A sand medium high-efficiency sedimentation tank control system based on visual recognition neural network

Abstract

The application discloses a sand medium high-efficiency sedimentation tank control system based on visual recognition neural network, and relates to the technical field of water treatment.The system collects sand and mud flow images and pressure data through an industrial camera, extracts features by using an improved Mask R-CNN, positions a separation interface and evaluates sand purity in combination with a visual-pressure fusion algorithm, then corrects control parameters by using a purity memory self-tuning algorithm, generates adjustment instructions for the inlet pressure of a hydraulic cyclone and the height of an overflow pipe, dynamically adjusts by an executing mechanism, and finally feeds back the execution result to the front end to form a closed loop.The application accurately identifies the sand and mud separation state and quantifies the sand purity by fusing visual and pressure data, provides a basis for intelligent regulation and control, dynamically adjusts operation parameters based on historical and real-time data, ensures safety through multiple protection, and continuously optimizes the algorithm with the help of closed-loop feedback, thereby improving the sedimentation treatment efficiency and stability, reducing manual intervention, and promoting the development of sand medium treatment towards intelligence and refinement.

Description

Technical Field

[0001] This invention relates to the field of water treatment technology, specifically to a high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network. Background Technology

[0002] As a key piece of equipment in water treatment, environmental protection, and resource utilization, the core function of a high-efficiency sand media sedimentation tank is to achieve efficient separation of sand particles and sludge, and to recycle the sand media. The separation effect and sand purity directly affect the overall treatment efficiency, resource recovery rate, and operating costs. With the continuous improvement of environmental protection requirements and the in-depth promotion of the concept of resource recycling, related industries have put forward higher requirements for the accuracy of sand and sludge separation, the purity of recycled sand particles, and the stability of system operation. The traditional control system, which relies on a single detection method and fixed control logic, can no longer meet the treatment needs under complex working conditions. The deep integration of visual recognition technology, neural network algorithms, and industrial control technology provides important support for the intelligent upgrading of sand media sedimentation treatment. Against this background, a control system that integrates multi-source monitoring data, optimizes feature recognition algorithms, and constructs a dynamic control mechanism has become an important direction for industry development. It aims to break through the limitations of traditional treatment modes through technological innovation and achieve a refined, efficient, and sustainable sand media sedimentation treatment process.

[0003] Traditional sand media sedimentation tank control systems have many limitations in practical applications, making it difficult to adapt to complex and ever-changing treatment conditions. In the sand and sludge identification stage, they rely heavily on single pressure sensor data or manual observation, lacking comprehensive extraction and analysis of multi-dimensional features such as sand and sludge color and edge contours. This results in inaccurate positioning of the sand-sludge separation interface and an inability to accurately capture the true state of sludge adhesion on the sand particle surface. Regarding purity assessment, there is a lack of a systematic quantitative evaluation system, relying mostly on qualitative descriptions or simple numerical estimations, making it difficult to scientifically classify sand particle purity levels. Consequently, control commands lack specificity, affecting the quality of sand reuse. In the control decision-making stage, control parameters are mostly fixed settings or adjusted based on manual experience, without dynamic optimization based on historical adjustment data and real-time operating status. This results in an inability to respond promptly to changes in operating conditions and insufficient stability of the adjustment effect. Furthermore, traditional systems lack a robust operational protection mechanism, making them prone to problems such as parameter exceeding limits and actuator failures. They also lack a complete closed-loop optimization process, have weak technological iteration capabilities, and struggle to continuously improve treatment efficiency, failing to fully meet the industry's demands for resource recycling and efficient treatment. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network. This system acquires sand and mud images and pressure data by deploying a high-speed industrial camera at the outlet of a hydrocyclone, extracts color and edge features using an improved Mask R-CNN neural network, locates the sand and mud interface and evaluates sand particle purity using a visual-pressure fusion algorithm, dynamically corrects control parameters based on purity level and historical data using a purity memory self-tuning algorithm, generates adjustment commands for inlet pressure and overflow pipe height, and executes closed-loop regulation through multiple protection mechanisms.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network, the system comprising: Visual acquisition module: High-speed industrial cameras are deployed at the sand and sludge discharge ports of the hydrocyclone. The cameras are equipped with polarizing filters to eliminate water mist interference, and waterproof and dustproof protective covers are installed on the outside and anti-fogging devices are set up. The cameras acquire high-definition images of sand and sludge flow and real-time inlet pressure data of the hydrocyclone. The acquired images are marked with timestamps and invalid frames are automatically filtered. Sand and mud identification module: The acquired images are preprocessed, and the color and edge features of sand and mud are extracted by improving the Mask R-CNN neural network. The location of the sand and mud separation interface is determined by the visual-pressure fusion positioning algorithm. At the same time, a sand purity evaluation system is constructed to complete the calculation of sand and sludge adhesion rate and purity level determination. Algorithm decision module: retrieves historical adjustment data, uses purity memory self-tuning algorithm to generate control parameter self-tuning coefficients, corrects PID+fuzzy control parameters, and generates adjustment instructions for hydrocyclone operation parameters by combining the sand-mud separation interface position and sand purity level. The execution adjustment module receives adjustment commands and dynamically adjusts the inlet pressure and overflow pipe height of the hydrocyclone through a dedicated actuator. Multiple protection mechanisms are set up during the adjustment process to ensure stable and reliable adjustment. After the adjustment is completed, the real-time operating status of the actuator is recorded simultaneously. Closed-loop optimization module: Feeds back the operating status and adjustment results of the actuator to the vision acquisition module to form a complete control closed loop. At the same time, it continuously stores the entire process operation data, analyzes the data regularly, and iteratively optimizes the relevant algorithms and control logic to ensure the long-term stable operation of the system.

[0006] Furthermore, the visual acquisition module, with high-speed industrial cameras deployed at the sand and sludge discharge ports of the hydrocyclone, features a polarizing filter on the camera lens to eliminate water mist interference. A waterproof and dustproof protective cover with an internal anti-fogging device is also installed on the camera's exterior. The high-speed industrial camera uses a global shutter to acquire high-definition images of the sand and sludge flow, simultaneously acquiring real-time inlet pressure data of the hydrocyclone. The acquired high-definition images are timestamped, and invalid frames are automatically filtered. The high-definition images and real-time inlet pressure data are then categorized and stored. The high-definition images contain morphological features of the sand and sludge, while the real-time inlet pressure data contains real-time pressure changes. The images and pressure data are transmitted via gigabit wired network to the subsequent processing module, ensuring no data loss or tampering during transmission, providing raw data support for subsequent sand and sludge feature identification and purity evaluation.

