Intelligent control system for mixed bed resin separation tower of nuclear power plant

Abstract

The present application belongs to the field of nuclear power plant chemical water treatment and industrial automation control technology, and particularly relates to an intelligent control system for a mixed bed resin separation tower in a nuclear power plant. A multi-spectral panoramic vision acquisition module: non-contact real-time monitoring of resin layering state inside the resin separation tower; a fluid mechanics pulse control module: generating pulsating water flow with specific waveforms by adjusting the operating frequency of the backwash flow regulating valve and backwash water pump during the backwash layering stage; a data transmission and preprocessing module: time stamp alignment and cleaning of the collected visual data and process parameters in the DCS system; a deep learning interface prediction module: processing the spatiotemporal sequence data of resin sedimentation using a hybrid architecture combining a U-Net semantic segmentation network and a long short-term memory network, combining historical regeneration data and real-time image data to predict the final stable position of the anion-cation resin interface and the interface drop rate during the conveying process. The present application improves the safety of the secondary circuit water quality in the nuclear power plant and the intelligent level of system operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of nuclear power plant chemical water treatment and industrial automation control technology, specifically relating to an intelligent control system for a mixed bed resin separation tower in a nuclear power plant. Background Technology

[0002] As energy facilities with high safety and reliability requirements, the purity of the secondary loop water in nuclear power plants directly affects the corrosion resistance and service life of core equipment such as steam generators and turbines. The condensate polishing system in a nuclear power plant plays a crucial role, like the kidneys, removing corrosion products and ionic impurities from the condensate to ensure the feedwater meets extremely stringent chemical standards. The external regeneration process of the mixed-bed resin is a core component ensuring the system's continuous operation, while the resin separation tower located within the regeneration system is a key device determining the regeneration effect. Its main task is to precisely separate the anion, cation, and inert resin layers using hydraulic backwashing after resin deterioration, so that they can be subsequently sent to the anion / cation regeneration towers for chemical regeneration. However, during the resin separation process, nuclear power plant condensate polishing systems often face challenges such as blurred resin interfaces, difficulty in depolymerizing aged resin agglomerates, and difficulties in positioning intermediate layers. These changes in physical properties can lead to incomplete separation of anion and cation resins (cross-contamination), resulting in excessive sulfate or sodium ion levels in the regenerated effluent, seriously threatening the water chemical safety of the nuclear power plant's secondary loop.

[0003] In the past, the resin separation process in condensate polishing systems of nuclear power plants typically relied on manual visual observation and fixed-sequence control to address resin stratification and transport issues. This method depended on experienced operators using multiple sight glasses mounted on the sidewalls of the separation tower to visually determine the location of the stratification interface by observing the color differences between the anion and cation resins. Backwash water pumps and grease discharge valves were then controlled manually or semi-automatically. However, manual monitoring and operation suffer from significant subjective errors, response delays, and difficulties in nighttime observation. Especially after prolonged resin operation, when the cation resin darkens, the anion resin changes color, or both clump together due to electrostatic adsorption, traditional visual identification becomes extremely difficult to distinguish clearly, easily leading to misoperation. This can result in the mixing of cation resin into the anion resin or vice versa, reducing regeneration efficiency and increasing acid and alkali consumption and wastewater discharge costs.

[0004] Furthermore, nuclear power plants have extremely high requirements for the automation and fault tolerance of their regeneration systems. Any failure in the regeneration process can lead to insufficient spare beds, affecting the unit's full-power operation at full flow. Because current technology lacks the ability to monitor and predict the microscopic state of resin particles and the sedimentation trend of macroscopic interfaces in real time, poor stratification often necessitates repeated backwashing, sedimentation, and air rinsing. This not only prolongs the regeneration cycle but may also accelerate resin breakage due to repeated mechanical friction. While existing single-point photoelectric switches or densitometers can provide local signals, they often issue erroneous signals when faced with complex mixed resin layers or window fouling, failing to fully meet the high-precision separation and high-reliability requirements of nuclear power plants for resin regeneration.

[0005] To address these issues, researchers have successively attempted to introduce technologies such as turbidimeter monitoring, ultrasonic interface detection, and time-series-based automatic control logic to assist in the management of the resin separation process. However, relying solely on physical instrument monitoring is easily affected by water turbidity and cannot identify the chromatographic characteristics of the resin; while pure time-based control logic simplifies operation, its accuracy and adaptability remain insufficient when dealing with nonlinear changes in sedimentation rate caused by resin performance degradation.

