areas: The coarse selection operation employs a control strategy combining feedforward compensation and feedback correction. The control formula is as follows: In the formula, Add a pump frequency command to the inhibitor at time t. This represents the real-time content of impurities in the raw ore. This serves as a benchmark reference value for the impurity content of the raw ore. Forward compensation coefficient, This represents the deviation between the measured and target values ​​of impurity content in the roughing concentrate. , These are the proportional and integral coefficients of the PID controller, respectively. The selected assignments employ an incremental gradient control algorithm, with the control formula as follows: ; ; In the formula, The optimal amount of collector added at time t. To control the increment in this cycle, To improve the P2O5 grade of the refined concentrate, The target grade of P2O5 in the concentrate is [missing information]. To adjust the step size coefficient, The response exponent is 0 < β < 1; The sweeping operation employs nonlinear threshold gain control, and the control formula is as follows: ; In the formula, The amount of sweep inhibitor added at time t. This is the basic dosage of medication under normal operating conditions. To improve the P2O5 grade of the tailings, Set the P2O5 grade for tailings. γ is the radical gain coefficient, where γ > 1.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention, by deploying a real-time sensing layer, achieves real-time online detection of multiple nodes and elements in the phosphate rock flotation process. The single-channel, single-analysis time (including sampling, measurement, and cleaning) does not exceed 5 minutes, overcoming the hours-long lag caused by traditional production methods that rely on manual timed sampling, sample preparation, and laboratory analysis. This transforms production control from post-event remediation to pre-event prevention and in-process regulation. The digital twin layer, based on a grade prediction model built from real-time sensing data, can perform online soft measurement of key process indicators and predict future trends in real time, providing precise data support for the intelligent decision-making layer, thereby significantly improving the timeliness and accuracy of production control.

[0019] 2. This invention, through closed-loop control, can quickly respond to fluctuations in raw ore properties and changes in equipment operating conditions, significantly reducing the fluctuation range of P2O5 grade and impurity content in concentrate products, thereby improving product quality stability and pass rate. The relative error between predicted and laboratory values ​​can be controlled within P2O5≤5%, MgO≤10%, Fe2O3≤10%, Al2O3≤10%, and SiO2≤10%, effectively improving concentrate quality. Simultaneously, precise control of reagent addition avoids reagent waste, reduces production costs, and optimizes recovery rates to reduce phosphorus resource loss in tailings, improving the comprehensive utilization level of resources. After the system was put into operation, the standard deviation of concentrate P2O5 grade fluctuation decreased by approximately 40%, the MgO content pass rate increased from approximately 80% before operation to over 95%, average reagent consumption decreased by approximately 8%, and flotation recovery rate increased by 2 percentage points.

[0020] 3. The intelligent decision-making layer of this invention adopts a hierarchical decoupled control strategy, combining an expert rule base and a multi-objective optimization algorithm. It can generate optimal operational variable settings based on the characteristics of different work areas, significantly reducing reliance on operator experience and improving the automation and intelligence level of production management. The automatic execution layer accurately sends optimization instructions to the field actuators, forming a complete closed-loop control circuit, realizing unattended operation and precise control of the production process. Furthermore, through digital twins and remote monitoring cloud services, this invention achieves complete transparency of the production process. Managers can remotely monitor the production status and equipment health in real time, facilitating fault diagnosis and system maintenance, and significantly reducing operation and maintenance costs and labor intensity. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are explained in detail together with the embodiments of the invention, but do not constitute a limitation thereof.

