Ore crushing-intelligent pre-concentration separation integrated system based on intelligent linkage control and control method

Abstract

The application discloses an ore crushing-intelligent pre-concentration separation integrated system and control method based on intelligent linkage control, and belongs to the field of automatic control technology. The system comprises a mine buffer bin and a mine receiving unit, a mine crushing unit and an intelligent detection and separation unit which are sequentially connected, and a pre-concentration storage and conveying unit and a tailing on-site disposal unit connected with the separation unit. The intelligent control system comprises an intelligent linkage control module and a separation parameter collaborative optimization module. A multi-modal deep fusion identification module is arranged in the intelligent detection and separation unit, and the network structure and classification threshold value of the multi-modal deep fusion identification module can be adaptively adjusted according to the belt speed. The control method is based on the integrated system and realizes linkage control. The application realizes material balance linkage of the whole process of underground crushing and separation and adaptive collaboration of separation parameters, improves the system operation efficiency and equipment utilization rate, solves the problem that the fluctuation of the belt speed leads to the decrease of the separation precision, and has remarkable resource saving and safety benefits.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, specifically to an integrated system and control method for ore crushing-intelligent pre-enrichment and sorting based on intelligent linkage control. Background Technology

[0002] Traditional mining typically employs an "underground mining, surface beneficiation" operation model. This involves transporting raw ore from the underground mining face to the surface via a hoisting system, where it undergoes crushing, sorting, and pre-enrichment processing at a surface beneficiation plant. In this model, the raw ore often contains a large amount of waste rock (20%–40%), which is hoisted and transported to the surface along with the ore. This significantly increases the load on hoisting equipment and transportation energy consumption, as well as raising the processing costs and tailings storage fees at the surface beneficiation plant. Furthermore, the tailings generated during surface beneficiation require dedicated tailings dams, posing ecological and environmental risks such as land occupation, leakage, and dam failure. In addition, during long-distance hoisting and transfer, the ore is prone to mudification and breakage, leading to the loss of some target minerals and further exacerbating resource waste.

[0003] To reduce waste rock lifting costs and alleviate environmental pressure on the surface, some mines have begun to try moving some mineral processing steps underground. For example, Chinese invention patent application CN118060039A discloses a modular mineral processing equipment system adapted to underground mining, beneficiation, and backfilling. This system adopts a modular design, arranging screens, crushing modules, screening modules, pre-selection modules, ore storage modules, roller mill modules, and sorting modules underground. After the raw ore is crushed and screened, it enters the pre-selection module to discharge waste rock, which is directly transported to the goaf for backfilling. The pre-selected concentrate is then further processed by roller milling and sorting to obtain the final concentrate, initially realizing the integrated operation of "crushing-benefiting-enrichment-waste rock backfilling" underground.

[0004] However, existing integrated underground mining, beneficiation, and filling solutions still have certain limitations. On the one hand, there is a lack of effective collaborative control mechanisms between various work units. For example, when the level of the buffer ore bin fluctuates, the operating speed of the sorting unit cannot be adaptively adjusted, often relying on manual intervention, which not only affects work efficiency but also brings additional safety risks to underground workers. On the other hand, existing sorting units mostly use a combination of XRT online detection and image recognition for ore classification. However, this type of detection method is quite sensitive to the belt conveyor speed. When the belt speed changes due to system linkage adjustments, the XRT detection time window will be affected, leading to distorted detection data. If the sorting model still uses fixed parameters, it is difficult to maintain stable sorting accuracy, especially under high-volume conditions where a significant decrease in accuracy is likely to occur. Summary of the Invention

[0005] This invention proposes an integrated system and control method for ore crushing-intelligent pre-enrichment and sorting based on intelligent linkage control. Its purpose is to overcome the shortcomings of existing integrated underground mining, beneficiation and filling schemes, such as poor unit coordination, reliance on manual intervention, low operating efficiency and safety hazards. At the same time, it solves the technical problems of distorted detection data caused by belt conveyor speed fluctuations and poor adaptability of sorting models that affect sorting accuracy.

[0006] The technical solution of this invention is as follows:

[0007] An integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control includes a buffer ore bin and ore receiving unit set underground, an underground crushing unit and a sorting unit, a pre-enrichment storage and conveying unit and a tailings on-site disposal unit set underground, and an intelligent control system for intelligent linkage control of each unit.

[0008] The buffer ore bin is sequentially connected to the ore receiving unit, the underground crushing unit, and the sorting unit. The sorting unit is further connected to the pre-enrichment storage and transportation unit and the tailings on-site disposal unit, respectively. The sorting unit is an intelligent detection and sorting unit.

[0009] The intelligent control system includes an intelligent linkage control module and a sorting parameter collaborative optimization module;

[0010] The intelligent linkage control module is used to coordinate and control the feeding speed of the vibrating feeder of the buffer ore bin and the receiving unit and the belt conveyor speed of the automated sorting actuator of the intelligent detection and sorting unit according to the operating status and material buffer status of each link of the system.

[0011] The sorting parameter collaborative optimization module is used to optimize the belt conveyor speed after the intelligent linkage control module adjusts the belt conveyor speed of the intelligent detection and sorting unit, and to send a speed signal to the intelligent detection and sorting unit to control the intelligent detection and sorting unit to adjust adaptively.

[0012] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, the intelligent linkage control module adopts a coordinated control strategy based on model predictive control, specifically including:

[0013] (1) Define variables

[0014] State variables include:

[0015] : The buffer ore bin level between the buffer ore bin and the receiving unit, in tons;

[0016] : The level of pre-enriched ore bins in the pre-enriched storage and conveying unit at all times, in tons;

[0017] The known input variables include:

[0018] : The raw ore flow rate of the mining conveying unit input buffer ore bin is measured in tons per hour, and the mining scheduling system provides the planned sequence within the future predicted time domain.

[0019] : The ore flow rate output from the conveying equipment to the ground is constantly increased, in tons per hour, and the ground dispatch system provides a planned sequence within the future forecast time domain based on the processing needs of the ore processing plant;

[0020] Control variables include:

[0021] : The feed flow rate of the vibrating feeder from the buffer ore bin to the underground crushing unit is measured in tons per hour.

[0022] : The incremental feed flow rate, expressed in tons per hour, is the decision variable for model predictive control.

[0023] : The desired belt conveyor speed of the sorting unit is constantly detected and measured in meters per second, which is uniquely determined by the feed flow rate through speed matching constraints.

[0024] System parameters include:

[0025] Control cycle;

[0026] : Overall output rate, which represents the proportion of feed flow rate that finally enters the pre-enriched ore bin after crushing, grading and intelligent sorting, is determined based on ore grade characteristics and historical operating data;

[0027] : Qualified block size output rate, which represents the proportion of the feed flow that enters the intelligent detection and sorting unit after crushing and grading, and is preset according to the characteristics of the ore;

[0028] : Bulk density of ore, in tons per cubic meter;

[0029] : Effective load-bearing cross-sectional area of ​​the belt, in square meters;

[0030] (2) Establish the state space model of system material flow.

[0031] The dynamic equation for the buffer ore bin level is:

[0032]

[0033] In the above formula, for The buffer ore level at any given time can be recursively calculated using this formula. Buffer ore bin level at any time , It is a positive integer;

[0034] The dynamic equation for the pre-enriched ore bin level is:

[0035]

[0036] In the above formula, for The pre-enriched ore bin level at any given time can be recursively calculated using this formula. Pre-enriched ore bin level at any time , It is a positive integer;

[0037] (3) Define the relationship between feed flow rate and feed flow rate increment.

[0038] For any The feed flow rate at any given time is calculated recursively using the feed flow rate increment:

[0039]

[0040] Within the prediction time domain, given the initial feed flow rate and incremental sequence It can recursively calculate the feed flow rate at each future time point:

[0041]

[0042] In the above formula, To predict the length of the time domain;

[0043] (4) Define velocity matching constraints

[0044] The relationship between belt conveyor speed and feed flow rate is as follows:

[0045]

[0046] (5) Obtain the predicted input quantity

[0047] The intelligent linkage control module obtains the planned sequence of raw ore input flow in the future predicted time domain from the mining scheduling system:

[0048]

[0049] The intelligent linkage control module obtains the planned sequence of increased output flow in the future predicted time domain from the ground dispatch system:

[0050]

[0051] (6) Define the optimization objective of model predictive control and construct the optimization objective function of the model predictive controller;

[0052] (7) Define control constraints

[0053] Material level constraints:

[0054]

[0055]

[0056] In the above formula, To buffer the maximum capacity of the ore bin, This is the maximum capacity of the pre-enriched ore bin;

[0057] Feed flow rate constraints:

[0058]

[0059] In the above formula, and These are the minimum and maximum feed flow rates of the vibrating feeder, respectively.

[0060] Belt conveyor speed constraints:

[0061]

[0062] In the above formula, and These are the lower and upper limits of belt conveyor speed, respectively.

[0063] Feed flow rate change constraint:

[0064]

[0065] In the above formula, This represents the maximum allowable change in feed flow rate within a single control cycle.