[0006] For the reasons mentioned above, a comprehensive solution integrating machine vision perception and intelligent decision support is needed. This solution should be able to achieve panoramic real-time monitoring and feature extraction of the resin interface within the separation tower through a multispectral vision array, accurately predict the interface behavior of aging resins through advanced deep learning models, and combine this with an adaptive pulse hydraulic control mechanism to achieve efficient, precise, and intelligent closed-loop control of the resin separation process. By deploying this optimization method in the condensate polishing system of nuclear power plants, the problem of resin cross-contamination can be effectively solved, significantly improving the automation level of the regeneration system and the guarantee capability of secondary loop water quality. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent control system for a mixed-bed resin separation tower in a nuclear power plant. By integrating machine vision panoramic monitoring, multispectral imaging analysis, deep learning interface prediction, pulsed microbubble deagglomeration, and millisecond-level valve interlock control technologies, it can effectively address the interface identification challenges caused by resin aging and discoloration, electrostatic adsorption and agglomeration, and nonlinear changes in sedimentation rate. This enables precise closed-loop control of the resin stratification and transport process, ensuring the efficient and automated operation of the resin regeneration system, significantly reducing the resin cross-contamination rate, and improving the safety of the secondary loop water quality and the intelligent operation and maintenance level of the system.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A smart control system for a mixed-bed resin separation tower in a nuclear power plant includes a multispectral panoramic vision acquisition module for non-contact real-time monitoring of the resin stratification state inside the resin separation tower; and a fluid dynamics pulse control module for generating pulsating water flow with a specific waveform by adjusting the backwash flow regulating valve and the operating frequency of the backwash water pump during the backwash stratification stage. Data transmission and preprocessing module: performs timestamp alignment and cleaning on the collected visual data and process parameters in the DCS system; Deep learning interface prediction module: uses a hybrid architecture combining U-Net semantic segmentation network and long short-term memory network to process the spatiotemporal sequence data of resin sedimentation, and combines historical regeneration data and real-time image data to predict the final stable position of the cation and anion resin interface and the interface descent rate during the transportation process; Real-time decision support module: automatically assesses the current resin separation risk level based on the output of the deep learning prediction model and calculates the optimal transportation strategy; Manual confirmation mechanism module: sends a confirmation request to the main control room operator when preparing to open key valves or determining the end of stratification; Adaptive backwash stratification module: dynamically controls the backwash water pump in a closed loop according to the real-time expansion rate of the resin; Intelligent transportation and interface locking module: achieves zero cross-contamination control during the transportation stage; Feedback and optimization mechanism module: continuously optimizes the control strategy by performing correlation analysis on the effluent quality after each regeneration and the resin separation process data, using reinforcement learning algorithms.

[0009] Multispectral panoramic visual acquisition module: acquires chromatographic characteristics, particle packing density, and dynamic interface changes of cation exchange resin, anion exchange resin, inert resin, and mixed resin layers inside the resin separation tower.

[0010] Multispectral panoramic vision acquisition module: By deploying a high-resolution industrial camera array at 9 standard viewing mirrors on the side wall of the separation tower, covering the visible and near-infrared bands, it supports real-time acquisition of chromatographic characteristics, particle packing density and interface dynamic changes of cation resin, anion resin, inert resin and mixed lipid layer in the tower at a frame rate of 30fps or higher.

[0011] Multispectral panoramic vision acquisition module: Equipped with a ring polarization light source and adaptive exposure control algorithm, it ensures high-quality image data acquisition even at night or under conditions of slight window contamination; it is equipped with a backup imaging unit for critical windows, and if a node fails or a window is severely contaminated, it automatically switches to the interpolation algorithm mode of the adjacent window and records the fault information in the log; it uses adaptive exposure and bandpass filtering algorithms to eliminate interference from changes in ambient light or reflective interference, preserving the key spectral characteristics of the resin layer.

[0012] Fluid dynamics pulse control module: With a built-in fluid dynamics model, it can automatically calculate the required shear force based on the degree of resin agglomeration. Through flow fluctuations in the form of sine waves or square waves, it breaks the van der Waals forces and electrostatic adsorption between aged resins, achieving complete depolymerization and stratification of the resin.

[0013] Data transmission and preprocessing module: Process parameters include backwash flow rate, tower temperature, and pressure; Gaussian filtering and histogram equalization algorithms are used to enhance image contrast and eliminate bubble occlusion interference, and edge detection algorithms are used to initially extract the resin layer contours. A panoramic thermal map of the tower is constructed through image stitching technology.