[0007] Furthermore, the improved Mask R-CNN neural network used in the sand and mud identification module adds a color feature branch to the original network structure while retaining the original edge feature extraction structure. It first receives a pre-processed high-definition image of the sand and mud flow, performs preliminary feature extraction on the image through the network backbone, and then filters and analyzes the color information of each pixel in the image through the newly added color feature branch. It extracts the color feature information corresponding to sand particles and sludge and converts it into color similarity matching values. Simultaneously, it captures the contour edges in the sand and mud flow image through the original edge feature extraction structure of the network, extracts the sand-mud interface and the edge contour information of sand particles and sludge, and converts it into edge gradient values. The extracted color feature information and edge feature parameters are fused to output standardized sand and mud feature data. This provides feature support for the visual-pressure fusion localization algorithm to determine the location of the sand-mud separation interface and provides basic feature data for the sand particle purity evaluation system to calculate the sand-sludge adhesion rate and determine the purity level.

[0008] Furthermore, in the sand and mud identification module, the mathematical expression of the vision-pressure fusion localization algorithm is:

[0009] in, This is the dynamic positioning coefficient of the sand-mud separation interface, with a value range of 0-1. The closer the value is to 1, the closer the interface is to the suitable position. The color similarity matching value of the sand and mud region in the visual image is output by the improved Mask R-CNN neural network, and the value range is 0-1. This represents the edge gradient value of the sand-mud interface, ranging from 0 to 1. This is the deviation between the inlet pressure and the rated pressure of the hydrocyclone. The rated inlet pressure of the hydrocyclone; These are dynamic weighting coefficients, adaptively adjusted by the system based on real-time operating conditions, and satisfying the following conditions: .

[0010] Furthermore, the sand and mud identification module and the sand particle purity evaluation system are multi-dimensional evaluation mechanisms used to quantitatively assess the degree of sludge adhesion on the sand particle surface and classify sand particle purity levels. Based on high-definition images of sand and mud flow acquired by the visual acquisition module and the sand-mud separation interface position output by the visual-pressure fusion positioning algorithm, the system is constructed as follows: Independent sand particle regions are segmented from the pre-processed sand and mud flow image using an improved MaskR-CNN neural network; the total projected area of ​​each sand particle and the area of ​​the surface sludge adhesion region are extracted; and the sand particle sludge adhesion rate, i.e., the ratio of the sludge adhesion region area to the total projected area of ​​the sand particle, is calculated. Simultaneously, a dynamic positioning coefficient of the sand and mud separation interface is introduced to correct the adhesion rate, thereby eliminating the interference of interface position fluctuations on purity evaluation. Subsequently, different purity levels are classified according to the purity requirements for sand medium recycling.

[0011] Furthermore, in the sand and sludge identification module, the purity level is determined by quantitatively classifying the sand based on preset multi-range sludge adhesion rate thresholds. Specifically, based on the sludge adhesion rate of the sand particles, the percentage of sludge area on the surface of the sand particles is calculated and output, with four preset continuous threshold ranges. Among them, the high purity level corresponds to a sludge adhesion rate of <5%, which means that the surface of the sand particles is clean and there is almost no sludge adhesion, and it can directly meet the reuse requirements; the medium purity level corresponds to a sludge adhesion rate of 5% ≤ sludge adhesion rate < 15%, which means that there is a small amount of sludge adhesion on the surface of the sand particles, and it can be reused after self-purification by optimizing the inlet pressure parameters of the hydrocyclone; the low purity level corresponds to a sludge adhesion rate of 15% ≤ sludge adhesion rate < 30%, which means that there is a lot of sludge adhesion on the surface of the sand particles, and it is necessary to trigger the control command to adjust the operating conditions of the hydrocyclone; the poor purity level corresponds to a sludge adhesion rate of ≥ 30%, which means that the surface of the sand particles is covered with a large amount of sludge, and it is necessary to carry out deep treatment or replace the sand particles.

[0012] Furthermore, in the algorithm decision module, the mathematical expression for the purity memory self-tuning algorithm is:

[0013] in, The self-tuning coefficients of the control parameters are directly used for parameter correction in PID+fuzzy control and are the core adjustment coefficients of the algorithm output. The real-time sand particle sludge adhesion rate reflects the current degree of sludge adhesion on the surface of the sand particles; The target sludge adhesion rate is the desired target for sand particle cleanliness. The dynamic positioning coefficient of the sand-mud separation interface is output by the vision-pressure fusion positioning algorithm, reflecting the degree of deviation between the sand-mud separation interface and the suitable position. The historical adjustment effect coefficient is calculated from the purity change rate of the past 5 adjustments, and its value ranges from 0 to 1. It is used to incorporate historical adjustment experience to optimize the current control strategy. , , For the weighting coefficients, satisfying + + =1, used to balance the influence of sludge adhesion rate deviation, interface positioning deviation and historical adjustment effect on the control coefficient.

[0014] Furthermore, in the algorithm decision module, the adjustment command is a quantitative control command generated by the algorithm decision module, used to dynamically adjust the operating parameters of the hydrocyclone. It uses the self-tuning coefficient of the control parameters generated by the purity memory self-tuning algorithm as the core correction basis, combined with the sand-sludge separation interface position, sand particle purity level, and historical adjustment effect coefficients. It includes three core parameters: inlet pressure adjustment amplitude, overflow valve opening correction value, and underflow discharge frequency. When the sand particle purity is low or poor, the command prioritizes increasing the hydrocyclone inlet pressure to enhance the centrifugal separation effect. When the sand-sludge separation interface deviates from the appropriate position, the overflow valve opening is adjusted synchronously to optimize the flow field distribution. The underflow discharge frequency is dynamically adapted according to the real-time sludge adhesion rate to ensure stable recovery of high-purity sand particles. All adjustment commands are sent to the execution unit in real time via the industrial bus, with an execution priority tag to ensure priority response of key parameters, achieving adaptive optimization of the hydrocyclone's operating state.

[0015] Furthermore, the execution adjustment module incorporates a multi-layered protection mechanism comprising five protection logics: parameter over-limit protection, actuator fault protection, adjustment rate limitation, emergency fault-tolerant switching, and status feedback verification. Parameter over-limit protection presets an upper limit of 0.8 MPa for the hydrocyclone inlet pressure and a ±200 mm limit for the overflow pipe height adjustment. When the adjustment command triggers parameters approaching the threshold, it automatically cuts off the adjustment range and triggers an early warning to prevent equipment overload. Actuator fault protection monitors the electric actuator status in real time using current and position sensors. When jamming or overload signals are detected, it immediately pauses adjustment and switches to manual emergency mode, while simultaneously uploading a fault alarm. The adjustment rate limitation sets an upper limit of ≤0.02 MPa / s for the inlet pressure change rate and ≤5 mm / s for the overflow pipe height change rate to prevent equipment impact caused by sudden flow field changes. The emergency fault-tolerant switching configuration includes a primary and backup actuator. When the primary actuator fails, it automatically switches to the backup actuator to maintain adjustment capability. The status feedback verification verifies the execution effect after adjustment using visual and pressure sensors. If the deviation exceeds 5%, it automatically triggers a secondary correction, ensuring stable and reliable adjustment throughout the entire process.