[0022] Figure 1 This is a diagram illustrating the overall architecture of the intelligent auxiliary control system for the phosphate rock flotation process based on online LIBS according to the present invention. Figure 2This is a schematic diagram of the online LIBS analysis system of the present invention. In the diagram, 1 is a knife-edge sampler, 2 is a slurry conveying pipeline, 3 is a multi-channel reducer, 4 is an online laser spectrometer, and 5 is the original process pipeline. Figure 3 This is a flowchart illustrating the construction of a quality prediction model that integrates mechanism and data-driven approaches in the digital twin layer of this invention. Figure 4 This is a control logic block diagram of the collaboration between expert rules and multi-objective optimization in the intelligent decision-making layer of this invention; Figure 5 This is a curve comparing the online predicted values ​​and laboratory test values ​​of P2O5, MgO, and Al2O3 grades in the concentrate of this invention. Figure 6 This is a flow chart of the flotation process of the present invention; Figure 7 The diagram shows the connection diagram of the equipment and the schematic diagram of the sampling point layout of the present invention. In the diagram, (1) raw ore; (2) rough concentrate; (3) fine tailings; (4) fine middlings; (5) concentrate tailings; (6) fine I concentrate; (7) fine II concentrate; (8) pre-selected tailings; (9) pre-selected concentrate; (10) re-selected concentrate; and (11) re-selected tailings. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] This intelligent auxiliary control system and method for phosphate rock flotation based on LIBS online detection technology constructs a closed-loop intelligent auxiliary control system for phosphate rock flotation, encompassing "real-time sensing, digital twin, intelligent decision-making, and automatic execution." Its core is to achieve real-time detection and closed-loop optimization of multiple nodes and elements through specific device deployment, hybrid model construction, and hierarchical control strategies. The following details the system's implementation process, key parameters, and operational verification process in conjunction with a specific application scenario: 1. Real-time sensing layer: A comprehensive and high-precision online LIBS analysis and control network is designed and deployed for key control points in the phosphate rock flotation process. This layer is the foundation of the system and is responsible for collecting data on the core parameters for controlling the flotation process.

[0025] (1) Sampling network design: According to the flotation process flow, scraper-type pneumatic knife-edge samplers are installed at key process nodes such as raw ore, rougher concentrate, scavenger concentrate, total concentrate, rougher tailings, scavenger tailings, and total tailings. The samplers directly extract representative slurry samples from the main process pipeline.

[0026] (2) Sample transport and pretreatment: A slurry intelligent multi-channel divider is adopted, which is controlled by PLC program to realize the automatic switching of 10 slurry samples according to a predetermined time sequence. The multi-channel divider has a built-in buffer tank, slag screen and tuning fork level switch to ensure that the slurry flow rate entering the analyzer is stable, free of large particulate impurities and constant liquid level, and overcome the interference of slurry fluctuation on analysis.

[0027] (3) Core analytical equipment: An online flotation laser spectrometer is used as the core detection unit. Its key technical features include: (4) Dual-pulse LIBS technology: Two pulsed lasers with energy of 100 mJ each and frequency of 10 Hz are used to overcome the quenching effect of water on plasma, enhance signal strength, and improve signal-to-noise ratio.

[0028] (5) Automatic rinsing and calibration: The analyzer integrates an automatic rinsing system to clean the sample chamber and pipelines before and after each measurement to prevent cross-contamination. It is equipped with an online concentration meter to monitor the slurry concentration in real time for compensation and correction of spectral data.

[0029] (6) High-performance spectral acquisition: A high-resolution fiber optic spectrometer with a wavelength range of 207–430 nm is used to cover the characteristic spectral lines of phosphate rock grade targets such as P2O5, MgO, Fe2O3, Al2O3, SiO2, and CaO.

[0030] (7) Rapid detection: The single-channel single-analysis time (including sampling, measurement and cleaning) is ≤5 minutes, which meets the timeliness requirements of process control.

[0031] 2. Digital Twin Layer: A process performance prediction model that integrates mechanism and data-driven approaches. This layer is the core of the system and is responsible for transforming the sensed data into process control twin data.

[0032] (1) Multi-source data fusion: The digital twin synchronously collects and integrates LIBS spectral data and concentration data from the real-time sensing layer, as well as process parameters such as drug delivery frequency, flotation liquid level, and aeration volume from the factory DCS system.

[0033] (2) Hybrid modeling method: Mechanism basis: Based on prior knowledge of flotation kinetics, mineralogy, and fluid mechanics, a macroscopic framework for the process is constructed.

[0034] Data-driven core mathematical model: Establishing a grade prediction model based on machine learning. A hybrid algorithm combining radial basis function (RBF) neural network and partial least squares regression (PLSR) is employed to address the characteristics of different elements and channels.

[0035] Total input to the model: (1) In the formula, The excitation spectra collected by the online laser spectrometer for pre-treatment flotation and the grade comparison data analyzed offline.

[0036] Total model output: (2) In the formula, This is a complete set of elemental grades for the slurry at the current testing point.

[0037] RBF neural networks are used to process nonlinear and complex mapping relationships of major elements such as P, Mg, and Fe. They are effective at approximating nonlinear functions due to their fast convergence speed.

[0038] RBF neural networks consist of an input layer, hidden layers (radial base layers), and an output layer. (3) In the formula, This represents the feature state of each node in the hidden layer. The width of the basis functions. This represents feature similarity; the smaller the value, the higher the similarity. The degree of membership is denoted by 1. The closer the degree is to 1, the more likely the current slurry belongs to the j-th type of mineral sample.