[0014] The deep learning interface prediction module, U-Net, is used for pixel-level classification of single-frame images, identifying the distribution of light-colored regions in anion resin, dark-colored regions in cation resin, and intermediate-colored regions in inert resin. U-Net is also used to extract the spatial distribution features of the resin; its basic convolution operation formula is as follows: These are the pixel values ​​of the output feature map, representing the probability of resin category. I It is the input image. K It is a convolution kernel.

[0015] Deep learning interface prediction module: Long Short-Term Memory network is used to capture the dynamic trend of interface changes over time. The interface position prediction model is represented as: The predicted interface height for the next moment. The height of the current screen. To transmit flow, The resin aging characteristic coefficient is color / particle size. These are the model weight parameters.

[0016] Real-time decision support module: Risk level R is calculated using the following formula: For interface clarity probability, This is the normalized value of the distance from the current interface to the target outlet. To reduce the interface lag, , , These are the weighting coefficients.

[0017] Real-time decision support module: When R When the speed is below the set threshold, it is considered a safe operating condition, and full-speed conveying is maintained; when R When the load is raised, the deceleration logic is automatically triggered or the conveying is paused, waiting for the interface to reset.

[0018] Manual confirmation mechanism module: The operation interface displays the enhanced resin interface image and AI confidence score in real time. After the operator confirms via physical button or touch screen, the next operation will be executed. If no confirmation is made within the specified time, the current safe state will be maintained or an alarm will be triggered.

[0019] Adaptive backwashing stratification module: The expansion rate E is calculated using the following formula: The dynamic resin layer height for visual recognition. The static resin layer height is determined by adjusting the backwash flow rate using a PID algorithm to precisely maintain the expansion rate within the specified range of 50%-80%, and automatically executes a linear deceleration strategy at the end of the layering process to prevent resin disturbance caused by abrupt pump stoppage.

[0020] Intelligent conveying and interface locking module: During the process of conveying anion resin to the anion regeneration tower, the position of the anion-inert interface is tracked in real time. When the interface drops to a preset safe distance from the middle outlet, a millisecond-level interlock is automatically triggered, the conveying valve is immediately closed, and the backwash water replenishment valve is opened at the same time to compensate for the liquid level, ensuring that the cation resin never enters the anion regeneration tower.

[0021] Feedback and optimization mechanism module: Effluent water quality includes sulfate and sodium ion data.

[0022] Feedback and optimization mechanism module: If the effluent indicators are abnormal after a regeneration, the system automatically reviews the video recordings and control parameters of that separation, adjusts the interface judgment threshold or backwash intensity for the next regeneration, and optimizes the model parameters. i Online updates: or For learning rate, This is a loss function based on water quality deviation.

[0023] The beneficial effects achieved by this invention are as follows: This invention solves the problems of large errors, high resin cross-contamination rates, low regeneration efficiency, and lack of real-time feedback regulation in traditional resin separation processes by integrating multi-window online visual data, convolutional neural network image segmentation algorithms, pulsed backwashing regulation mechanisms, and millisecond-level valve interlocking control. It relies heavily on manual experience and visual judgment, leading to these issues. Applicable to pressurized water reactor nuclear power plants in scenarios requiring high water quality standards for condensate polishing resin regeneration, spent resin stratification, and transport control, this invention significantly improves the intelligent operation and maintenance capabilities of chemical water treatment systems and the safety of secondary loop water quality.

[0024] This invention integrates a multispectral panoramic visual acquisition module with image enhancement technology to achieve comprehensive real-time monitoring of the microstructure, aging chromatographic characteristics, and stratification interfaces of resin within the separation tower. This effectively overcomes the subjective errors and blind spots inherent in traditional manual observation through a sieves. A deep learning interface prediction model improves the accuracy and response speed in predicting blurred interfaces and resin settling trends. When the vision system detects abnormalities such as resin agglomeration or unclear interfaces, it rapidly analyzes and automatically adjusts the backwash flow regulating valve and pulse control module, utilizing adaptive pulsating water flow to break up agglomerates and ensure optimal stratification, significantly reducing the risk of regeneration failure due to human error.