[0016] Compared with existing technologies, this high-efficiency sedimentation tank control system for sand media based on visual recognition neural networks has the following advantages: I. This invention deploys visual acquisition equipment at the key discharge port of a hydrocyclone, optimizes image filtering and data synchronous acquisition mechanisms, combines improved neural network extraction of sand and mud color and edge features, and integrates pressure data to construct a positioning algorithm. This achieves precise capture of the sand-mud separation interface and multi-dimensional quantitative evaluation of sand particle purity. By leveraging the dual extraction of color and edge features and the deep fusion of visual and pressure data, it breaks through the limitations of traditional single detection methods, improves the accuracy of sand and mud identification and the scientific nature of purity determination, accurately controls the sludge adhesion state on the sand particle surface, and provides a clear basis for subsequent control through a graded and well-defined purity evaluation system. This avoids inaccurate adjustments due to identification deviations, effectively ensures the sand-mud separation effect, lays a reliable foundation for sand-mud recycling, improves the stability and accuracy of the sedimentation process, and promotes the development of sand-mud sedimentation treatment towards refinement and intelligence.

[0017] II. This invention integrates historical adjustment data with real-time monitoring information, uses a purity memory self-tuning algorithm to correct control parameters, and generates targeted adjustment commands based on relevant control logic. This achieves dynamic adaptation of hydrocyclone operating parameters and relies on multiple protection mechanisms to prevent equipment risks and parameter over-limit issues during the adjustment process, ensuring the safety and continuity of system operation. Simultaneously, a closed-loop feedback mechanism continuously stores full-process operating data and iteratively optimizes the algorithm, driving continuous improvement of the control logic. This collaborative model, encompassing identification, decision-making, and execution, can dynamically adjust the control strategy based on sand purity levels and separation interface conditions, while continuously improving control accuracy through data iteration and optimization. This enhances the efficiency and quality of sand media sedimentation treatment, reduces the need for manual intervention, and strengthens the system's adaptability to different operating conditions, providing strong support for the stable operation of high-efficiency sand media sedimentation tanks.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0020] Figure 1 This is a flowchart of a high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network. Figure 2This is a data transmission diagram of a high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network. Figure 3 This is a schematic diagram of data transmission for sand and mud feature extraction using the improved Mask R-CNN neural network of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] Example 1: Sand media sedimentation treatment scenario in urban wastewater treatment plants This embodiment applies to a high-efficiency sand media sedimentation tank in an urban wastewater treatment plant. In this scenario, the wastewater composition is complex, containing domestic sludge, organic impurities, etc., which easily adhere to the surface of the sand particles, affecting the quality of sand media recycling. Furthermore, the wastewater treatment volume varies over time, and operating conditions fluctuate frequently. Therefore, real-time monitoring and precise control of sand-sludge separation are necessary to ensure treatment efficiency and resource recovery. Figure 1 As shown.

[0023] After system startup, the visual acquisition module operates according to the preset plan. High-speed industrial cameras are fixedly deployed at designated locations at the sand and sludge discharge ports of the hydrocyclone. This deployment method directly corresponds to the discharge path after the separation of sand and sludge, enabling immediate capture of the sand and sludge flow status at the separation terminal and ensuring that the acquired data directly reflects the separation effect. A polarizing filter is installed at the camera lens to specifically eliminate water mist formed by water vapor evaporation during wastewater treatment, preventing image blurring and ensuring clear presentation of the morphological characteristics and color differences of the sand and sludge flow. This allows for accurate capture of key information during subsequent feature extraction. A waterproof and dustproof protective cover installed on the outside of the camera, along with a built-in anti-fogging device, provides dual protection, specifically designed for the humid and dusty operating environment of wastewater treatment plants. This effectively prevents damage to the camera from wastewater splashes and dust corrosion, while also preventing fogging of the lens due to changes in ambient temperature and humidity, extending the camera's lifespan and preventing data acquisition interruptions due to equipment failure. High-resolution images of sand and sludge flow are continuously acquired using a high-speed industrial camera with a global shutter. The global shutter design rapidly captures the instantaneous state of the high-speed sand and sludge flow, avoiding motion blur and ensuring the complete preservation of the morphological details and color differences of sand particles and sludge in the images. Even when faced with changes in sand and sludge flow velocity caused by fluctuations in wastewater flow rate, clear images can be stably acquired. Real-time inlet pressure data from the hydrocyclone is simultaneously acquired. This data directly reflects the operating conditions of the hydrocyclone, providing key physical parameters for subsequent fusion analysis. Each frame of the high-resolution sand and sludge flow image is automatically timestamped, establishing a precise time correspondence between the image and the corresponding inlet pressure data. This provides a time reference for subsequent multi-dimensional data fusion analysis, ensuring data correlation. Simultaneously, invalid frames are automatically filtered out, eliminating abnormal images caused by equipment vibration, sudden changes in lighting, instantaneous water flow impact, etc., reducing interference from invalid data in subsequent processing and improving data processing efficiency. The high-definition images of sand and sludge flow and the real-time inlet pressure data of the hydrocyclone are classified and stored. The image data includes visual information such as the morphological characteristics and color distribution of sand and sludge, while the pressure data includes the real-time changes in inlet pressure. This classified storage facilitates subsequent targeted processing. At the same time, the data is transmitted to the subsequent processing module via gigabit wired network. The gigabit network has a stable transmission rate and strong anti-interference capability, ensuring that a large number of high-definition images and real-time pressure data can be transmitted quickly and without delay. This provides timely data support for real-time monitoring and control, and avoids untimely control due to data transmission lag.