[0039] (4) In the formula, Here, b is the weight value, and b is the bias term. This is the predicted output value for the grade of phosphorus, magnesium, and iron elements.

[0040] PLSR model: used to handle elements such as Al, Si, and Ca, and performs stably, especially in datasets with severe collinearity.

[0041] The PLSR model finds the direction with the maximum covariance between the high-dimensional spectral matrix X and the concentration matrix Y by simultaneously decomposing them: (5) In the formula, T is the spectral principal component score matrix, representing the basic features of the input data, and E is the input residual matrix, representing the main features extracted from the original spectral matrix X. The remaining part that cannot be explained by the model, P is the characteristic wavelength load matrix, corresponding to the characteristic emission peak position of the predicted element.

[0042] (6) In the formula, U is the score matrix corresponding to the output variable Y, and Q is the component correlation mapping coefficient. For laboratory analysis of elemental true values, This is the laboratory truth error matrix.

[0043] (7) In the formula, These are regression coefficients, corresponding to the changes in Al / Si / Ca grades. The output is the predicted grade values ​​of aluminum, silicon, and calcium elements. This is the error term in the regression prediction equation.

[0044] (3) Model functions: Online soft measurement: Outputs the element grade of each node in real time as a feedback signal for the control system.

[0045] Ultra-real-time forecasting: Based on current operating conditions, it predicts the trend of concentrate and tailings grades over a future period, providing a basis for forward-looking control.

[0046] Model self-updating: The system periodically compares laboratory test data with online prediction data. When the deviation exceeds the threshold, the model update mechanism is automatically triggered to use new data to incrementally learn the model and ensure long-term prediction accuracy.

[0047] 3. Intelligent Decision-Making Layer: Rule-Based and Optimization-Based Collaborative Control Strategies This layer is the system's command center, responsible for developing optimized control strategies.

[0048] (1) Expert rule base: Based on domain expert knowledge and historical best operation data, an IF-THEN rule base is constructed. For example: IF The predicted value of MgO grade in concentrate > the set upper limit AND The MgO grade of raw ore is stable, THEN Increase the amount of inhibitor added.

[0049] If the predicted P2O5 grade in the tailings exceeds the set upper limit, then check and adjust the collector dosage and flotation machine level.

[0050] (2) Multi-objective optimization algorithm: A multi-objective optimization model is established with the objectives of maximizing the P2O5 grade of concentrate, minimizing the MgO content of impurities, and minimizing reagent consumption. Combining the prediction function of the digital twin, model predictive control (MPC) or evolutionary algorithm is used to calculate the optimal setpoint sequence of operating variables for the next few control cycles.

[0051] (3) Intelligent identification and self-adaptation of operating conditions: Real-time analysis of parameters such as flotation machine power, stirring tank speed, aeration volume, and LIBS spectral signal stability to identify abnormal operating conditions such as overflow and settling. When an abnormality is detected, the system can automatically switch to the preset safety control mode or adjust the control parameters to ensure system safety.

[0052] 4. Automatic Execution Layer: This layer is responsible for the precise and reliable issuance and execution of instructions. It acts as the system's hands and feet, turning decisions into actions.

[0053] (1) Control command issuance: The optimized set values ​​(such as reagent addition amount, flotation machine speed, liquid level set value) generated by the intelligent decision layer are issued to the underlying DCS or PLC system through standard interfaces (such as OPC UA, Modbus TCP).

[0054] (2) Actuators: The underlying control system drives the corresponding actuators, such as frequency converters, metering pumps (to control reagents), pneumatic regulating valves (to control liquid level and air volume), etc., to complete the precise adjustment of production parameters.

[0055] (3) Closed-loop control loop: forming a complete closed loop. For example, the online LIBS analyzer detects an increase in the MgO grade of the concentrate → the digital twin model predicts the future trend → the intelligent decision layer calculates the increment of the inhibitor → the automatic execution layer adjusts the frequency of the metering pump → the amount of reagent added increases → the LIBS detection data of the next cycle verifies the control effect, and so on.