[0066] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, the optimization objective function is:

[0067]

[0068] In the above formula:

[0069] To predict the time domain length, the value ranges from 10 to 30 control cycles;

[0070] To control the time domain length, the value ranges from 3 to 10 control cycles, and ;

[0071] To buffer the target material level in the ore bin, it is typically set to 40% to 60% of the ore bin's capacity.

[0072] The target material level for the pre-enriched ore bin is typically set at 40% to 60% of the bin's capacity.

[0073] , , These are weighting coefficients used to balance the relative importance of various optimization objectives;

[0074] The decision variable is the sequence of feed flow rate increments. ,total There are several decision variables; when the prediction time exceeds the control time domain, i.e. season .

[0075] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, the intelligent linkage control module performs the following steps in each control cycle:

[0076] Step 1: Collect the current buffer ore bin level Pre-enriched ore bin material level and current feed flow rate ;

[0077] Step 2: Obtain the future raw ore input flow plan sequence from the mining scheduling system. Obtain the future output flow plan sequence from the ground dispatch system. ;

[0078] Step 3: Based on the state-space model, constraints, and optimization objective function, a quadratic programming algorithm is used to solve for the optimal feed flow rate increment sequence.

[0079]

[0080] Step 4: Calculate the target feed flow rate for the current control cycle based on the first optimal increment:

[0081]

[0082] Step 5: Set the target feed flow rate Send to the vibrating feeder for execution;

[0083] Step 6: Calculate the corresponding target belt conveyor speed based on the speed matching constraints:

[0084]

[0085] Step 7: Set the target conveyor belt speed The data is sent to the sorting parameter collaborative optimization module, triggering adaptive adjustment of the sorting parameters;

[0086] Step 8: Enter the next control cycle and repeat the above steps to achieve rolling time-domain optimization control.

[0087] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, the sorting parameter collaborative optimization module establishes a rate-accuracy prediction model, the expression of which is:

[0088]

[0089] In the above formula, For belt conveyor speed The predicted sorting accuracy is as follows. For the highest sorting accuracy, For accuracy attenuation coefficient, The preset optimal belt conveyor speed;

[0090] When the intelligent linkage control module calculates the target belt conveyor speed This leads to changes in predictive sorting accuracy. Below the threshold At the same time, the sorting parameter collaborative optimization module corrects the target belt conveyor speed:

[0091]

[0092] In the above formula, The corrected target belt conveyor speed. To meet the maximum allowable rate constrained by sorting accuracy:

[0093]

[0094] If no correction is needed, then directly set... ;

[0095] Then press Control the belt operation of the intelligent detection and sorting unit, and The data is sent to the rate adaptive parameter regulator and classification decision layer of the intelligent detection and sorting unit.

[0096] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control: the intelligent detection and sorting unit includes an intelligent detection module, the multimodal deep fusion recognition module, and the automated sorting execution mechanism;

[0097] The intelligent detection module is used to collect multi-source detection data of the ore to be sorted in real time; the multimodal deep fusion recognition module is used to receive the data collected by the intelligent detection module, and perform fusion analysis and recognition based on a deep neural network model to output the ore category determination result; the automated sorting execution mechanism is used to perform physical sorting actions according to the determination result output by the multimodal deep fusion recognition module. It includes the belt sorter and the solenoid valve injection mechanism. The belt conveying speed of the belt sorter is given by the sorting parameter collaborative optimization module in the intelligent control system after calculation based on the overall material balance state of the system; the intelligent detection module collects data from the ore on the belt sorter.

[0098] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control: the intelligent detection module includes an XRT online detector and a high-definition industrial camera;

[0099] The XRT online detector is used to collect ore elemental composition data, ore density data, and ore spot condition data. The elemental composition data includes the target element content value and element distribution uniformity index; the density data is the ore bulk density calculated based on the X-ray transmission attenuation coefficient; and the ore spot condition data includes ore spot distribution density and ore spot average size.

[0100] The high-definition industrial camera is used to acquire RGB images of the ore surface;

[0101] The multimodal deep fusion recognition module adopts a dual-branch deep neural network architecture, including an XRT feature extraction branch and an image feature extraction branch;

[0102] In the XRT feature extraction branch, elemental composition data, density data, and mineral patch state data are combined to form an input vector. XRT feature vectors are extracted through a multi-layer fully connected network. :

[0103]

[0104] In the above formula, ,in The target element content value, For elemental distribution uniformity, Density of the ore The density of mineral patches. This represents the average size of the mineral deposits; , , Here are the weight matrices for each layer. , , For each layer's bias vector, For activation functions;

[0105] The image feature extraction branch uses a convolutional neural network to process the input image. The image is processed to extract image feature vectors containing texture information. :

[0106]

[0107] In the above formula, Indicates the first Layer convolution operation, The number of convolutional layers. This indicates a global pooling operation.

[0108] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, the multimodal deep fusion recognition module also includes a cross-modal attention fusion layer;

[0109] The cross-modal attention fusion layer is used to perform deep fusion of XRT feature vectors and image feature vectors; specifically, using features from one modality as query vectors and features from another modality as key-value pairs, cross-modal attention weights are calculated.

[0110]

[0111]

[0112] In the above formula, The attention weights are calculated using XRT features as the query pair image features. The attention weights are calculated for XRT features using image features as the query. and , and These are the projection matrices of the query and the key in two directions, respectively, and are learnable parameters; is the dimension of the key vector.

[0113] As a further improvement to the integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, the multimodal deep fusion recognition module also includes a rate adaptive parameter regulator and a classification decision layer.

[0114] The rate adaptive parameter regulator receives the belt conveyor rate sent by the intelligent control system. Calculate the rate influence factor :

[0115]

[0116] In the above formula, For nominal transport speed, This is the adjustment coefficient;

[0117] Then calculate the rate-adaptive fusion features. :

[0118]

[0119] In the above formula, and For rate-related residual weights:

[0120]

[0121]

[0122] in, The projection matrix is ​​the value of , and the parameters are learnable.

[0123] The classification decision layer receives the rate-adaptive fusion features. The probability distribution of ore categories is output through a fully connected layer and a softmax function:

[0124]

[0125] In the above formula, This is the classification weight matrix. As the bias vector, the ore category probability distribution includes two categories: target minerals and waste rock;

[0126] Meanwhile, the classification decision layer determines the belt conveyor speed based on the information sent by the intelligent control system. Dynamically adjust classification threshold :

[0127]

[0128] In the above formula, Basic classification threshold; The threshold adjustment range; Adjust the rate coefficient for the threshold; This represents the system's minimum transport rate.

[0129] If the probability of the target mineral category in the ore category probability distribution is greater than the threshold If the ore is found to be the target mineral, then it is determined to be waste rock; otherwise, it is determined to be waste rock.

[0130] This invention also provides a control method for an integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, as described above, comprising the following steps:

[0131] Step 1: Continuously collect on-site data;

[0132] Step 2: Intelligent linkage and coordination control, specifically including:

[0133] Step 2.1: The intelligent linkage control module operates at a fixed cycle. Perform coordination and control operations;

[0134] Step 2.2: At the beginning of each control cycle, obtain the current material level of each ore bin. , and current feed flow rate ;

[0135] Step 2.3: Obtain the future prediction time domain from the mining scheduling system The raw ore input flow planning sequence Obtain future prediction time domain from ground dispatch system Internal increase output flow plan sequence ;

[0136] Step 2.4: Based on the material flow state-space model, constraints, and optimization objective function, a quadratic programming algorithm is used to obtain the optimal feed flow increment sequence. ,in To control the length of the time domain;

[0137] Step 2.5: Calculate the target feed flow rate for the current control cycle based on the first optimal increment:

[0138]

[0139] Step 2.6: Set the target feed flow rate Send to the vibrating feeder for execution;

[0140] Step 2.7: Calculate the corresponding target belt conveyor speed based on the speed matching constraints:

[0141]

[0142] Step 2.8: Set the target belt conveyor speed Send to the sorting parameter collaborative optimization module;

[0143] Step 3: Coordinated adjustment of sorting parameters;

[0144] Step 3.1: The sorting parameter collaborative optimization module receives the target belt conveyor speed of the intelligent detection and sorting unit calculated by the intelligent linkage control module, denoted as... ;

[0145] Step 3.2: Calculate the predicted sorting accuracy at this speed based on the rate-accuracy prediction model. ;

[0146] Step 3.3: Determine whether the sorting accuracy meets the constraints and make corrections accordingly to obtain the corrected belt conveyor speed of the intelligent detection and sorting unit. ;

[0147] Step 3.4: Press Control the belt operation of the intelligent detection and sorting unit, and The data is sent to the rate adaptive parameter regulator and classification decision layer of the intelligent detection and sorting unit.