[0024] The sand and mud identification module receives pre-processed high-resolution images of sand and mud flows. The pre-processing stage has completed image noise reduction and enhancement operations, eliminating interference for subsequent feature extraction. Feature extraction of the image is performed by improving the Mask R-CNN neural network. This network adds a color feature branch to the original structure while retaining the original edge feature extraction structure, forming a dual feature extraction mechanism. The network backbone first performs preliminary feature extraction on the image, filtering redundant information to lay the foundation for subsequent accurate feature separation. A newly added color feature branch specifically filters and analyzes the color differences between sand and sludge, accurately capturing subtle differences in hue and saturation. Even when faced with color fluctuations caused by changes in sludge concentration, it can stably identify these differences. The converted color similarity matching value provides an intuitive quantitative basis for distinguishing between sand and sludge, avoiding misjudgments that may occur with single edge feature extraction. The original edge feature extraction structure focuses on capturing the contour morphology of sand-sludge flow, accurately identifying the dynamic changes at the sand-sludge separation interface and the edge contours of individual sand grains and sludge clumps. Even when a small amount of sludge adheres to the sand grain surface, causing blurred edges, it can effectively capture its complete contour. The converted edge gradient value clearly reflects the smoothness of the interface, the integrity of the sand grains, and the shape and size of the sludge clumps. The extracted color feature information and edge feature parameters are fused, integrating two key types of information from the visual dimension to overcome the limitations of single feature extraction. This makes the output standardized sand-sludge feature data more comprehensive and representative, providing reliable data support for subsequent separation interface localization and purity evaluation. Subsequently, a visual-pressure fusion localization algorithm was used to determine the location of the sand-mud separation interface. The mathematical expression of the visual-pressure fusion localization algorithm is as follows:

[0025] in, The dynamic positioning coefficient of the sand-mud separation interface; The color similarity matching value of the sand and mud regions in the visual image; The edge gradient value of the sand-mud interface; This is the deviation between the inlet pressure and the rated pressure of the hydrocyclone. The rated inlet pressure of the hydrocyclone; Using dynamic weighting coefficients, this algorithm organically combines visually acquired color similarity matching values, edge gradient values, and real-time inlet pressure data. This overcomes the limitations of single visual or single pressure detection, enabling dynamic perception of the impact of changing operating conditions on the separation interface. Even when faced with inlet pressure changes caused by fluctuations in wastewater flow, it can accurately pinpoint the separation interface location, ensuring that the positioning results are not affected by fluctuations in a single factor, significantly improving stability and accuracy. Simultaneously, a sand purity evaluation system is constructed. This system, based on high-definition images acquired by the visual acquisition module and the sand-sludge separation interface location output by the visual-pressure fusion positioning algorithm, is a multi-dimensional mechanism for quantitatively assessing the degree of sludge adhesion on the sand particle surface. By improving the Mask R-CNN neural network, independent sand grain regions are segmented from the preprocessed sand-sludge flow image, which can accurately isolate individual sand grains and avoid mutual interference between sand grains or between sand grains and sludge clumps, ensuring that the analysis of each sand grain is not affected by the surrounding environment. The total projected area of ​​each sand grain and the area of ​​the sludge-attached area on the surface are extracted to calculate the sand grain sludge adhesion rate. This ratio directly reflects the cleanliness of the sand grain surface. At the same time, a dynamic positioning coefficient of the sand-sludge separation interface is introduced to correct the adhesion rate, so that the purity assessment results not only reflect the surface state of the sand grains, but also relate to the overall separation effect of the separation interface, making the assessment more comprehensive and scientific. Based on the purity requirements for sand media recycling, four different purity levels are defined. Quantitative grading is achieved through preset multi-range sludge adhesion rate thresholds. High purity levels correspond to low sludge adhesion rates, indicating clean sand surfaces suitable for direct reuse; medium purity levels correspond to small amounts of sludge adhesion, which can be autonomously purified and reused after parameter optimization; low purity levels correspond to significant sludge adhesion, requiring triggering control commands; and the lowest purity levels correspond to extensive sludge coating, requiring deep treatment. Clear grading standards provide a clear basis for subsequent control, avoiding blind adjustments. Figure 3 As shown.

[0026] After the algorithm decision-making module is activated, it first retrieves historical adjustment data. This data includes the sand particle purity level, separation interface location, control parameters, and final treatment effects under different past operating conditions. This data can provide a reference for the control of the current operating conditions, reduce repeated trial and error, and improve control efficiency. The purity memory self-tuning algorithm is used to generate the self-tuning coefficients of the control parameters. The mathematical expression of the purity memory self-tuning algorithm is:

[0027] in, These are the self-tuning coefficients for the control parameters; The real-time sand particle sludge adhesion rate reflects the current degree of sludge adhesion on the surface of the sand particles; The target sludge adhesion rate; The dynamic positioning coefficient of the sand-mud separation interface; This represents the historical adjustment effect coefficient. , , Using the weighting coefficient, this algorithm comprehensively considers the difference between the real-time sand-sludge adhesion rate and the target adhesion rate, the dynamic positioning coefficient of the sand-sludge separation interface, and the historical adjustment effect coefficient. The generated self-tuning coefficient can accurately reflect the deviation between the current operating condition and the ideal state, while also taking into account historical control experience to ensure the scientific nature and adaptability of the coefficient. Based on this self-tuning coefficient as the core correction basis, the PID+fuzzy control parameters are modified, allowing the control logic to possess both the precision of PID control, enabling stable tracking of the target value, and the flexibility of fuzzy control to cope with complex operating conditions, quickly adapting to fluctuations in operating conditions and avoiding lag or overshoot in fixed parameters when operating conditions change. Combining the location of the sand-sludge separation interface and the sand purity level, adjustment instructions for the hydrocyclone operating parameters are generated. These instructions include three core parameters: inlet pressure adjustment range, overflow valve opening correction value, and underflow discharge frequency, forming a comprehensive control scheme. When the sand particles are of low or poor purity, the command prioritizes increasing the inlet pressure of the hydrocyclone to enhance the sand-sludge separation effect by increasing centrifugal force and quickly peeling off the sludge attached to the surface of the sand particles. When the sand-sludge separation interface deviates from the appropriate position, the overflow valve opening is adjusted simultaneously to optimize the internal flow field distribution of the hydrocyclone, allowing the sand-sludge separation to take place in a more reasonable flow field environment and improving the separation efficiency. The discharge frequency at the underflow outlet is dynamically adapted according to the real-time sludge adhesion rate to avoid excessive discharge causing sand particle loss or excessively slow discharge causing sludge accumulation that affects the separation effect, ensuring that the control measures are targeted and effective.