[0056] The invented intelligent decision-making layer adopts a hierarchical decoupling control strategy, constructing independent mathematical control models for three different working areas: coarse selection, fine selection, and sweep selection. 1. The main task of the roughing process is to suppress apatite (such as P2O5). To overcome the lag caused by fluctuations in the raw ore, the system adopts a control strategy combining feedforward compensation and feedback correction: (8) In the formula: Uinh(t) is the reagent addition pump frequency command (%) at time t, Craw(t) is the real-time content of impurities (Mg or Fe) in the raw ore detected online by LIBS, Cbase is the baseline reference value (design value) of the impurity content in the raw ore, Kff is the feedforward compensation coefficient, which represents the amount of inhibitor adjustment required per unit of raw ore impurity fluctuation, e(t) is the deviation between the measured value and the set target value of impurity content in the roughing concentrate, and KP and KI are the proportional coefficient and integral coefficient of the PID controller, respectively, used to eliminate steady-state error.

[0057] 2. The fine-grained process is extremely sensitive to collectors. An incremental gradient control algorithm is used to approach the optimal grade with small step sizes, avoiding the deterioration of concentrate quality caused by excessive reagent dosage. ; ; (9) In the formula: The optimal amount of collector added at time t. This is the control increment for this cycle. The P2O5 grade of the selected concentrate detected by LIBS. The target grade for P2O5 in the concentrate. The sign function determines the direction of increase or decrease: add drug when the grade is low, and reduce drug when the grade is high. Adjust the step size coefficient; The response index (usually taken as 0 < 0) <1), to prevent overshoot oscillation.

[0058] 3. The core of the scavenging operation is to reduce the grade of tailings. Nonlinear threshold gain control is employed; when the risk of tailings runoff is detected, the reagent dosage is increased exponentially.

[0059] The control formula is as follows: (10) In the formula: The amount of sweep inhibitor added at time t. This is the basic dosage of medication under normal operating conditions. The P2O5 grade of the roughing tailings (i.e., scavenging feed) detected by LIBS. Set the P2O5 grade for tailings. This is the radical gain coefficient (γ > 1).

[0060] When the tailings grade is lower than the set value At the same time, the system maintains a basic dosage to save costs; once the grade exceeds the standard, the dosage will increase rapidly according to the square of the deviation, achieving strong interception of useful minerals.

[0061] Example 1 Application Background This embodiment is applied to the Jinning Mineral Processing Branch of Yunnan Phosphate Group. The company processes 3 million tons of raw ore annually and adopts a process flow of reverse flotation roughing, reverse flotation cleaning I, reverse flotation cleaning II, and scavenging. The raw ore has a P2O5 grade of 22-25% and an MgO grade of 1.5-2.5%. The original production mode relied on manual sampling and testing every 2 hours, with a detection lag of up to 3 hours. The MgO qualification rate of the concentrate was only 80%, and the reagent consumption was 12 kg / ton of ore.

[0062] System Deployment Customized knife-edge samplers were installed at 11 key nodes: raw ore, rougher concentrate, rougher tailings, cleaner I concentrate, cleaner I tailings, cleaner II concentrate, cleaner II tailings, scavenger concentrate, scavenger tailings, total concentrate, and total tailings. A dedicated online control room was constructed. Two multi-channel reducers were installed on the upper level of the instrument working area, and an online laser spectrometer was installed on the lower level. All equipment was connected through a DN65 slurry delivery pipeline to form a complete detection link. The online laser spectrometer is equipped with a dual-pulse laser, a single-energy 100mJ, 10Hz, 207-430nm high-resolution fiber optic spectrometer, and a single-channel, single-analysis time of 4.5 minutes.

[0063] Model building and parameter setting After the system was running, a total of 12,000 valid data pairs were collected. An RBF neural network model was established for P, Mg, and Fe elements, with 25 hidden layer nodes, a basis function width σ=1.0, a learning rate of 0.03, and 1000 iterations. The output formula is as follows: Weight values ​​in the formula The value range is 0.05-0.12, and the bias term b=0.02. When the input spectral feature vector X and the center vector... When the Euclidean distance is 0.5, the radial basis function output is exp(-0.5). 2 / (2×1.0 2 ))=exp(-0.125)=0.882, substitute into the weight =0.08, the calculated contribution value of this node is 0.08×0.882=0.0706. After superimposing the contribution values ​​of all nodes and the bias term, the predicted value of P element grade is 23.5%.

[0064] A PLSR model was established for Al, Si, and Ca elements. The cumulative variance contribution rate of the eight principal components was 96.2%. The regression formula is as follows: In the formula, in the regression coefficient matrix B_{PLS}, Al corresponds to 0.08, Si corresponds to 0.06, and Ca corresponds to 0.07, and the error term... =0.01. When the value of a certain row in the preprocessed spectral data matrix X is 1.2, the predicted Al element grade is 1.2×0.08+0.01=0.096+0.01=0.106, which is 10.6%.