[0148] Step 4: Intelligent detection and sorting unit adaptive recognition, specifically including:

[0149] Step 4.1: The intelligent detection module continuously collects ore data passing through the belt separator, including elemental composition data, density data, and ore spot condition data collected by the XRT online detector, as well as RGB images of the ore surface collected by a high-definition industrial camera. ;

[0150] Step 4.2: The rate adaptive parameter regulator adjusts the speed according to the received belt conveyor speed. Calculate the rate influence factor ;

[0151] Step 4.3: The XRT feature extraction branch extracts features from the XRT detection data to obtain the XRT feature vector. ;

[0152] Step 4.4: The image feature extraction branch extracts features from the ore surface image to obtain the image feature vector. ;

[0153] Step 4.5: The cross-modal attention fusion layer adaptively weights and fuses the features of the two modalities based on the current rate influence factor to obtain the rate-adaptive fusion features. And calculate the probability distribution of ore categories based on rate adaptive fusion features. ;

[0154] Step 4.6: The classification decision layer calculates the adaptive classification threshold based on the current belt conveyor speed. ;

[0155] Step 4.7: If the probability distribution of ore types... The probability of the target mineral category being greater than the threshold If the ore is positive, it is determined to be the target mineral; otherwise, it is determined to be waste rock.

[0156] Step 5: Sorting and Material Diversion: The automated sorting execution mechanism controls the solenoid valve injection mechanism based on the judgment results of the multimodal deep fusion identification module. For ores identified as target minerals, the solenoid valve injection mechanism does not operate, and the ore is normally transported along the belt to the pre-enrichment storage and conveying unit. For ores identified as waste rock, the solenoid valve injection mechanism operates, and the waste rock is injected and diverted to the tailings on-site disposal unit.

[0157] Compared with the prior art, the present invention has the following beneficial effects:

[0158] 1. This invention addresses the problems of poor unit coordination and reliance on manual intervention in existing integrated underground mining, beneficiation, and filling systems. By introducing an intelligent control system, it achieves coordinated control of all equipment throughout the entire process, including the buffer ore bin, receiving unit, underground crushing unit, and intelligent detection and sorting unit. Specifically, the system uses an intelligent linkage control module to collect the material level status of each ore bin in real time. Combined with planning information from the mining and surface scheduling systems, it employs a model predictive control-based coordinated control strategy to continuously optimize the feeding speed of the vibrating feeder and the belt conveyor speed of the sorting unit. Compared to traditional methods that rely on manual experience for adjustment, this mechanism can proactively respond to fluctuations in upstream ore supply and downstream hoisting demands, automatically maintaining the material levels in the buffer and pre-enrichment ore bins within a safe and reasonable range. This avoids the risk of downtime due to excessively high material levels or idle capacity due to excessively low material levels, thereby significantly improving the overall operating efficiency and equipment utilization rate of the system while ensuring the continuity and safety of underground operations.

[0159] 2. This invention constructs a precise material flow state-space model, incorporating upstream mining plans and downstream hoisting demands as known inputs into the model's predictive control framework. This strategy not only considers current operating conditions but also proactively responds to fluctuations in incoming ore and hoisting plans over a future period, making the adjustment of the vibrating feeder and belt speed more predictable and comprehensive. Compared to the traditional method of feedback adjustment based solely on current material level deviation, this rolling optimization mechanism incorporating feedforward information can more smoothly respond to system disturbances, effectively suppressing large fluctuations in material level, thereby further improving the stability and anti-interference capability of the entire process operation.

[0160] 3. In the objective function of model predictive control, this invention not only includes a penalty term for tracking the target values ​​of the material levels in the two ore bins, but also specifically introduces a penalty term for the rate of change of feed flow rate. By balancing the level deviation with the severity of the control action in the objective function, the controller can automatically generate a gradually changing feed flow trajectory while maintaining stable material levels. This method effectively avoids mechanical shocks and process disturbances to downstream crushing, screening, and sorting equipment caused by sudden changes in feed rate, extends equipment lifespan, and ensures the stability and reliability of system operation under continuous variable operating conditions.

[0161] 4. This invention, while achieving speed linkage adjustment, specifically addresses the core problem of decreased sorting accuracy caused by frequent changes in belt conveyor speed. The sorting parameter collaborative optimization module in the intelligent control system establishes a speed-accuracy prediction model, performing secondary verification and correction on the target speed issued by the linkage control module. When the target speed might cause the sorting accuracy to fall below a preset threshold, this module can automatically calculate and execute a maximum permissible speed that meets the accuracy requirements, thus achieving a dynamic balance between system throughput and sorting quality. This collaborative optimization mechanism ensures that even when the entire system adjusts its speed according to material balance needs, the intelligent detection and sorting unit can still operate stably within an acceptable accuracy range, overcoming the poor adaptability of traditional fixed-parameter sorting models under varying operating conditions.

[0162] 5. This invention incorporates a multimodal deep fusion recognition module and a rate adaptive mechanism within the intelligent detection and sorting unit, enabling the sorting unit to better adapt to rate changes caused by system linkage. Specifically, this module not only deeply mines the complementary features of XRT detection data (such as element content and density) and high-definition image data (such as texture and mineral spots) through a cross-modal attention fusion layer, improving the recognition ability of complex ores and boundary-grade ores; it also introduces a rate adaptive parameter regulator, which dynamically adjusts the fusion weights of cross-modal features and the judgment threshold of the classification decision layer according to the current conveyor belt speed. When the conveyor belt speed deviates from the nominal value, the recognition model can adaptively strengthen the feature contribution of the less rate-affected modalities and appropriately adjust the classification threshold to compensate for the impact of changes in the detection time window, thereby ensuring high-precision recognition results under different processing throughputs and truly achieving efficient collaboration and adaptive operation. Attached Figure Description

[0163] Figure 1 This is a flowchart of the control method for an integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control. Detailed Implementation

[0164] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0165] I. Example of an Integrated Ore Crushing-Intelligent Pre-enrichment and Sorting System Based on Intelligent Linkage Control

[0166] This system adopts a modular layout design, adaptable to the limited working space underground. All equipment is explosion-proof for mining, capable of adapting to the harsh underground environment of humidity, dust, and gas. The system includes a buffer ore bin and ore receiving unit connected in sequence, an underground crushing unit, and an intelligent detection and sorting unit. The intelligent detection and sorting unit is further connected to the pre-enrichment storage and transportation unit and the tailings on-site disposal unit. It also includes an intelligent control system for intelligent linkage control of each unit, as well as an auxiliary support system.

[0167] (a) Buffer ore bins and ore receiving units

[0168] The buffer ore bin and receiving unit are used to receive raw ore transported from the underground mining face through the underground mining conveying unit, and to buffer it to balance the supply. The buffer ore bin and receiving unit include a buffer ore bin, a vibrating feeder, and a material level sensor.

[0169] The buffer ore bin is a mine-use explosion-proof steel structure storage bin with an effective volume greater than or equal to 200 cubic meters. The bin walls are lined with wear-resistant plates, and a conical discharge port is provided at the bottom. A material level sensor is installed inside the buffer ore bin for real-time monitoring of the material level.

[0170] The vibrating feeder is located below the buffer ore bin and is used to uniformly feed the raw ore in the bin into the subsequent underground crushing unit. The vibrating feeder is a mining explosion-proof electromagnetic vibrating feeder with a feeding capacity of 600 tons per hour or more. Its feeding flow rate is given by the intelligent control system to adapt to the processing capacity of the subsequent crushing unit.

[0171] (ii) Downhole crushing unit

[0172] The underground crushing unit is used to crush the raw ore in the buffer ore bin into qualified block size required by the subsequent intelligent detection and sorting unit, while realizing intelligent adjustment of the crushing process, overload protection, and control of dust and noise in the underground crushing area.

[0173] The underground crushing unit includes a coarse crusher, a medium crusher, a vibrating screen, and a dust removal and noise reduction device.

[0174] The primary crusher is a mining explosion-proof jaw crusher with a feed opening size of 800 x 1000 mm, a processing capacity of 600 tons per hour or more, an adjustable discharge opening range of 150 to 200 mm, a motor power of 315 kW, and is equipped with a hydraulic overload protection device to prevent hard rock from jamming. The feed particle size of the primary crusher is less than or equal to 1200 mm, and the discharge particle size is less than or equal to 200 mm. The ore crushed by the primary crusher enters the subsequent intermediate crusher.

[0175] The intermediate crusher is a mining-grade explosion-proof hydraulic cone crusher with a processing capacity of 500 tons per hour or more. Its discharge opening can be adjusted via an intelligent control system to accommodate variations in ore hardness. The feed particle size is 200 mm or less, and the discharge particle size is 100 mm or less. The ore crushed by the intermediate crusher then enters the subsequent vibrating screen grading process.

[0176] The vibratory screening uses a double-layer circular vibrating screen with screen sizes of 60 mm and 18 mm (the material size can be adjusted by changing the screen), and a processing capacity of 600 tons per hour or more. After classification by the vibrating screen, qualified blocky ore with a particle size in the range of 18 to 60 mm is conveyed to the subsequent intelligent detection and sorting unit, fine ore with a particle size of less than 18 mm directly enters the pre-enrichment storage and conveying unit, and coarse particles with a particle size of more than 60 mm are returned to the cone crusher for further crushing via a belt conveyor.

[0177] The dust removal and noise reduction device includes a mining pulse bag filter and a soundproof cover. The dust removal efficiency is greater than or equal to 99%, and it is used to control the dust concentration in the crushing workshop to be less than or equal to 10 milligrams per cubic meter and the noise to be less than or equal to 85 decibels, meeting the occupational health requirements underground.