[0028] After receiving the adjustment commands generated by the algorithm decision module, the execution adjustment module dynamically adjusts the inlet pressure and overflow pipe height of the hydrocyclone through a dedicated actuator. This dedicated actuator boasts fast response and high adjustment precision, accurately executing the parameter adjustment requirements in the commands to ensure timely implementation of control measures. Multiple protection mechanisms are implemented during the adjustment process, forming a five-layer protection logic to comprehensively ensure system operational safety and adjustment effectiveness. Parameter over-limit protection presets reasonable operating thresholds. When the adjustment command triggers parameters approaching the threshold, it automatically cuts off the adjustment range and triggers an early warning to prevent excessive inlet pressure or excessive overflow pipe height adjustment from causing equipment overload and avoiding damage to the hydrocyclone due to excessive pressure. Actuator fault protection monitors the electric actuator status in real time through current and position sensors. When abnormal signals such as jamming or overload are detected, adjustment is immediately paused and switched to manual emergency mode, while simultaneously uploading a fault alarm for timely troubleshooting by staff, preventing the fault from escalating and affecting the overall system operation. Adjustment rate limiting sets reasonable upper limits for the inlet pressure change rate and overflow pipe height change rate to ensure parameter... The adjustment process ensures a smooth transition, preventing drastic fluctuations in the internal flow field of the hydrocyclone due to sudden parameter changes, which could affect sand and mud separation. It also reduces mechanical wear caused by frequent and drastic adjustments. An emergency fault-tolerant switching configuration with primary and backup actuators automatically switches to the backup actuator in case of primary failure, maintaining uninterrupted adjustment capability and ensuring continuous system operation under complex conditions. Status feedback verification, after adjustment, uses both visual acquisition module images and pressure sensor data to verify the execution effect. If the deviation exceeds the set range, a secondary correction is automatically triggered to ensure the control measures truly achieve the expected results and prevent control failure due to actuator errors or sudden changes in operating conditions. The real-time operating status of the actuators is recorded synchronously after adjustment, including adjusted parameter values, actuator operating current, and position information, providing a basis for subsequent data traceability and algorithm optimization.

[0029] The closed-loop optimization module feeds back the operating status and adjustment results of the actuators to the visual acquisition module, forming a complete control closed loop. This allows the system to perceive the control effect in real time and dynamically adjust the focus and frequency of subsequent data acquisition based on the effect, achieving closed-loop management of the entire process from data acquisition, analysis, decision-making, execution to feedback. Simultaneously, it continuously stores all process operation data, including image data, pressure data, feature extraction results, control parameters, execution status, and adjustment effects at each step, accumulating a massive amount of operational cases and data materials. Regular analysis of the stored data uncovers the correlation between changes in operating conditions and control effects, identifies optimization space in algorithms and control logic, and iteratively optimizes relevant algorithms and control logic, allowing the system to continuously improve recognition accuracy, control accuracy, and operational stability over long-term operation. Through continuous optimization, the system can better adapt to changes in influent water quality and quantity at different times in urban wastewater treatment plants, continuously ensuring the sand media separation effect and reuse quality, improving the overall efficiency of the sedimentation treatment process, reducing the need for manual intervention and operation and maintenance costs, and providing solid support for the efficient and intelligent operation of urban wastewater treatment.

[0030] Example 2: Mining wastewater purification sand media separation scenario This embodiment applies to a high-efficiency sedimentation tank using sand media in the purification process of mine wastewater. Mine wastewater, due to mining operations, contains a large amount of mineral sludge, rock fragments, and other impurities. The sludge has a complex composition and strong viscosity, and mineral sludge easily adheres to the surface of sand particles, forming a coating layer. This necessitates strict requirements for the purity of the recycled sand media. Simultaneously, the wastewater discharge and pollutant concentration fluctuate significantly during mining operations, resulting in complex and variable operating environments. Some areas also experience high levels of water mist and dust concentration, placing higher demands on the adaptability, stability, and anti-interference capabilities of the control system. Figure 2 As shown.

[0031] During system operation, the vision acquisition module is activated first and begins data acquisition. High-speed industrial cameras are deployed at preset positions at the sand and sludge discharge ports of the hydrocyclone. These positions directly correspond to the discharge paths of the separated sand and sludge, enabling immediate capture of the sand and sludge flow status at the separation terminal. This ensures that the acquired data directly reflects the separation effect, providing first-hand information for subsequent accurate analysis. Addressing the harsh environment of mine wastewater treatment sites with high levels of water mist and dust, a polarizing filter is installed at the camera lens to eliminate the refraction and scattering of light by water mist, preventing image clarity degradation caused by water mist. This ensures that the color differences and morphological details of sand and mineral sludge are clearly distinguishable, allowing for stable acquisition of high-quality images even in high-humidity environments. A waterproof and dustproof protective cover on the outside of the camera effectively blocks dust erosion and wastewater splashes generated during mining operations, while a built-in anti-fogging device keeps the inside of the lens dry, preventing fogging due to changes in ambient temperature and humidity. This dual-protection structure ensures long-term stable operation of the camera in complex and harsh mining environments, reducing equipment maintenance frequency and downtime, and guaranteeing continuous data acquisition. High-resolution images of sand and mud flows are acquired using a high-speed industrial camera with a global shutter. The global shutter design allows for rapid capture of the instantaneous state of high-speed sand and mud flows, addressing velocity variations caused by fluctuations in mine wastewater flow, avoiding motion blur, and ensuring that each frame fully preserves the morphological characteristics and positional relationships of sand particles and sludge. Even in the face of turbulent sand and mud flows caused by instantaneous water flow impacts, key details can be captured. Simultaneously, real-time inlet pressure data from the hydrocyclone is acquired. This data directly reflects the hydrocyclone's operating conditions and provides timely feedback on fluctuations, offering crucial physical parameter support for subsequent fusion analysis. The acquired high-resolution images of the sand and mud flows are timestamped, establishing a precise temporal correspondence between each frame and its corresponding inlet pressure data. This provides a reliable time reference for subsequent visual-pressure data fusion analysis, ensuring the correlation and consistency of multi-source data. Invalid frames are automatically filtered out, specifically removing abnormal images caused by sudden changes in operating conditions, equipment vibration, drastic light changes, or momentary dust obstruction, preventing invalid data from consuming processing resources and improving the efficiency and accuracy of subsequent data processing. The high-definition images of sand and mud flow collected are classified and stored together with the real-time inlet pressure data of the hydrocyclone. The image data includes visual information such as the morphological characteristics, color distribution, and adhesion status of sand and mineral sludge, while the pressure data includes the real-time changes in inlet pressure. This classified storage facilitates subsequent targeted processing. At the same time, the data is transmitted to the subsequent processing module via gigabit wired network. Gigabit network transmission has the advantages of high transmission rate and strong anti-interference capability, which can quickly and stably transmit a large number of high-definition images and real-time pressure data, avoiding data transmission interruptions or delays caused by the complex electromagnetic environment at the mine site. This ensures that the subsequent processing module can obtain data in a timely manner, providing time guarantee for real-time control.