[0025] This invention reduces manual intervention, lowers operating costs and acid / alkali consumption, and achieves zero cross-contamination during resin transport by precisely controlling the anion resin discharge valve through millisecond-level interlocking and automatically performing liquid level compensation, thus maintaining the efficient operation of the subsequent regeneration tower. The feedback and optimization mechanism continuously improves the system's intelligence level through self-learning, and optimizes the interface identification threshold and control parameters based on the effluent quality data after regeneration, ensuring that nuclear power plants can maintain high-standard fine treatment operation stability even under complex environments such as resin aging or fluctuating operating conditions. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of an intelligent control system for a mixed-bed resin separation tower in a nuclear power plant. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0028] like Figure 1 As shown, an intelligent control system for a mixed-bed resin separation tower in a nuclear power plant is a comprehensive solution integrating multi-level sensing and collaboration, full-process closed-loop control, and adaptive decision-making adjustment. It aims to achieve efficient separation and precise delivery of failed resin in the condensate polishing system of a nuclear power plant by deeply integrating multiple aspects such as multispectral visual acquisition of the resin's microstructure and macroscopic interface within the separation tower, image data processing, deep learning interface prediction, decision support, pulsed hydraulic depolymerization, and precision valve interlocking control. It is particularly suitable for regeneration control under conditions of resin aging, severe agglomeration, and interface ambiguity. The system includes: a multispectral panoramic visual acquisition module, a fluid dynamics pulse control module, a data transmission and preprocessing module, a deep learning interface prediction module, a real-time decision support module, a manual confirmation mechanism module, an adaptive backwashing and stratification module, an intelligent delivery and interface locking module, and a feedback and optimization mechanism module.

[0029] The multispectral panoramic vision acquisition module serves as the system's data sensing entry point, responsible for non-contact real-time monitoring of the resin stratification state inside the resin separation tower. It uniformly acquires data on the chromatographic characteristics, particle packing density, and dynamic interface changes of the cation exchange resin, anion exchange resin, inert resin, and mixed resin layers within the tower. Specifically, the system deploys a high-resolution industrial camera array at nine standard viewing mirrors on the sidewall of the separation tower, covering the visible and near-infrared bands. This array supports real-time acquisition of the chromatographic characteristics, particle packing density, and dynamic interface changes of the cation exchange resin, anion exchange resin, inert resin, and mixed resin layers at 30fps or higher frame rates. To overcome uneven lighting and window reflection interference, the acquisition module is equipped with a ring polarization light source and an adaptive exposure control algorithm, ensuring high-quality image data acquisition even at night or under conditions of slight window contamination. In the data transmission stage, the system transmits multiple high-definition video streams to the edge computing unit in real time via a fiber optic ring network, and employs hardware encryption protocols to ensure the data security of the nuclear power plant's industrial control network. To improve monitoring accuracy, when abnormal operating conditions are detected (such as severe resin clumping or blurred interfaces), the system will automatically increase the sampling frequency and trigger subsequent analysis modules for emergency processing.

[0030] The system incorporates a redundancy mechanism, configuring backup imaging units for critical windows. If a node fails or a window becomes severely contaminated, the system automatically switches to the interpolation algorithm mode of the adjacent window and logs the fault information. For changes in ambient lighting or reflective interference, adaptive exposure and bandpass filtering algorithms are employed to eliminate interference while preserving key spectral characteristics of the resin layers. Data synchronization utilizes high-precision protocols such as IEEE 1588PTP to ensure unified time alignment between multiple visual data streams and DCS process parameters (such as backwash flow), supporting high-quality training and prediction of deep learning models.

[0031] The fluid dynamics pulse control module is the system's execution and regulation unit. During the backwash stratification stage, it generates a specific waveform of pulsating water flow by adjusting the backwash flow regulating valve and the operating frequency of the backwash water pump. This module incorporates a fluid dynamics model that automatically calculates the required shear force based on the resin's agglomeration degree (feedback from the vision module). Through sinusoidal or square wave flow fluctuations, it breaks down the van der Waals forces and electrostatic adsorption between aged resins, replacing the traditional air scrubbing step that relies on manual experience, thus achieving complete depolymerization and stratification of the resin.

[0032] The data transmission and preprocessing module is responsible for timestamping and cleaning the acquired visual data and process parameters (such as backwash flow rate, tower temperature, and pressure) in the DCS system. The system preferably uses the IEEE 1588 PTP protocol for nanosecond-level time synchronization to ensure that the interface position captured by the vision strictly corresponds to the current valve opening and pump speed. During preprocessing, Gaussian filtering and histogram equalization algorithms are used to enhance image contrast, eliminate bubble occlusion interference, and edge detection algorithms are used to initially extract the resin layer contours, providing standardized tensor data for subsequent deep learning models. A panoramic heat map of the tower is constructed using image stitching technology to ensure the data quality for subsequent analysis.