[0178] (III) Intelligent Detection and Sorting Unit

[0179] The intelligent detection and sorting unit is used to detect and intelligently identify qualified block ore after crushing in real time, accurately distinguish target minerals from waste rock, and realize automated sorting with a sorting accuracy greater than or equal to 95%.

[0180] The intelligent detection and sorting unit includes an intelligent detection module, a multimodal deep fusion recognition module, and an automated sorting execution mechanism. The intelligent detection module collects multi-source detection data of the ore to be sorted in real time. The multimodal deep fusion recognition module receives the data collected by the intelligent detection module, performs fusion analysis and recognition based on a deep neural network model, and outputs a determination result of the ore category. The automated sorting execution mechanism performs physical sorting actions based on the determination result output by the multimodal deep fusion recognition module. It includes a mining explosion-proof belt separator and a solenoid valve injection mechanism. The belt conveyor speed of the belt separator is calculated and issued by the sorting parameter collaborative optimization module in the intelligent control system based on the overall material balance of the system, ensuring accurate execution of the sorting action and dynamic matching with the system's processing capacity. The intelligent detection module collects data from the ore on the belt separator.

[0181] 1. Intelligent detection module

[0182] The intelligent detection module includes an XRT online detector and a high-definition industrial camera, both equipped with explosion-proof protective housings.

[0183] The XRT online detector is used to collect ore elemental composition data, ore density data, and ore spot condition data. The elemental composition data includes the content value of the target element and the elemental distribution uniformity index; the density data is the ore bulk density calculated based on the X-ray transmission attenuation coefficient; and the ore spot condition data includes the ore spot distribution density and the average size of the ore spots.

[0184] The high-definition industrial camera is used to acquire RGB images of the ore surface. The acquisition frame rate of the high-definition industrial camera is adjustable, ranging from 30 to 120 frames per second, to adapt to different belt conveyor speeds.

[0185] 2. Multimodal deep fusion recognition module

[0186] The multimodal deep fusion recognition module adopts a dual-branch deep neural network architecture, receives multi-source detection data collected by the intelligent detection module, and outputs ore category determination results and classification confidence scores, which are used to control the automated sorting execution mechanism to perform corresponding sorting actions.

[0187] The dual-branch deep neural network includes an XRT feature extraction branch, an image feature extraction branch, a cross-modal attention fusion layer, a rate adaptive parameter regulator, and a classification decision layer.

[0188] The XRT feature extraction branch is used to extract features from the structured data acquired by the XRT online detector. Elemental composition data, density data, and mineral deposit state data are combined to form an input vector. XRT feature vectors are extracted through a multi-layer fully connected network. :

[0189]

[0190] In the above formula, ,in The target element content value, For elemental distribution uniformity, Density of the ore The density of mineral patches. This represents the average size of the mineral deposits; , , Here are the weight matrices for each layer. , , For each layer's bias vector, The activation function is preferably the ReLU function.

[0191] The image feature extraction branch is used to process RGB images of the ore surface captured by a high-definition industrial camera. Feature extraction is performed. A convolutional neural network is used to process the input image. The image is processed to extract image feature vectors containing texture information. :

[0192]

[0193] In the above formula, Indicates the first Layer convolution operation, The number of convolutional layers is preferably 4. This indicates a global pooling operation.

[0194] The cross-modal attention fusion layer is used to perform deep fusion of XRT feature vectors and image feature vectors. To overcome the problem that simple feature concatenation cannot effectively mine complementary information between different modalities, this scheme designs a cross-modal attention mechanism. Specifically, using features from one modality as the query vector and features from another modality as key-value pairs, the cross-modal attention weights are calculated:

[0195]

[0196]

[0197] In the above formula, The attention weights are calculated using XRT features as the query pair image features. The attention weights are calculated for XRT features using image features as the query. and , and These are the projection matrices of the query and the key in two directions, respectively, and are learnable parameters; The dimension of the key vector is kept consistent in both calculations to ensure scale compatibility.

[0198] The rate adaptive parameter regulator is used to dynamically adjust network parameters according to the current belt conveyor speed of the automated sorting actuator to ensure optimal recognition accuracy under different conveyor speeds. The rate adaptive parameter regulator receives the belt conveyor speed sent by the intelligent control system. Calculate the rate influence factor :

[0199]

[0200] In the above formula, For nominal transport speed, This is an adjustment coefficient, with a value ranging from 0.5 to 2.0.

[0201] To overcome the problems of image acquisition quality degradation and XRT detection time window shortening caused by changes in transport speed, the fusion weight ratio of the cross-modal attention fusion layer is adjusted according to the rate influence factor. When the belt transport speed increases, the reliability of XRT detection data decreases faster than that of image data; therefore, the fusion weights of the two modalities are dynamically adjusted to obtain rate-adaptive fusion features. :

[0202]

[0203] In the above formula, and For rate-related residual weights:

[0204]

[0205]

[0206] in, The projection matrix is ​​the value of , and the parameters are learnable. The last two terms of the calculation formula are rate-adaptive residual connections used to preserve the original feature information.

[0207] This design increases the weight of image features as the belt conveyor speed increases, thereby compensating for the decrease in XRT detection accuracy.

[0208] The classification decision layer receives the rate-adaptive fusion features. The probability distribution of ore categories is output through a fully connected layer and a softmax function:

[0209]

[0210] In the above formula, This is the classification weight matrix. As the bias vector, the ore category probability distribution includes two categories: target minerals and waste rock.

[0211] Meanwhile, the classification decision layer determines the belt conveyor speed based on the information sent by the intelligent control system. Dynamically adjust classification threshold :

[0212]

[0213] In the above formula, The basic classification threshold ranges from 0.5 to 0.7. The threshold adjustment range is 0.05 to 0.15. Adjust the rate coefficient for the threshold; This represents the system's minimum transport rate. If the probability of the target mineral category in the ore category probability distribution is greater than a threshold... If the ore is found to be the target mineral, then it is determined to be waste rock; otherwise, it is determined to be waste rock.

[0214] 3. Automated sorting actuator

[0215] The automated sorting actuator includes a mining explosion-proof belt sorter and a solenoid valve injection mechanism.

[0216] The belt conveyor speed of the belt sorting machine is 1.5 to 2 meters per second, and the conveying speed can be adjusted by the intelligent control system according to the overall operating status of the system.

[0217] The electromagnetic valve injection mechanism, based on the judgment results output by the multimodal deep fusion identification module, diverts the ore identified as the target mineral to the subsequent pre-enrichment storage and transportation unit, and diverts the ore identified as waste rock to the tailings on-site disposal unit.

[0218] (iv) Pre-enrichment storage and delivery unit

[0219] The pre-enrichment storage and transport unit is used to store the target mineral ore sorted by the intelligent detection and sorting unit and transport it to the surface.

[0220] The pre-enrichment storage and conveying unit includes a pre-enrichment ore bin, a ore bin level sensor, and a hoisting and conveying device. The effective volume of the pre-enrichment ore bin is greater than or equal to 150 cubic meters. The ore bin level sensor is used to monitor the ore level in the pre-enrichment ore bin in real time and transmit the level signal to the intelligent control system. The hoisting and conveying device is used to lift the ore from the pre-enrichment ore bin to the surface.

[0221] (v) Tailings on-site disposal unit

[0222] The tailings on-site disposal unit receives waste rock separated by the intelligent detection and sorting unit and performs on-site disposal underground. The unit includes a waste rock buffer silo and a backfill material preparation device. The buffer silo temporarily stores the separated waste rock. The backfill material preparation device processes the waste rock into backfill material for underground goaf areas, realizing the resource utilization of the waste rock.

[0223] (vi) Intelligent control system

[0224] The intelligent control system adopts a three-layer distributed control architecture, including a field control layer, a regional control layer, and a ground monitoring layer.

[0225] The field control layer is deployed in each core unit, and collects equipment operation data and material status data in real time through PLC controllers and sensors to execute control commands. The sensors include level sensors, pressure sensors, temperature sensors, concentration sensors, and rate sensors.

[0226] The area control layer is equipped with an explosion-proof control station in the well, which collects field data, realizes the linkage control of equipment in the area, and has local manual operation function as an emergency backup.

[0227] The ground monitoring layer is equipped with a central control room, which communicates with the underground control station through a mine explosion-proof switch and industrial Ethernet to realize full-process data visualization monitoring, parameter setting, fault alarm and historical data query, and supports remote control.

[0228] From a functional perspective, the intelligent control system includes an intelligent linkage control module, a sorting parameter collaborative optimization module, a fault early warning and self-diagnosis module, and a safety interlock control module.

[0229] 1. Intelligent linkage control module

[0230] The intelligent linkage control module is used to coordinate and control the feeding speed of the vibrating feeder of the buffer ore bin and the receiving unit and the belt conveyor speed of the automated sorting actuator of the intelligent detection and sorting unit according to the operating status and material buffer status of each link of the system, so as to achieve material flow balance and avoid system efficiency loss caused by overflow or emptying of the buffer ore bin.