[0032] The sand and mud identification module receives pre-processed high-resolution images of sand and mud flows. The pre-processing stage includes image denoising, enhancement, and dehazing, effectively improving image quality and eliminating interference for subsequent feature extraction. Feature extraction is performed on the images using an improved Mask R-CNN neural network. This network adds a color feature branch to the original structure while retaining the original edge feature extraction structure, forming a dual feature extraction mechanism specifically designed to address the complex composition of mine wastewater sludge. The network backbone first performs preliminary feature extraction on the image, filtering redundant information to lay the foundation for subsequent accurate feature separation. The newly added color feature branch is specifically designed to screen and analyze the color differences between mineral sludge and sand particles in mine wastewater. Mineral sludge and sand particles have obvious differences in hue and brightness. Even when faced with color fluctuations caused by changes in sludge concentration, this branch can accurately capture the differences. The converted color similarity matching value provides an intuitive quantitative basis for distinguishing between sand and sludge, effectively avoiding confusion caused by the complex composition and similar colors of sludge. The original edge feature extraction structure focuses on capturing the contour morphology of sand and sludge flow. The complex composition of mine wastewater sludge leads to irregular sand and sludge separation interfaces and blurred edge contours after sludge adheres to the surface of sand particles. This structure can penetrate these interferences and accurately identify the dynamic changes of the sand and sludge separation interface, the complete contour of a single sand particle, and the edge morphology of sludge clumps. Even when sand particles are partially covered by sludge, their core contours can be effectively captured. The converted edge gradient value can clearly reflect the fluctuation of the interface, the integrity of the sand particles, and the size and shape of the sludge clumps. The extracted color features and edge parameters are fused to integrate the two core types of visual information, overcoming the limitations of single feature extraction. This results in more comprehensive and accurate standardized sand and mud feature data, providing solid data support for subsequent separation interface location and purity evaluation. A visual-pressure fusion positioning algorithm is then used to determine the sand-mud separation interface location. This algorithm organically combines visually acquired color similarity matching values, edge gradient values, and real-time inlet pressure data. Given the large fluctuations in mine operating conditions leading to easily changing inlet pressure, this algorithm dynamically correlates pressure changes with visual features to accurately pinpoint the sand-mud separation interface location. Even in the face of sudden flow surges or pressure fluctuations, it can quickly adjust the positioning results, avoiding positioning deviations that occur with single visual or pressure detection under fluctuating operating conditions, ensuring the stability and accuracy of interface positioning. Simultaneously, a sand purity evaluation system is constructed. This system, based on high-definition images acquired by the visual acquisition module and the sand-mud separation interface location output by the visual-pressure fusion positioning algorithm, is a multi-dimensional mechanism for quantitatively evaluating the degree of mineral sludge adhesion on the sand particle surface.By improving the Mask R-CNN neural network to segment independent sand grain regions from preprocessed sand-sludge flow images, it can effectively isolate sand grains that are wrapped in sludge or adhere to each other, avoiding interference between sand grains and between sand grains and sludge clumps. This ensures that the morphological characteristics and sludge adhesion of each sand grain can be analyzed individually, and accurate segmentation can be achieved even when facing sand grain aggregation in high-concentration sludge environments. The total projected area of ​​each sand grain and the area of ​​the sludge-attached area on its surface are extracted to calculate the sand grain sludge adhesion rate. This ratio directly reflects the cleanliness of the sand grain surface and can quantitatively assess the adhesion of mineral sludge. At the same time, a dynamic positioning coefficient of the sand-sludge separation interface is introduced to correct the adhesion rate, so that the purity assessment results not only reflect the surface state of the sand grains, but also relate to the overall separation effect of the separation interface, making the assessment more comprehensive and scientific, and avoiding the neglect of the overall working condition due to focusing only on individual sand grains. Based on the stringent requirements for sand media recycling, four different purity levels are defined. Quantitative grading is achieved through preset multi-range sludge adhesion rate thresholds. High purity level corresponds to a low sludge adhesion rate, indicating clean sand particle surfaces that directly meet recycling requirements. Medium purity level corresponds to a small amount of sludge adhesion, which can be autonomously purified and reused after optimizing hydrocyclone parameters. Low purity level corresponds to a large amount of sludge adhesion, requiring triggering control commands to adjust operating conditions. Low purity level corresponds to a large amount of sludge coating, requiring deep treatment. Clear grading standards can quickly identify sand particles requiring enhanced control or deep treatment, providing a clear basis for subsequent precise control and ensuring that the purity of recycled sand media meets mine production requirements.

[0033] After the algorithm decision-making module is activated, it first retrieves historical adjustment data. This data includes sand particle purity levels, separation interface locations, control parameters, and final treatment effects under different mine operating conditions. It covers control experience under different wastewater discharge volumes and pollutant concentrations, providing a reference for current operating conditions, reducing trial-and-error costs, improving control efficiency, and avoiding blind adjustments when facing new operating conditions. A purity memory self-tuning algorithm is used to generate self-tuning coefficients for control parameters. This algorithm comprehensively considers the difference between real-time sand particle-sludge adhesion rate and target adhesion rate, the dynamic positioning coefficient of the sand-sludge separation interface, and historical adjustment effect coefficients. The generated self-tuning coefficients accurately reflect the deviation between the current operating conditions and the ideal state, while also taking into account historical control experience, ensuring the scientific validity and adaptability of the coefficients, and can specifically address the characteristics of large fluctuations in mine operating conditions. Using this self-tuning coefficient as the core correction basis, the PID+fuzzy control parameters are modified, allowing the control logic to possess both the precision of PID control, enabling stable tracking of target values, and the flexibility of fuzzy control to handle complex operating conditions. This allows it to quickly adapt to fluctuations in mine wastewater flow and pollutant concentration, avoiding lag or overshoot in fixed parameters when operating conditions change, thus ensuring the stability and accuracy of control. Control commands are generated by combining the location of the sand-sludge separation interface, sand particle purity level, and historical adjustment effect coefficients. These commands include three core parameters: inlet pressure adjustment range, overflow valve opening correction value, and underflow discharge frequency, forming a comprehensive control scheme. For low-purity and inferior-purity sand particles commonly found in mine wastewater treatment, the instructions prioritize increasing the inlet pressure of the hydrocyclone. By increasing centrifugal force, the sand-sludge separation effect is enhanced, quickly peeling off the mineral sludge adhering to the surface of the sand particles. This specifically addresses the problem of strong adhesion and difficulty in peeling off mineral sludge. If the sand-sludge separation interface deviates from the appropriate position, the overflow valve opening is adjusted simultaneously to optimize the internal flow field distribution of the hydrocyclone, improve the separation environment, and enhance separation efficiency, allowing sand and sludge to achieve efficient separation in a more reasonable flow field. The discharge frequency at the underflow outlet is dynamically adapted according to the real-time sludge adhesion rate to avoid sludge accumulation or sand particle loss due to improper discharge frequency, ensuring the continuity and stability of the separation process and adapting to the large fluctuations in sludge concentration in mine wastewater.