[0065] Control strategy implementation The intelligent decision-making layer adopts a hierarchical decoupling control strategy, and the control formulas and parameters for each work area are as follows: Coarse selection: Feedforward compensation coefficient Kff = 0.4, PID proportional coefficient KP = 1.3, integral coefficient KI = 0.15, control formula is: When the real-time MgO content of the raw ore =2.0%, benchmark value =1.8%, the deviation of MgO in the roughing concentrate e(t) = 0.1%, and the integral term ∫e(t)dt = 0.5, are calculated as follows: =0.4×(2.0-1.8)+1.3×0.1+0.15×0.5=0.08+0.13+0.075=0.285; That is, the pump frequency command for adding inhibitors is 28.5%.

[0066] Selected assignment: Adjust the step size coefficient α=0.9, the response index β=0.6, and the control formula is: When the real-time grade of P2O5 in the selected concentrate =30.5%, target grade =31.0%, the amount of collector added in the previous cycle When the flow rate is 7.2 L / h, the sign function outputs 1, |31.0-30.5|. 0.6 =0.5 0.6 ≈0.76, Δ =0.9×1×0.76≈0.68L / h, current cycle addition amount =7.2 + 0.68 = 7.88 L / h.

[0067] Scavenging operation: Basic reagent dosage Ubase = 6 L / h, aggressive gain coefficient γ = 1.8, tailings P2O5 setpoint Tsafe = 0.8%, control formula is: When the P2O5 grade of the tailings is scavenged When the value is 1.0%, it exceeds the set value, and the calculation is as follows: =6×1.8×(1.0-0.8) 2 =6×1.8×0.04=0.432L / h, meaning the current amount of sweep inhibitor added is 6.432L / h.

[0068] Running effect After 6 months of stable operation, the relative error of P2O5 was ≤4.2%, the relative error of MgO was ≤8.5%, and the relative errors of Fe2O3, Al2O3, SiO2, and CaO were all ≤9.3%. The P2O5 grade of the concentrate stabilized at 30-32%, the standard deviation of fluctuation decreased from 0.8 to 0.48, and the qualified rate of MgO content increased to 95.2%. The reagent consumption decreased to 11.04 kg / ton of ore, a reduction of 8%, the P2O5 loss rate of tailings decreased to 0.96%, the flotation recovery rate increased by 2.1 percentage points, and the annual economic benefits increased by approximately 12 million yuan. Unmanned automatic sampling and analysis were achieved, and the workload of the laboratory was reduced by 70%.

[0069] Example 2 Application Background This embodiment is applied to a medium-sized phosphate mine beneficiation plant in Hubei Province. The plant processes 800,000 tons of raw ore annually and adopts a three-stage flotation process of roughing, cleaning, and scavenging. The raw ore has a P2O5 grade of 19-23% and an MgO grade of 2.2-3.0%. The original production mode relied on manual sampling and testing every 3 hours, with a detection lag of up to 4 hours. The standard deviation of the concentrate P2O5 grade fluctuation was 0.95, the MgO qualification rate was only 75%, and the reagent consumption was 15 kg / ton of ore.

[0070] System Deployment Knife-edge samplers are installed at nine key nodes: raw ore inlet, rougher concentrate outlet, rougher tailings outlet, cleaner concentrate outlet, cleaner tailings outlet, scavenger concentrate outlet, scavenger tailings outlet, total concentrate outlet, and total tailings outlet. A multi-channel reducer is installed on the upper level of the instrument working area in the online control room, and an online laser spectrometer is installed on the lower level. All equipment is connected via a DN50 slurry delivery pipeline. The online laser spectrometer is equipped with a dual-pulse laser, a single-unit high-resolution fiber optic spectrometer with an energy of 100mJ and a frequency of 10Hz (207-430nm), and a single-channel, single-analysis time of 4.8 minutes.

[0071] Model building and parameter setting After the system was running, a total of 8000 valid data pairs were collected. An RBF neural network model was established for P, Mg, and Fe elements, with 20 hidden layer nodes, a basis function width σ=0.9, a learning rate of 0.025, and 800 iterations. The output formula is as follows: Weight values ​​in the formula The value range is 0.04-0.10, and the bias term b=0.015. When the input spectral feature vector X and the center vector... When the Euclidean distance is 0.4, the radial basis function output is exp(-0.4). 2 / (2×0.81))=exp(-0.16 / 1.62)≈exp(-0.0988)=0.906, substitute into the weight =0.07, the calculated contribution value of this node is 0.07×0.906=0.0634. After superimposing the contribution values ​​of all nodes and the bias term, the predicted value of Mg element grade is 2.3%.