[0231] The intelligent linkage control module adopts a coordinated control strategy based on model predictive control, specifically including:

[0232] (1) Define variables

[0233] State variables include:

[0234] : The buffer level of the ore bin is measured in tons.

[0235] : The level of the enriched ore bin is measured in tons.

[0236] The known input variables include:

[0237] : The raw ore flow rate of the mining conveying unit input buffer ore bin is measured in tons per hour, and the mining scheduling system provides the planned sequence within the future predicted time domain.

[0238] : The ore flow rate output from the conveying equipment to the ground is constantly increased, measured in tons per hour. The ground dispatch system provides a planned sequence within the future forecast time domain based on the processing needs of the ore processing plant.

[0239] Control variables include:

[0240] : The feed flow rate from the buffer bin to the crushing unit by the vibrating feeder is measured in tons per hour.

[0241] : The incremental feed flow rate, expressed in tons per hour, is the decision variable for model predictive control.

[0242] : The desired belt conveyor speed of the sorting unit is constantly detected and measured in meters per second, which is uniquely determined by the feed flow rate through speed matching constraints.

[0243] System parameters include:

[0244] : Control cycle, with a value range of 5 to 30 seconds;

[0245] : Overall output rate, which represents the proportion of the feed flow rate that finally enters the pre-enriched ore bin after crushing, grading and intelligent sorting. It is calibrated according to the ore grade characteristics and historical operating data, and the value ranges from 0.5 to 0.8.

[0246] : Qualified block size output rate, which represents the proportion of the mass of the feed that has passed crushing and grading and enters the intelligent detection and sorting unit. It is preset according to the characteristics of the ore and has a value range of 0.7 to 0.9.

[0247] : Bulk density of ore, in tons per cubic meter;

[0248] : Effective cross-sectional area of ​​the belt, in square meters.

[0249] (2) Establish the state space model of system material flow.

[0250] The dynamic equation for the buffer ore bin level is:

[0251]

[0252] In the above formula, dividing by 3600 converts the flow rate unit (tons / hour) and control cycle unit (seconds) into a unified amount of material level change (tons). for The buffer ore level at any given time can be recursively calculated using this formula. Buffer ore bin level at any time , It is a positive integer.

[0253] The dynamic equation for the pre-enriched ore bin level is:

[0254]

[0255] The above formula shows that the input flow rate of the pre-enriched ore bin is the feed flow rate multiplied by the overall output rate, and the output flow rate is the lift flow rate. for The pre-enriched ore bin level at any given time can be recursively calculated using this formula. Pre-enriched ore bin level at any time , It is a positive integer.

[0256] (3) Define the relationship between feed flow rate and feed flow rate increment.

[0257] For any The feed flow rate at any given time is calculated recursively using the feed flow rate increment:

[0258]

[0259] Within the prediction time domain, given the initial feed flow rate and incremental sequence It can recursively calculate the feed flow rate at each future time point:

[0260]

[0261] In the above formula, To predict the length of the time domain.

[0262] (4) Define velocity matching constraints

[0263] Since the underground crushing unit has sufficient processing capacity (the coarse crusher has a processing capacity of 600 tons per hour or more), the crushing process does not constitute a system bottleneck. In this control model, the crushing unit is considered a processing stage that does not require speed adjustment, and its processing capacity is equal to... .

[0264] To ensure that the intelligent detection and sorting unit does not accumulate or interrupt the material flow, its processing capacity should match the flow rate of qualified block-sized ore output from the crushing unit. The relationship between the belt conveyor speed and the feed flow rate is as follows:

[0265]

[0266] In the above formula, dividing by 3600 converts the flow rate unit (tons / hour) into a dimension that matches the belt conveyor speed unit (meters / second).

[0267] (5) Obtain the predicted input quantity

[0268] The intelligent linkage control module obtains the planned sequence of raw ore input flow in the future predicted time domain from the mining scheduling system:

[0269]

[0270] When mining planning data is unavailable, a sliding window mean method is used to make predictions based on historical input flow data.

[0271] The intelligent linkage control module obtains the planned sequence of increased output flow in the future predicted time domain from the ground dispatch system:

[0272]

[0273] Both sequences mentioned above are used as known boundary conditions for model predictive control.

[0274] (6) Define the model predictive control optimization objective

[0275] Construct the optimization objective function for the model predictive controller:

[0276]

[0277] In the above formula:

[0278] To predict the time domain length, the value ranges from 10 to 30 control cycles;

[0279] To control the time domain length, the value ranges from 3 to 10 control cycles, and ;

[0280] To buffer the target material level in the ore bin, it is typically set to 40% to 60% of the ore bin's capacity.

[0281] The target material level for the pre-enriched ore bin is typically set at 40% to 60% of the bin's capacity.

[0282] , , These are weighting coefficients used to balance the relative importance of various optimization objectives.

[0283] The decision variable is the sequence of feed flow rate increments. ,total There are several decision variables. When the prediction time exceeds the control time domain, i.e. season .

[0284] (7) Define control constraints

[0285] Material level constraints:

[0286]

[0287]

[0288] In the above formula, The maximum capacity of the buffer ore bin (the mass corresponding to a volume greater than or equal to 200 cubic meters). The maximum capacity of the pre-enriched ore bin (the mass corresponding to a volume greater than or equal to 150 cubic meters).

[0289] Feed flow rate constraints:

[0290]

[0291] In the above formula, and These are the minimum and maximum feed flow rates of the vibrating feeder, respectively.

[0292] Belt conveyor speed constraints:

[0293]

[0294] In the above formula, and These are the lower and upper limits of the belt conveyor speed, respectively, with values ​​ranging from 1.5 to 2 meters per second.

[0295] Feed flow rate change constraint:

[0296]

[0297] In the above formula, This is the maximum allowable change in feed flow rate within a single control cycle, used to avoid drastic fluctuations in feed rate that could impact the equipment.

[0298] (8) Control the execution process

[0299] In each control cycle, the intelligent linkage control module performs the following steps:

[0300] Step 1: Collect the current buffer ore bin level Pre-enriched ore bin material level and current feed flow rate ;

[0301] Step 2: Obtain the future raw ore input flow plan sequence from the mining scheduling system. Obtain the future output flow plan sequence from the ground dispatch system. ;

[0302] Step 3: Based on the state-space model, constraints, and optimization objective function, a quadratic programming algorithm is used to solve for the optimal feed flow rate increment sequence.

[0303]

[0304] Step 4: Calculate the target feed flow rate for the current control cycle based on the first optimal increment:

[0305]

[0306] Step 5: Set the target feed flow rate Send to the vibrating feeder for execution;

[0307] Step 6: Calculate the corresponding target belt conveyor speed based on the speed matching constraints:

[0308]

[0309] Step 7: Set the target conveyor belt speed The data is sent to the sorting parameter collaborative optimization module, triggering adaptive adjustment of the sorting parameters;

[0310] Step 8: Enter the next control cycle and repeat the above steps to achieve rolling time-domain optimization control.

[0311] 2. Sorting Parameter Collaborative Optimization Module

[0312] The sorting parameter collaborative optimization module is used to optimize the belt conveyor speed of the intelligent detection and sorting unit after the intelligent linkage control module adjusts it, and sends a speed signal to the multimodal deep fusion recognition module of the intelligent detection and sorting unit, so that the recognition module can adaptively adjust the network parameters and classification threshold according to the current speed.

[0313] The sorting parameter collaborative optimization module establishes a rate-accuracy prediction model to constrain the adjustment range of the belt conveyor rate of the intelligent detection and sorting unit in the intelligent linkage control module. The prediction model expression is:

[0314]

[0315] In the above formula, For belt conveyor speed The predicted sorting accuracy is as follows. For the highest sorting accuracy ( ), For accuracy attenuation coefficient, , This is the preset optimal belt conveyor speed.

[0316] When the intelligent linkage control module calculates the target belt conveyor speed This leads to changes in predictive sorting accuracy. Below the threshold When the value is between 0.92 and 0.95, the sorting parameter collaborative optimization module corrects the target belt conveyor speed:

[0317]

[0318] In the above formula, The corrected target belt conveyor speed. To meet the maximum allowable rate constrained by sorting accuracy:

[0319]

[0320] If no correction is needed, then directly set... .

[0321] Then press Control the belt operation of the intelligent detection and sorting unit, and The data is sent to the rate adaptive parameter regulator and classification decision layer of the intelligent detection and sorting unit.

[0322] 3. Fault Early Warning and Self-Diagnosis Module

[0323] The fault warning and self-diagnosis module monitors the parameters of each unit device in real time. When the parameters exceed the normal range, it issues an alarm and completes a simple fault self-diagnosis.

[0324] 4. Safety interlock control module

[0325] The safety interlock control module monitors environmental parameters such as gas concentration and dust concentration, and triggers equipment interlock shutdown and personnel evacuation warnings in emergency situations.

[0326] (vii) Auxiliary support system

[0327] The auxiliary support system includes a ventilation and dust prevention system, a power and water supply system, and a safety protection system.

[0328] The ventilation and dust control system is equipped with an underground local ventilator and a pulse bag filter, with an air volume greater than or equal to 200 cubic meters per minute, to achieve ventilation and dust control.