[0034] After receiving the adjustment commands generated by the algorithm decision-making module, the execution adjustment module dynamically adjusts the inlet pressure and overflow pipe height of the hydrocyclone through a dedicated actuator. This specially adapted actuator is designed to withstand harsh mining environments, offering fast response times and high adjustment precision. It accurately executes the parameter adjustment requirements in the commands, ensuring timely implementation of control measures and rapid response to changes in operating conditions. Multiple protection mechanisms are implemented during the adjustment process, forming a five-layer protection logic to comprehensively guarantee system operational safety and adjustment effectiveness, specifically addressing the challenges of complex mining environments and equipment susceptibility to failure. The parameter over-limit protection system presets reasonable operating thresholds. When the adjustment command triggers a parameter approaching the threshold, it automatically cuts off the adjustment range and triggers an early warning to prevent excessive inlet pressure or excessive overflow pipe height adjustment from causing equipment overload, thus avoiding damage to the hydrocyclone due to excessive pressure and ensuring safe equipment operation. The actuator fault protection system monitors the operating status of the electric actuator in real time through current and position sensors. Harsh mining environments can easily lead to actuator jamming and overload. This protection mechanism can detect abnormal signals immediately, suspend adjustment, switch to manual emergency mode, and upload fault alarms for timely troubleshooting and to prevent the fault from escalating and affecting the overall system operation. The adjustment rate limit sets reasonable upper limits for the inlet pressure change rate and overflow pipe height change rate to ensure smooth parameter adjustment transitions and avoid damage caused by excessive pressure. Sudden parameter changes cause drastic fluctuations in the internal flow field of the hydrocyclone, affecting the sand-sludge separation effect. Simultaneously, reducing mechanical wear caused by frequent and drastic adjustments extends equipment lifespan and lowers on-site maintenance costs in the mine. An emergency fault-tolerant switching configuration with primary and backup actuators ensures seamless takeover in case of primary failure, guaranteeing uninterrupted adjustment and continuous system operation under complex conditions, preventing wastewater treatment interruptions due to equipment failure. Status feedback verification, after adjustment, uses high-definition images from a visual acquisition module and data from pressure sensors to verify the execution effect, ensuring the control measures achieve the expected goals. If the deviation exceeds the set range, a secondary correction is automatically triggered, preventing control failure due to actuator errors or sudden changes in operating conditions, and ensuring the stability of sand-sludge separation. Real-time operating status of the actuators is recorded synchronously after adjustment, including adjusted parameter values, operating current, position information, and operating temperature, providing detailed data for subsequent data traceability and algorithm optimization.

[0035] The closed-loop optimization module feeds back the operating status and adjustment results of the actuators to the vision acquisition module, forming a complete control closed loop. This allows the system to perceive the control effect in real time and dynamically adjust the focus and frequency of subsequent data acquisition based on the effect. This achieves closed-loop management of the entire process from data acquisition, analysis, decision-making, execution to feedback, ensuring that each link can be dynamically optimized based on actual results. Simultaneously, it continuously stores the entire process operation data, including image data, pressure data, feature extraction results, control parameters, execution status, and adjustment effects at each step, accumulating a massive amount of operational cases and data materials for mine wastewater treatment. Regular analysis of the stored data uncovers the correlation between changes in different mine operating conditions and control effects, identifies optimization space in algorithms and control logic, and iteratively optimizes related algorithms and control logic. This allows the system to continuously adapt to the characteristics of large fluctuations in mine wastewater treatment conditions and complex sludge composition, continuously improving recognition accuracy, control accuracy, and operational stability. Through continuous optimization, the system can steadily improve the efficiency of sand and mud separation and the purity of recycled sand particles, ensuring that the sand medium always meets the recycling requirements of mine production. At the same time, it reduces the need for manual intervention and equipment wear and tear, lowers the cost of mine wastewater treatment, improves the efficiency of resource recycling, and provides reliable support for the stable operation and green development of mine environmental protection.

[0036] This embodiment precisely adapts to the stringent requirements of sand separation in mine wastewater purification, specifically addressing the challenges of complex sludge composition, large fluctuations in operating conditions, and harsh environments. The visual acquisition module, through enhanced protection and efficient transmission, ensures effective data acquisition and transmission even in harsh environments. The sand and sludge identification module utilizes improved neural networks and fusion algorithms to overcome interference from complex sludge, achieving accurate sand and sludge differentiation and scientific purity grading. The algorithm decision-making module combines historical data and real-time operating conditions to generate control commands adapted to fluctuating scenarios. The execution and adjustment module relies on dedicated mechanisms and multiple protections to ensure safe and continuous control. The closed-loop optimization module enhances the system's adaptability to complex operating conditions through data accumulation and algorithm iteration. The system effectively improves sand and sludge separation efficiency and sand reuse purity, reduces manual intervention and equipment wear, and provides a reliable solution for the stable operation and resource recycling of mine wastewater purification.

[0037] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A control system for a high-efficiency sedimentation tank for sand media based on a visual recognition neural network, characterized in that, The system includes: Visual acquisition module: High-speed industrial cameras are deployed at the sand discharge port and sludge discharge port of the hydrocyclone to acquire high-definition images of sand and sludge flow and real-time inlet pressure data of the hydrocyclone. The acquired images are timestamped and invalid frames are automatically filtered. Sand and mud identification module: The acquired images are preprocessed, and the color and edge features of sand and mud are extracted by improving the Mask R-CNN neural network. The location of the sand and mud separation interface is determined by the visual-pressure fusion positioning algorithm. At the same time, a sand purity evaluation system is constructed to complete the calculation of sand and sludge adhesion rate and purity level determination. Algorithm decision module: retrieves historical adjustment data, uses purity memory self-tuning algorithm to generate control parameter self-tuning coefficients, corrects PID+fuzzy control parameters, and generates adjustment instructions for hydrocyclone operation parameters by combining the sand-mud separation interface position and sand purity level. The control module receives control commands and dynamically adjusts the inlet pressure and overflow pipe height of the hydrocyclone through a dedicated actuator. Multiple protection mechanisms are set up during the control process, and the real-time operating status of the actuator is recorded synchronously after the control is completed. Closed-loop optimization module: Feeds back the operating status and adjustment results of the actuator to the vision acquisition module to form a complete control closed loop. At the same time, it continuously stores the entire process operation data, analyzes the data regularly, and iteratively optimizes the relevant algorithms and control logic.