[0072] A PLSR model was established for Al, Si, and Ca elements. The cumulative variance contribution rate of the six principal components was 95.1%. The regression formula is as follows: In the formula, the regression coefficient matrix Al corresponds to 0.07, Si to 0.05, and Ca to 0.06; error term =0.012. When the value of a certain row in the preprocessed spectral data matrix X is 1.1, the predicted Si grade is 1.1×0.05+0.012=0.055+0.012=0.067, or 6.7%.

[0073] Control strategy implementation The intelligent decision-making layer adopts a hierarchical decoupling control strategy, and the control formulas and parameters for each work area are as follows: Coarse selection: Feedforward compensation coefficient Kff = 0.35, PID proportional coefficient KP = 1.2, integral coefficient KI = 0.12, control formula is: When the real-time MgO content of the raw ore =2.3%, benchmark value =2.0%, the deviation of MgO in the roughing concentrate e(t) = 0.15%, and the integral term ∫e(t)dt = 0.4, the calculation yields: Uinh(t) = 0.35 × (2.3 - 2.0) + 1.2 × 0.15 + 0.12 × 0.4 = 0.105 + 0.18 + 0.048 = 0.333, which means the inhibitor addition pump frequency command is 33.3%.

[0074] Selected assignment: Adjust the step size coefficient α = 0.8, the response index β = 0.55, and the control formula is: When the real-time grade of P2O5 in the selected concentrate =29.5%, target grade =30.0%, the amount of collector added in the previous cycle When the flow rate is 6.5 L / h, the sign function outputs 1, |30.0-29.5|. 0.55 =0.5 0.55 ≈0.74, =0.8×1×0.74≈0.59L / h, current cycle addition amount =6.5 + 0.59 = 7.09 L / h.

[0075] Sweeping and selection operation: Basic drug dosage =5L / h, radical gain coefficient γ=1.6, tailings P2O5 setpoint =0.9%, the control formula is: When the P2O5 grade of the tailings is scavenged When the value is 1.1%, it exceeds the set value. The calculation is as follows: = 5 × 1.6 × (1.1 - 0.9) 2 =5×1.6×0.04=0.32L / h; That is, the current addition rate of the sweep inhibitor is 5.32 L / h.

[0076] Running effect After four months of stable operation, the relative error of P2O5 was ≤4.8%, the relative error of MgO was ≤9.2%, and the relative errors of other elements were all ≤9.8%. The P2O5 grade of the concentrate stabilized at 28-30%, and the standard deviation of fluctuation decreased from 0.95 to 0.57, a reduction of 40%. The qualified rate of MgO content increased to 92.5%. The reagent consumption decreased to 13.2 kg / ton of ore, a reduction of 12%. The P2O5 loss rate of tailings decreased to 1.05%, and the flotation recovery rate increased by 1.8 percentage points, resulting in an annual increase in economic benefits of approximately 1.8 million yuan. The system deployment cost was reduced by 30% compared to large-scale plants, and the worker training cycle was ≤3 days, making it suitable for the technical personnel configuration needs of small and medium-sized plants.

[0077] The advantages of the intelligent auxiliary control system and method for phosphate rock flotation process based on LIBS online detection technology proposed in this invention are as follows: This invention, by deploying a real-time sensing layer, enables real-time online monitoring of multiple nodes and elements in the phosphate rock flotation process. The single-channel, single-analysis time (including sampling, measurement, and cleaning) does not exceed 5 minutes. This overcomes the hour-long lag caused by traditional production methods that rely on manual timed sampling, sample preparation, and laboratory analysis, shifting production control from reactive remediation to proactive prevention and in-process regulation. The digital twin layer, based on a grade prediction model built from real-time sensing data, can perform online soft measurement of key process indicators and predict future trends in real time, providing precise data support for the intelligent decision-making layer, thereby significantly improving the timeliness and accuracy of production control.