[0329] The power supply system adopts a mine-use explosion-proof transformer, distribution cabinet and dual-circuit power supply, with a total power supply capacity of greater than or equal to 1500 kilowatts, ensuring continuous power supply.

[0330] The water supply system is equipped with an underground static pressure water tank and a water supply network, with a water consumption of 50 cubic meters per hour or more, which meets the needs of dust removal, stirring and cooling.

[0331] The safety protection system includes an infrared monitoring camera, a gas detector, a carbon monoxide detector, and an emergency stop button, enabling 24-hour real-time monitoring. All equipment is explosion-proof, moisture-proof, and dustproof.

[0332] II. Implementation Examples of Control Methods for an Integrated Ore Crushing-Intelligent Pre-enrichment and Separation System Based on Intelligent Linkage Control

[0333] like Figure 1 As shown, this method is applied to the above-mentioned integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, and includes the following steps:

[0334] Step 1: Continuously collect on-site data.

[0335] Step 1.1: Start all unit devices of the system and enter standby mode;

[0336] Step 1.2: The sensors in the field control layer continuously collect operational data from each unit, including: buffer ore bin level. Pre-enriched ore bin material level Waste rock buffer bin level, vibrating feeder feed flow rate The capacity of the coarse crusher, the capacity of the medium crusher, and the belt conveyor speed of the intelligent detection and sorting unit. And the status parameters of each device, such as temperature and pressure;

[0337] Step 1.3: The collected data is aggregated from the regional control layer to the ground monitoring layer for data preprocessing and storage.

[0338] Step 2: Intelligent linkage and coordination control.

[0339] Step 2.1: The intelligent linkage control module operates at a fixed cycle. (Period length is 5 to 30 seconds) Perform coordination control calculations;

[0340] Step 2.2: At the beginning of each control cycle, obtain the current material level of each ore bin. , and current feed flow rate ;

[0341] Step 2.3: Obtain the future prediction time domain from the mining scheduling system The raw ore input flow planning sequence Obtain future prediction time domain from ground dispatch system Internal increase output flow plan sequence ;

[0342] Step 2.4: Based on the material flow state-space model, constraints, and optimization objective function, a quadratic programming algorithm is used to obtain the optimal feed flow increment sequence. ,in To control the length of the time domain;

[0343] Step 2.5: Calculate the target feed flow rate for the current control cycle based on the first optimal increment:

[0344]

[0345] Step 2.6: Set the target feed flow rate Send to the vibrating feeder for execution;

[0346] Step 2.7: Calculate the corresponding target belt conveyor speed based on the speed matching constraints:

[0347]

[0348] Step 2.8: Set the target belt conveyor speed Send to the sorting parameter collaborative optimization module.

[0349] Step 3: Coordinated adjustment of sorting parameters.

[0350] Step 3.1: The sorting parameter collaborative optimization module receives the target belt conveyor speed of the intelligent detection and sorting unit calculated by the intelligent linkage control module, denoted as... ;

[0351] Step 3.2: Calculate the predicted sorting accuracy at this speed based on the rate-accuracy prediction model. ;

[0352] Step 3.3: Determine whether the sorting accuracy meets the constraints and make corrections accordingly to obtain the corrected belt conveyor speed of the intelligent detection and sorting unit. ;

[0353] Step 3.4: Press Control the belt operation of the intelligent detection and sorting unit, and The data is sent to the rate adaptive parameter regulator and classification decision layer of the intelligent detection and sorting unit.

[0354] Step 4: Intelligent detection and sorting unit adaptive recognition.

[0355] Step 4.1: The intelligent detection module continuously collects ore data passing through the belt separator, including elemental composition data, density data, and ore spot condition data collected by the XRT online detector, as well as RGB images of the ore surface collected by a high-definition industrial camera. ;

[0356] Step 4.2: The rate adaptive parameter regulator adjusts the speed according to the received belt conveyor speed. Calculate the rate influence factor ;

[0357] Step 4.3: The XRT feature extraction branch extracts features from the XRT detection data to obtain the XRT feature vector. ;

[0358] Step 4.4: The image feature extraction branch extracts features from the ore surface image to obtain the image feature vector. ;

[0359] Step 4.5: The cross-modal attention fusion layer adaptively weights and fuses the features of the two modalities based on the current rate influence factor to obtain the rate-adaptive fusion features. And calculate the probability distribution of ore categories based on rate adaptive fusion features. ;

[0360] Step 4.6: The classification decision layer calculates the adaptive classification threshold based on the current belt conveyor speed. ;

[0361] Step 4.7: If the probability distribution of ore types... The probability of the target mineral category being greater than the threshold If the ore is found to be the target mineral, then it is determined to be waste rock; otherwise, it is determined to be waste rock.

[0362] Step 5: Sorting and Material Diversion: The automated sorting execution mechanism controls the solenoid valve injection mechanism based on the judgment result of the multimodal deep fusion identification module: For ores judged to be target minerals, the solenoid valve injection mechanism does not operate, and the ores are normally transported along the belt to the pre-enrichment storage and conveying unit; for ores judged to be waste rock, the solenoid valve injection mechanism operates, and the waste rock is injected and diverted to the tailings on-site disposal unit.

[0363] Step 6: Control effect feedback and model update.

[0364] Step 6.1: Record the actual sorting accuracy and system operating parameters for each control cycle;

[0365] Step 6.2: Periodically update the parameters in the rate-accuracy prediction model based on actual operating data. and To adapt to changes in ore properties;

[0366] Step 6.3: Optimize the weight parameters in the model predictive controller based on historical operating data. to and prediction time domain Control Time Domain .

[0367] Step 7: Fault monitoring and safety protection.

[0368] Step 7.1: The fault warning and self-diagnosis module continuously monitors the operating parameters of each device, and issues an alarm and records fault information when an abnormality is detected;

[0369] Step 7.2: The safety interlock control module continuously monitors environmental parameters such as gas concentration and dust concentration. When any parameter exceeds the safety threshold, it triggers the equipment interlock shutdown procedure and issues a personnel evacuation warning.

[0370] III. Examples of On-site Operation Results

[0371] (I) Example 1 (Application in iron ore mines)

[0372] A certain iron mine has a depth of 800 meters, a maximum ore block size of 1200 mm, a waste rock rate of 35%, and an annual ore processing capacity of 1.2 million tons. The integrated system and control method of this invention are used, and the specific configuration and operational effects are as follows:

[0373] 1. System Configuration

[0374] Buffer bins and receiving units: Scraper conveyors are used to receive raw ore from underground mining faces. Buffer bin volume: 200 cubic meters. Vibrating feeder has a feeding capacity of 600 tons per hour.

[0375] The underground crushing unit consists of an 800×1000 mm jaw crusher (600 tons per hour), a hydraulic cone crusher (500 tons per hour), a double-layer circular vibrating screen (600 tons per hour), and a pulse bag filter.

[0376] Intelligent detection and sorting unit: Built-in iron ore grade database, XRF online detector, and 20-megapixel industrial camera. The automated sorting actuator uses a mining explosion-proof belt separator (belt separator conveying speed 1.8 meters per second) and a solenoid valve injection mechanism.

[0377] Pre-enrichment storage and conveying unit: 150 cubic meter ore bin, steel wire rope core belt hoist (lifting capacity 500 tons per hour); equipped with ore bin level sensor and belt conveyor.

[0378] Tailings on-site disposal unit: small jaw crusher, filling and mixing station and hydraulic filling pump are selected; intelligent control system adopts PLC controller and linkage with ground monitoring center.

[0379] Auxiliary support system: equipped with a 200 cubic meter per minute ventilation fan, dual-circuit power supply system, infrared monitoring and gas detection equipment.

[0380] 2. Performance

[0381] After the system is put into operation, the sorting accuracy is 96.5%, the grade of pre-enriched ore is increased by 28%, the annual waste rock transportation volume is reduced by 420,000 tons, and the transportation cost is reduced by 18.9 million yuan; the on-site backfilling utilization rate of tailings is 100%, and no new tailings dam is needed; the number of underground workers is reduced by 60%, and the equipment fault-free operation time is 95.8%, which meets the requirements for the construction of green mines and smart mines.

[0382] (II) Example 2 (Application in Gold Mines)

[0383] A gold mine has a depth of 1000 meters, extracts gold ore with a grade of 0.3 grams per ton, a maximum ore block size of 1200 mm, a waste rock rate of 32%, and an annual ore processing capacity of 1 million tons. The system and control method of this invention are used, and the specific configuration and operational effects are as follows:

[0384] 1. System Configuration

[0385] Buffer bins and receiving units: Scraper conveyors are used to receive raw ore from underground mining faces. Buffer bin volume: 200 cubic meters. Vibrating feeder has a feeding capacity of 600 tons per hour.

[0386] The underground crushing unit consists of an 800×1000 mm jaw crusher (600 tons per hour), a hydraulic cone crusher (500 tons per hour), a double-layer circular vibrating screen (600 tons per hour), and a pulse bag filter.