2. The control system for a high-efficiency sedimentation tank for sand media based on a visual recognition neural network according to claim 1, characterized in that, The vision acquisition module includes high-speed industrial cameras deployed at the sand and sludge discharge ports of the hydrocyclone. Polarizing filters are installed on the camera lenses to eliminate water mist interference, and waterproof and dustproof protective covers with built-in anti-fogging devices are installed on the outside of the cameras. The high-speed industrial cameras acquire high-definition images of the sand and sludge flow through a global shutter, simultaneously acquiring real-time inlet pressure data of the hydrocyclone. The acquired high-definition images of the sand and sludge flow are timestamped, and invalid frames are automatically filtered. The acquired high-definition images of the sand and sludge flow and the real-time inlet pressure data of the hydrocyclone are classified and stored. The high-definition images of the sand and sludge flow contain morphological characteristics of the sand and sludge, while the real-time inlet pressure data of the hydrocyclone contains real-time changes in inlet pressure. The images and pressure data are transmitted to the subsequent processing module via gigabit wired network.

3. The control system for a high-efficiency sedimentation tank for sand media based on a visual recognition neural network according to claim 1, characterized in that, The sand and mud identification module employs an improved Mask R-CNN neural network. This network adds a color feature branch to the original network structure while retaining the original edge feature extraction structure. It first receives a pre-processed high-resolution image of the sand and mud flow, then performs preliminary feature extraction on the image through the network backbone. Next, the newly added color feature branch filters and analyzes the color information of each pixel in the image, extracting the color feature information corresponding to sand particles and sludge, and converting it into color similarity matching values. Simultaneously, the original edge feature extraction structure of the network captures the contour edges in the sand and mud flow image, extracting the sand-mud interface and the edge contour information of sand particles and sludge, and converting it into edge gradient values. The extracted color feature information and edge feature parameters are then fused to output standardized sand and mud feature data.

4. The control system for a high-efficiency sedimentation tank for sand media based on a visual recognition neural network according to claim 1, characterized in that, In the sand and mud identification module, the mathematical expression of the vision-pressure fusion localization algorithm is: ,in, The dynamic positioning coefficient of the sand-mud separation interface; The color similarity matching value of the sand and mud regions in the visual image; The edge gradient value of the sand-mud interface; This is the deviation between the inlet pressure and the rated pressure of the hydrocyclone. This refers to the rated inlet pressure of the hydrocyclone. These are dynamic weighting coefficients.

5. A high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network according to claim 1, characterized in that, The sand and mud identification module uses a sand particle purity evaluation system, a multi-dimensional evaluation mechanism for quantitatively assessing the degree of sludge adhesion on the sand particle surface and classifying sand particle purity levels. It is constructed based on high-definition images of sand and mud flow acquired by the visual acquisition module and the sand-mud separation interface location output by the visual-pressure fusion positioning algorithm. Specifically, it includes: segmenting independent sand particle regions from the pre-processed sand and mud flow image using an improved Mask R-CNN neural network; extracting the total projected area of ​​each sand particle and the area of ​​the surface sludge adhesion region; calculating the sand particle sludge adhesion rate, i.e., the ratio of the sludge adhesion region area to the total projected area of ​​the sand particle; simultaneously introducing a dynamic positioning coefficient of the sand and mud separation interface to correct the adhesion rate; and then classifying different purity levels according to the purity requirements for sand medium recycling.

6. The control system for a high-efficiency sedimentation tank for sand media based on a visual recognition neural network according to claim 1, characterized in that, In the sand and mud identification module, the purity level is determined by setting multiple interval sludge adhesion rate thresholds to achieve quantitative grading. Specifically, based on the sludge adhesion rate of sand particles, the sludge area ratio on the surface of sand particles is calculated and output. Four consecutive threshold intervals are preset. Among them, the high purity level corresponds to a sludge adhesion rate of <5%, which means that the surface of sand particles is clean and there is almost no sludge adhesion, which can directly meet the requirements for reuse. Medium purity level corresponds to 5% ≤ sludge adhesion rate < 15%, indicating that there is a small amount of sludge adhering to the surface of the sand particles. It can be reused after self-purification by optimizing the inlet pressure parameters of the hydrocyclone. Low purity level corresponds to 15% ≤ sludge adhesion rate < 30%, indicating that there is a lot of sludge adhering to the surface of the sand particles. It is necessary to trigger the control command to adjust the operating conditions of the hydrocyclone. Low purity level corresponds to sludge adhesion rate ≥ 30%, indicating that the surface of the sand particles is covered with a large amount of sludge and requires deep treatment.

7. A high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network according to claim 1, characterized in that, In the algorithm decision module, the mathematical expression of the purity memory self-tuning algorithm is: ,in, These are the self-tuning coefficients for the control parameters; The real-time sand particle sludge adhesion rate reflects the current degree of sludge adhesion on the surface of the sand particles; The target sludge adhesion rate; The dynamic positioning coefficient of the sand-mud separation interface; This represents the historical adjustment effect coefficient. , , These are the weighting coefficients.

8. A high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network according to claim 1, characterized in that, In the algorithm decision module, the adjustment command is a quantitative control command generated by the algorithm decision module, used to dynamically adjust the operating parameters of the hydrocyclone. The control parameter self-tuning coefficient generated by the purity memory self-tuning algorithm is used as the core correction basis. It is generated in combination with the sand-sludge separation interface position, sand particle purity level and historical adjustment effect coefficient. It includes three core parameters: inlet pressure adjustment range, overflow valve opening correction value and underflow discharge frequency. When the sand particle purity is low or poor, the command prioritizes increasing the hydrocyclone inlet pressure to enhance the centrifugal separation effect. When the sand-sludge separation interface deviates from the appropriate position, the overflow valve opening is adjusted synchronously to optimize the flow field distribution. The underflow discharge frequency is dynamically adapted according to the real-time sludge adhesion rate.

9. A high-efficiency sedimentation tank control system for sand media based on a visual recognition neural network according to claim 1, characterized in that, The execution adjustment module incorporates a multi-layered protection mechanism, including five layers of protection logic: parameter over-limit protection, actuator fault protection, adjustment rate limitation, emergency fault-tolerant switching, and status feedback verification. Parameter over-limit protection presets an upper limit of 0.8 MPa for the hydrocyclone inlet pressure and a ±200 mm limit for the overflow pipe height adjustment. When the adjustment command triggers parameters approaching the threshold, it automatically cuts off the adjustment range and triggers an early warning to prevent equipment overload. Actuator fault protection monitors the electric actuator's status in real time using current and position sensors. When jamming or overload signals are detected, it immediately pauses adjustment and switches to manual emergency mode, while simultaneously uploading a fault alarm. The adjustment rate limitation sets an upper limit of ≤0.02 MPa / s for the inlet pressure change rate and ≤5 mm / s for the overflow pipe height change rate. The emergency fault-tolerant switching configuration includes a primary and backup actuator; when the primary actuator fails, it automatically switches to the backup actuator to maintain adjustment capability. The status feedback verification verifies the execution effect after adjustment using visual and pressure sensors; if the deviation exceeds 5%, it automatically triggers a secondary correction.

Images