[0078] This invention, through closed-loop control, can rapidly respond to fluctuations in raw ore properties and changes in equipment operating conditions, significantly reducing the fluctuation range of P2O5 grade and impurity content in concentrate products, thereby improving product quality stability and pass rate. Examples show that the relative error between predicted and laboratory values ​​can be controlled within P2O5≤5%, MgO≤10%, Fe2O3≤10%, Al2O3≤10%, and SiO2≤10%, effectively improving concentrate quality. Simultaneously, precise control of reagent addition avoids reagent waste, reduces production costs, and optimizes recovery rates to reduce phosphorus resource loss in tailings, improving the overall utilization of resources. After the system was put into operation, the standard deviation of concentrate P2O5 grade fluctuation decreased by approximately 40%, the MgO content pass rate increased from approximately 80% before operation to over 95%, average reagent consumption decreased by approximately 8%, and flotation recovery rate increased by approximately 2 percentage points.

[0079] The intelligent decision-making layer of this invention adopts a hierarchical decoupled control strategy, combining an expert rule base and a multi-objective optimization algorithm. It can generate optimal operational variable setpoints based on the characteristics of different work areas, significantly reducing reliance on operator experience and improving the automation and intelligence level of production management. The automatic execution layer precisely sends optimization instructions to the field actuators, forming a complete closed-loop control circuit, realizing unattended operation and precise control of the production process. Furthermore, through digital twins and remote monitoring cloud services, this invention achieves complete transparency of the production process. Managers can remotely monitor the production status and equipment health in real time, facilitating fault diagnosis and system maintenance, and significantly reducing operation and maintenance costs and labor intensity.

[0080] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology, characterized in that, include: The real-time sensing layer is used to automatically sample and perform multi-element online laser spectral analysis of the slurry at multiple process nodes in the phosphate rock flotation process to obtain data on grade changes in phosphate rock flotation operations. The digital twin layer, based on the spectral data and production process data output by the real-time sensing layer, constructs and runs a process index prediction model to achieve soft measurement and trend prediction of grade. The intelligent decision-making layer, based on the output of the digital twin layer, combines expert rules and optimization algorithms to generate optimized control instructions for the parameters of each operational variable in the flotation process; The automatic execution layer connects to the underlying distributed control system via an industrial communication protocol, and sends the optimized control commands to the field actuators to complete the closed-loop optimized control of the flotation process.

2. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 1, characterized in that, The real-time sensing layer includes: Multiple knife-edge samplers (1) are deployed at key process nodes such as raw ore, rougher concentrate, scavenger concentrate, total concentrate, rougher tailings, scavenger tailings, and total tailings to directly extract representative slurry samples from the original process pipeline (5). The slurry conveying pipeline (2) is used to connect the knife-edge sampler (1) and the multi-channel divider (3) to achieve stable conveying of slurry samples; The multi-channel reducer (3) is connected to the slurry conveying pipeline (2), and has a built-in buffer tank, slag screen and tuning fork level switch. It is controlled by PLC program to realize the automatic switching of multiple slurry samples according to a predetermined time sequence, so as to ensure that the slurry flow entering the online laser spectrometer (4) is stable, free of large particle impurities and constant liquid level. An online laser spectrometer (4) is connected to the multi-channel reducer (3) and uses dual-pulse laser-induced breakdown spectroscopy to quantitatively analyze the P2O5, MgO, Fe2O3, Al2O3, SiO2, and CaO elements in the slurry.

3. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 2, characterized in that, The technical parameters of the online laser spectrometer (4) are as follows: it uses two pulsed lasers; it integrates an automatic rinsing system to clean the sample chamber and pipelines before and after each measurement; it is equipped with an online concentration meter to monitor the slurry concentration in real time for spectral data compensation and correction; it uses a high-resolution fiber optic spectrometer with a wavelength range of 207 to 430 nm; and the single-channel single-analysis time, including sampling, measurement, and cleaning, is ≤5 minutes.

4. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 1, characterized in that, The process index prediction model used in the digital twin layer is a hybrid model that combines a mechanism model and a data-driven model. The data-driven model is a combination of radial basis function (RBF) neural network and partial least squares regression (PLSR). The RBF neural network is used to predict the grades of P, Mg, and Fe elements, and its output formula is: ； In the formula, Here, b is the weight value, and b is the bias term. The output values ​​are the predicted grades of phosphorus, magnesium, and iron, representing the characteristic states of the hidden layer nodes. The width of the basis functions. For feature similarity; The PLSR model is used to predict the grades of Al, Si, and Ca elements. Its regression formula is as follows: In the formula, This is the preprocessed spectral data matrix. For regression coefficients, These are the predicted grades of Al, Si, and Ca elements. This represents the regression prediction error term.

5. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 1, characterized in that, The process performance prediction model of the digital twin layer has the following functions: The online soft measurement function outputs the current elemental grades of P2O5, MgO, Fe2O3, Al2O3, SiO2, and CaO at each process node in real time, as feedback signals for the control system. The ultra-real-time prediction function predicts the trend of changes in concentrate grade and tailings grade over a period of time based on current operating conditions. The model self-update function periodically compares laboratory test data with online prediction data. When the deviation exceeds a threshold, the incremental learning mechanism is automatically triggered to update the model parameters.

6. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 1, characterized in that, The intelligent decision-making layer includes: An expert rule base stores IF-THEN control rules constructed based on process knowledge. These rules include: If the predicted MgO grade in the concentrate is greater than the set upper limit AND the MgO grade in the raw ore is stable, then increase the amount of inhibitor added. Or, if the predicted P2O5 grade in the IF tailings exceeds the set upper limit, THEN adjust the collector dosage and flotation machine level. The multi-objective optimization unit takes maximizing the P2O5 grade of the concentrate, minimizing the MgO content of impurities, and minimizing reagent consumption as optimization objectives. It calculates the optimal setpoint sequence of operating variables by combining the prediction results of the digital twin layer.

7. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 2, characterized in that, The intelligent decision-making layer also includes a working condition identification unit, which identifies abnormal working conditions such as overflow and settling by analyzing the flotation machine power, stirring tank speed, aeration volume, and spectral signal stability parameters of the online laser spectral analyzer (4) in real time; when an abnormality is identified, it automatically switches to the preset safety control mode or adjusts the control parameters.

8. The intelligent auxiliary control system for phosphate rock flotation process based on LIBS online detection technology according to claim 2, characterized in that, The operational variables controlled by the automatic execution layer include the amount of collector added, the amount of inhibitor added, the flotation machine level, and the aeration rate; the industrial communication protocols include the OPC UA protocol and the Modbus TCP protocol; the real-time sensing layer constructs a closed-loop sampling channel from the original process pipeline (5) to the online laser spectrometer (4) through the slurry conveying pipeline (2).

9. A method for intelligent auxiliary control of the phosphate rock flotation process based on the system described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Data is collected through the real-time sensing layer. Specifically, the slurry samples of each process node are cut from the original process pipeline (5) using the knife-edge sampler (1), and transported to the multi-channel divider (3) through the slurry conveying pipeline (2) for flow stabilization, impurity removal and time sequence switching. Then, the samples are sent to the online laser spectrometer (4) to complete the acquisition of multi-element laser spectrometer data and simultaneously collect slurry concentration data. S2: Through the digital twin layer, based on the spectral data, slurry concentration data and real-time production process data output by the online laser spectrometer (4), the current and future element grades are predicted using the process index prediction model; the digital twin layer periodically uses laboratory test data to perform online calibration and update of the process index prediction model; S3: Through the intelligent decision-making layer, the predicted value of grade is compared with the target value, and the adjustment instructions of the operation variables are generated by using expert rules and multi-objective optimization algorithms. S4: The adjustment command is sent to the actuator through the automatic execution layer. Combined with the real-time data fed back by the sampling and detection link formed by the slurry conveying pipeline (2) and the multi-channel divider (3) through the real-time sensing layer, the closed-loop optimization control of the flotation process is realized.

10. The intelligent auxiliary control method for phosphate rock flotation process based on LIBS online detection technology according to claim 9, characterized in that, In step S3, the intelligent decision-making layer adopts a hierarchical decoupling control strategy for the coarse selection, fine selection, and sweeping operation areas: The coarse selection operation employs a control strategy combining feedforward compensation and feedback correction. The control formula is as follows: In the formula, Add a pump frequency command (%) to the inhibitor at time t. This represents the real-time content of impurities in the raw ore. This serves as a benchmark reference value for the impurity content of the raw ore. Forward compensation coefficient, This represents the deviation between the measured and target values ​​of impurity content in the roughing concentrate. , These are the proportional and integral coefficients of the PID controller, respectively. The selected assignments employ an incremental gradient control algorithm, with the control formula as follows: ； ； In the formula, The optimal amount of collector added at time t. To control the increment in this cycle, To refine the P2O5 grade of the concentrate, The target grade of P2O5 in the concentrate is [missing information]. To adjust the step size coefficient, The response exponent is 0 < β < 1; The sweeping operation employs nonlinear threshold gain control, and the control formula is as follows: ； In the formula, The amount of sweep inhibitor added at time t. This is the basic dosage of medication under normal operating conditions. To improve the P2O5 grade of the tailings, Set the P2O5 grade for tailings. γ is the radical gain coefficient, where γ > 1.

Images