[0387] Intelligent detection and sorting unit: Built-in gold ore grade standard database, XRF online detector, and 20-megapixel industrial camera. The automated sorting actuator uses a mining explosion-proof belt separator (belt separator conveying speed 1.8 meters per second) and a solenoid valve injection mechanism.

[0388] Pre-enrichment storage and conveying unit: 150 cubic meter ore bin, steel wire rope core belt hoist (lifting capacity 500 tons per hour); equipped with ore bin level sensor and belt conveyor.

[0389] Tailings on-site disposal unit: small jaw crusher, filling and mixing station and hydraulic filling pump are selected; intelligent control system adopts PLC controller and linkage with ground monitoring center.

[0390] Auxiliary support system: equipped with a 200 cubic meter per minute ventilation fan, dual-circuit power supply system, infrared monitoring and gas detection equipment.

[0391] 2. Performance

[0392] After the system was put into operation, the sorting accuracy of gold ore with a particle size of 18 to 60 mm reached 95.8%, and the grade of pre-enriched ore increased from 0.3 grams per ton of raw ore to 1.15 grams per ton, an increase of 283.3%. The annual waste rock transportation volume was reduced by 320,000 tons, reducing transportation and surface storage costs by approximately 16 million yuan. The on-site backfilling utilization rate of tailings was 100%, eliminating the need to build new surface tailings dams, saving approximately 80 acres of land resources, and completely eliminating ecological safety risks such as tailings dam leakage and dam failure. The automation level of the entire process was greater than or equal to 90%, the number of underground workers was reduced by 55%, and the average annual mean time between failures of the equipment reached 96.2%, effectively reducing the intensity of manual labor and the safety risks of underground operations, which fully meets the development needs of efficient development of gold mine resources and the construction of green and intelligent mines.

[0393] It should be noted that, as will be apparent to those skilled in the art, the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics thereof. The scope of the present invention is defined by the claims rather than the foregoing description.

Claims

1. An integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control, comprising a buffer ore bin and receiving unit installed underground, an underground crushing unit, and a sorting unit, characterized in that: It also includes a pre-enrichment storage and transportation unit and a tailings on-site disposal unit set up underground, as well as an intelligent control system for intelligent linkage control of each unit; The buffer ore bin is sequentially connected to the ore receiving unit, the underground crushing unit, and the sorting unit. The sorting unit is further connected to the pre-enrichment storage and transportation unit and the tailings on-site disposal unit, respectively. The sorting unit is an intelligent detection and sorting unit. The intelligent control system includes an intelligent linkage control module and a sorting parameter collaborative optimization module; The intelligent linkage control module is used to coordinate and control the feeding flow rate of the vibrating feeder of the buffer ore bin and the receiving unit and the belt conveying speed of the automated sorting actuator of the intelligent detection and sorting unit according to the operating status and material buffer status of each link of the system. The intelligent linkage control module adopts a coordinated control strategy based on model predictive control. This coordinated control strategy includes: defining variables, which include state variables, known input variables, control variables, and system parameters. The state variables include the buffer ore bin level and the pre-enrichment ore bin level. The known input variables include the upstream raw ore flow rate and the downstream hoisting output ore flow rate. The control variables include the vibrating feeder feed flow rate, the feed flow rate increment, and the belt conveyor speed of the intelligent detection and sorting unit. The system parameters include the control cycle, overall output rate, qualified block size output rate, ore bulk density, and the effective carrying cross-sectional area of ​​the belt conveyor. A system material flow state space model is established. This state space model includes a first dynamic equation describing the change in the buffer ore bin level and a second dynamic equation describing the change in the pre-enrichment ore bin level. The first dynamic equation reflects the relationship between the buffer ore bin level and the difference between the raw ore input flow rate and the feed flow rate. The second dynamic equation reflects the relationship between the pre-enrichment ore bin level and the buffer ore bin level. The model defines the relationship between the ore bin level and the difference between the input flow rate (calculated using the comprehensive output rate) and the lifting output flow rate; defines the recursive relationship between the feed flow rate and the feed flow rate increment; defines the speed matching constraint between the belt conveyor speed and the feed flow rate, which is based on the qualified block size output rate, ore bulk density, and effective bearing cross-sectional area of ​​the belt; obtains the planned sequence of raw ore input flow rate in the future prediction time domain from the mining scheduling system and the planned sequence of lifting output flow rate in the future prediction time domain from the ground scheduling system, as the predicted input of the model predictive control; defines the optimization objective of the model predictive control, which is constructed based on the degree of deviation of the state variables from the target value and the change amplitude of the control variables; defines the control constraints, which include the level constraint to maintain the buffer ore bin level and the pre-enriched ore bin level within their respective safe capacity ranges, the feed flow rate range constraint, the belt conveyor speed range constraint, and the feed flow rate change rate constraint. The sorting parameter collaborative optimization module establishes a rate-accuracy prediction model, which describes the relationship between sorting accuracy and belt conveyor speed: sorting accuracy equals the highest sorting accuracy minus the accuracy attenuation, and the accuracy attenuation is proportional to the square of the degree to which the belt conveyor speed deviates from the preset optimal belt conveyor speed; the sorting parameter collaborative optimization module is used to optimize the belt conveyor speed after the intelligent linkage control module adjusts the belt conveyor speed of the intelligent detection and sorting unit, and sends a rate signal to the intelligent detection and sorting unit to control the intelligent detection and sorting unit to adaptively adjust.

2. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 1, characterized in that, The intelligent linkage control module specifically includes: (1) Define variables State variables include: : The buffer ore bin level between the buffer ore bin and the receiving unit, in tons; : The level of the pre-enriched ore bin in the pre-enriched storage and conveying unit at all times, in tons; The known input variables include: : The raw ore flow rate of the mining conveying unit input buffer ore bin is measured in tons per hour, and the mining scheduling system provides the planned sequence within the future predicted time domain. : The ore flow rate output from the conveying equipment to the ground is constantly increased, in tons per hour, and the ground dispatch system provides a planned sequence within the future forecast time domain based on the processing needs of the ore processing plant; Control variables include: : The feed flow rate of the vibrating feeder from the buffer ore bin to the underground crushing unit is measured in tons per hour. : The incremental feed flow rate, expressed in tons per hour, is the decision variable for model predictive control. : The desired belt conveyor speed of the sorting unit is constantly detected and measured in meters per second, which is uniquely determined by the feed flow rate through speed matching constraints. System parameters include: Control cycle; : Overall output rate, which represents the proportion of feed flow rate that finally enters the pre-enriched ore bin after crushing, grading and intelligent sorting, is determined based on ore grade characteristics and historical operating data; : Qualified block size output rate, which represents the proportion of the feed flow that enters the intelligent detection and sorting unit after crushing and grading, and is preset according to the characteristics of the ore; : Ore bulk density, in tons per cubic meter; : Effective load-bearing cross-sectional area of ​​the belt, in square meters; (2) Establish the state space model of system material flow The dynamic equation for the buffer ore bin level is: ； In the above formula, for The buffer ore level at any given time can be recursively calculated using this formula. Buffer ore bin level at any time , It is a positive integer; The dynamic equation for the pre-enriched ore bin level is: ； In the above formula, for The pre-enriched ore bin level at any given time can be recursively calculated based on this formula. Pre-enriched ore bin level at any time , It is a positive integer; (3) Define the relationship between feed flow rate and feed flow rate increment. For any The feed flow rate at any given time is calculated recursively using the feed flow rate increment: ； Within the prediction time domain, given the initial feed flow rate and incremental sequence It can recursively calculate the feed flow rate at each future time point: ； In the above formula, To predict the length of the time domain; (4) Define velocity matching constraints The relationship between belt conveyor speed and feed flow rate is as follows: ； (5) Obtain the predicted input quantity The intelligent linkage control module obtains the planned sequence of raw ore input flow in the future predicted time domain from the mining scheduling system: ； The intelligent linkage control module obtains the planned sequence of increased output flow in the future predicted time domain from the ground dispatch system: ； (6) Define the optimization objective of model predictive control and construct the optimization objective function of the model predictive controller; (7) Define control constraints Material level constraints: ； ； In the above formula, To buffer the maximum capacity of the ore bin, This is the maximum capacity of the pre-enriched ore bin; Feed flow rate constraints: ； In the above formula, and These are the minimum and maximum feed flow rates of the vibrating feeder, respectively. Belt conveyor speed constraints: ； In the above formula, and These are the lower and upper limits of belt conveyor speed, respectively. Feed flow rate change constraint: ； In the above formula, This represents the maximum allowable change in feed flow rate within a single control cycle.

3. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 2, characterized in that, The optimization objective function is: ； In the above formula: To predict the time domain length, the value ranges from 10 to 30 control cycles; To control the time domain length, the value ranges from 3 to 10 control cycles, and ; To buffer the target material level in the ore bin, it is typically set to 40% to 60% of the ore bin's capacity. The target material level for the pre-enriched ore bin is typically set at 40% to 60% of the bin's capacity. , , These are weighting coefficients used to balance the relative importance of various optimization objectives; The decision variable is the sequence of feed flow rate increments. ,total There are several decision variables; when the prediction time exceeds the control time domain, i.e. season .

4. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 3, characterized in that, The intelligent linkage control module executes the following steps in each control cycle: Step 1: Collect the current buffer ore bin level Pre-enriched ore bin material level and current feed flow rate ; Step 2: Obtain the future raw ore input flow plan sequence from the mining scheduling system. Obtain the future output flow plan sequence from the ground dispatch system. ; Step 3: Based on the state-space model, constraints, and optimization objective function, a quadratic programming algorithm is used to solve for the optimal feed flow rate increment sequence. ； Step 4: Calculate the target feed flow rate for the current control cycle based on the first optimal increment: ； Step 5: Set the target feed flow rate Send to the vibrating feeder for execution; Step 6: Calculate the corresponding target belt conveyor speed based on the speed matching constraints: ； Step 7: Set the target conveyor belt speed The data is sent to the sorting parameter collaborative optimization module, triggering adaptive adjustment of the sorting parameters; Step 8: Enter the next control cycle and repeat the above steps to achieve rolling time-domain optimization control.

5. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 3 or 4, characterized in that, The sorting parameter collaborative optimization module establishes a rate-accuracy prediction model, and the prediction model expression is: ； In the above formula, For belt conveyor speed The predicted sorting accuracy is as follows. For the highest sorting accuracy, For accuracy attenuation coefficient, The preset optimal belt conveyor speed; When the intelligent linkage control module calculates the target belt conveyor speed This leads to changes in predictive sorting accuracy. Below the threshold At the same time, the sorting parameter collaborative optimization module corrects the target belt conveyor speed: ； In the above formula, The corrected target belt conveyor speed. To meet the maximum allowable rate constrained by sorting accuracy: ； If no correction is needed, then directly set... ; Then press Control the belt operation of the intelligent detection and sorting unit, and The data is sent to the rate adaptive parameter regulator and classification decision layer of the intelligent detection and sorting unit.

6. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 5, characterized in that: The intelligent detection and sorting unit includes an intelligent detection module, a multimodal deep fusion recognition module, and the automated sorting execution mechanism; The intelligent detection module is used to collect multi-source detection data of the ore to be sorted in real time; the multimodal deep fusion recognition module is used to receive the data collected by the intelligent detection module, and perform fusion analysis and recognition based on a deep neural network model to output the ore category determination result; the automated sorting execution mechanism is used to perform physical sorting actions according to the determination result output by the multimodal deep fusion recognition module. It includes a belt sorter and a solenoid valve injection mechanism. The belt conveying speed of the belt sorter is given by the sorting parameter collaborative optimization module in the intelligent control system after calculation based on the overall material balance state of the system; the intelligent detection module collects data from the ore on the belt sorter.

7. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 6, characterized in that: The intelligent detection module includes an XRT online inspection instrument and a high-definition industrial camera; The XRT online detector is used to collect ore elemental composition data, ore density data, and ore spot condition data. The elemental composition data includes the target element content value and element distribution uniformity index; the density data is the ore bulk density calculated based on the X-ray transmission attenuation coefficient; and the ore spot condition data includes ore spot distribution density and ore spot average size. The high-definition industrial camera is used to acquire RGB images of the ore surface; The multimodal deep fusion recognition module adopts a dual-branch deep neural network architecture, including an XRT feature extraction branch and an image feature extraction branch; In the XRT feature extraction branch, elemental composition data, density data, and mineral patch state data are combined to form an input vector. XRT feature vectors are extracted through a multi-layer fully connected network. : ； In the above formula, ,in The target element content value, For elemental distribution uniformity, Density of the ore The density of mineral patches. This represents the average size of the mineral deposits; , , Here are the weight matrices for each layer. , , For each layer's bias vector, For activation functions; The image feature extraction branch uses a convolutional neural network to process the input image. The image is processed to extract image feature vectors containing texture information. : ； In the above formula, Indicates the first Layer convolution operation, The number of convolutional layers. This indicates a global pooling operation.

8. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 7, characterized in that: The multimodal deep fusion recognition module also includes a cross-modal attention fusion layer; The cross-modal attention fusion layer is used to perform deep fusion of XRT feature vectors and image feature vectors; specifically, using features from one modality as query vectors and features from another modality as key-value pairs, cross-modal attention weights are calculated. ； ； In the above formula, The attention weights are calculated using XRT features as the query pair image features. The attention weights are calculated for XRT features using image features as the query. and , and These are the projection matrices of the query and the key in two directions, respectively, and are learnable parameters; is the dimension of the key vector.

9. The integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 7 or 8, characterized in that: The multimodal deep fusion recognition module also includes a rate adaptive parameter regulator and a classification decision layer; The rate adaptive parameter regulator receives the belt conveyor rate sent by the intelligent control system. Calculate the rate influence factor : ； In the above formula, For nominal transport speed, This is the adjustment coefficient; Then calculate the rate-adaptive fusion features. : ； In the above formula, and For rate-related residual weights: ； ； in, The projection matrix is ​​the value of , and the parameters are learnable. The classification decision layer receives the rate-adaptive fusion features. The probability distribution of ore categories is output through a fully connected layer and a softmax function: ； In the above formula, This is the classification weight matrix. As the bias vector, the ore category probability distribution includes two categories: target minerals and waste rock; Meanwhile, the classification decision layer determines the belt conveyor speed based on the information sent by the intelligent control system. Dynamically adjust classification threshold : ； In the above formula, Basic classification threshold; The threshold adjustment range; Adjust the rate coefficient for the threshold; This represents the system's minimum transport rate. If the probability of the target mineral category in the ore category probability distribution is greater than the threshold If the ore is found to be the target mineral, then it is determined to be waste rock; otherwise, it is determined to be waste rock.

10. A control method for an integrated ore crushing-intelligent pre-enrichment and sorting system based on intelligent linkage control as described in claim 9, characterized in that... Includes the following steps: Step 1: Continuously collect on-site data; Step 2: Intelligent linkage and coordination control, specifically including: Step 2.1: The intelligent linkage control module operates at a fixed cycle. Perform coordination and control operations; Step 2.2: At the beginning of each control cycle, obtain the current material level of each ore bin. , and current feed flow rate ; Step 2.3: Obtain the future prediction time domain from the mining scheduling system The raw ore input flow planning sequence Obtain future prediction time domain from ground dispatch system Internal increase output flow plan sequence ; Step 2.4: Based on the material flow state-space model, constraints, and optimization objective function, a quadratic programming algorithm is used to obtain the optimal feed flow increment sequence. ,in To control the length of the time domain; Step 2.5: Calculate the target feed flow rate for the current control cycle based on the first optimal increment: ； Step 2.6: Set the target feed flow rate Send to the vibrating feeder for execution; Step 2.7: Calculate the corresponding target belt conveyor speed based on the speed matching constraints: ； Step 2.8: Set the target belt conveyor speed Send to the sorting parameter collaborative optimization module; Step 3: Coordinated adjustment of sorting parameters; Step 3.1: The sorting parameter collaborative optimization module receives the target belt conveyor speed of the intelligent detection and sorting unit calculated by the intelligent linkage control module, denoted as... ; Step 3.2: Calculate the predicted sorting accuracy at this speed based on the rate-accuracy prediction model. ; Step 3.3: Determine whether the sorting accuracy meets the constraints and make corrections accordingly to obtain the corrected belt conveyor speed of the intelligent detection and sorting unit. ; Step 3.4: Press Control the belt operation of the intelligent detection and sorting unit, and The data is sent to the rate adaptive parameter regulator and classification decision layer of the intelligent detection and sorting unit. Step 4: Intelligent detection and sorting unit adaptive recognition, specifically including: Step 4.1: The intelligent detection module continuously collects ore data passing through the belt separator, including elemental composition data, density data, and ore spot condition data collected by the XRT online detector, as well as RGB images of the ore surface collected by a high-definition industrial camera. ; Step 4.2: The rate adaptive parameter regulator adjusts the speed according to the received belt conveyor speed. Calculate the rate influence factor ; Step 4.3: The XRT feature extraction branch extracts features from the XRT detection data to obtain the XRT feature vector. ; Step 4.4: The image feature extraction branch extracts features from the ore surface image to obtain the image feature vector. ; Step 4.5: The cross-modal attention fusion layer adaptively weights and fuses the features of the two modalities based on the current rate influence factor to obtain the rate-adaptive fusion features. And calculate the probability distribution of ore categories based on rate adaptive fusion features. ; Step 4.6: The classification decision layer calculates the adaptive classification threshold based on the current belt conveyor speed. ; Step 4.7: If the probability distribution of ore types... The probability of the target mineral category being greater than the threshold If the ore is positive, it is determined to be the target mineral; otherwise, it is determined to be waste rock. Step 5: Sorting and Material Diversion: The automated sorting execution mechanism controls the solenoid valve injection mechanism based on the judgment results of the multimodal deep fusion identification module. For ores identified as target minerals, the solenoid valve injection mechanism does not operate, and the ore is normally transported along the belt to the pre-enrichment storage and conveying unit. For ores identified as waste rock, the solenoid valve injection mechanism operates, and the waste rock is injected and diverted to the tailings on-site disposal unit.

Images