[0033] The deep learning interface prediction module is the core of the system's intelligent analysis. It employs a hybrid architecture combining the U-Net semantic segmentation network and the Long Short-Term Memory (LSTM) network to process the spatiotemporal sequence data of resin sedimentation. By combining historical regeneration data (including sedimentation curves of resins at different aging levels) and real-time image data, it predicts the final stable position of the cation and cation resin interface and the interface descent rate during the transport process.

[0034] U-Net is used for pixel-level classification of single-frame images, identifying the distribution of anion resin (light color), cation resin (dark color), and inert resin (intermediate color); LSTM is used to capture the dynamic trend of interface changes over time. The interface location prediction model can be represented as: (1) in, The predicted interface height for the next moment. The height of the current screen. To transmit flow, This represents the resin aging characteristic coefficient (color / particle size). These are the model weight parameters. Using this model, the system can still accurately calculate the true physical boundary line even when the interface is blurry or exhibits trailing artifacts.

[0035] U-Net is used to extract the spatial distribution features of resin. Its basic convolution operation formula is as follows: (2) in, These are the pixel values ​​of the output feature map (representing the probability of resin category). I It is the input image. K It is a convolutional kernel. By combining LSTM and CNN, the system can simultaneously capture temporal settlement dynamics and spatial hierarchical features, improving the accuracy and reliability of interface prediction.

[0036] The real-time decision support module is the core control unit of the system, integrating a fuzzy neural network and a rule engine. Based on the output of the deep learning prediction model, it automatically assesses the current resin separation risk level and calculates the optimal delivery strategy. The overall risk level R can be calculated using the following formula: (3) in: The probability of interface clarity is (0-1). This is the normalized value of the distance from the current interface to the target outlet. To reduce the interface lag, , , These are the weighting coefficients.

[0037] when R When the speed is below the set threshold, the system determines it to be a safe operating condition and maintains full-speed conveying; when R When the load is raised, the system automatically triggers deceleration logic or pauses the conveying process, waiting for the interface to reset.

[0038] The manual confirmation mechanism module is a crucial safety feature of the system, adhering to the "human-machine loop" operating principle of nuclear power plants. When the system is about to open a critical valve (such as the anion exchange resin transfer valve 40LDP21AA106 / 107) or when the stratification determination is complete, a confirmation request is sent to the operator in the main control room. The operating interface displays an enhanced image of the resin interface and an AI confidence score in real time. The system can only proceed to the next step after the operator confirms via a physical button or touchscreen. If no confirmation is received within the specified time, the system will maintain its current safe state or trigger an alarm.

[0039] The adaptive backwash stratification module is responsible for dynamically controlling the backwash water pump in a closed loop based on the real-time expansion rate of the resin. The expansion rate E is calculated using the following formula: (4) in, The dynamic resin layer height for visual recognition. This represents the static resin layer height. The system uses a PID algorithm to adjust the backwash flow rate, precisely maintaining the expansion rate within the specified range of 50%-80%. At the end of the separation process, it automatically implements a linear ramp-down strategy to prevent resin disturbance caused by abrupt pump shutdown.

[0040] The intelligent conveying and interface locking module is responsible for achieving "zero cross-contamination" control during the conveying stage. During the conveying of anion resin to the anion regeneration tower, the system tracks the position of the anion-inert interface in real time. When the interface drops to a preset safe distance (e.g., 50mm) from the intermediate discharge port (e.g., at the LDP21AX004 elevation), the system automatically triggers a millisecond-level interlock, immediately closing the conveying valve and simultaneously opening the backwash water supply valve for level compensation, ensuring that cation resin never enters the anion regeneration tower.

[0041] The feedback and optimization mechanism module analyzes the correlation between the effluent quality (sulfate and sodium ion data) after each regeneration and the resin separation process data, and continuously optimizes the control strategy using reinforcement learning algorithms. If the effluent indicators are abnormal after a regeneration, the system automatically reviews the video recordings and control parameters of that separation, adjusts the interface judgment threshold or backwash intensity for the next regeneration, and achieves online updates of the model parameter θ. (5) in, or For learning rate, This is a loss function based on water quality deviation.

[0042] Through the collaborative work of the above-mentioned multiple modules, this invention can upgrade the traditional open-loop manual separation process to a closed-loop intelligent control process without changing the original main equipment structure, significantly reducing the risk of manual operation, extending the service life of the resin, and ensuring the long-term stability of the secondary loop water quality in nuclear power plants.

Claims

1. An intelligent control system for a mixed-bed resin separation tower in a nuclear power plant, characterized in that: It includes a multispectral panoramic vision acquisition module for non-contact real-time monitoring of the resin stratification state inside the resin separation tower; and a fluid dynamics pulse control module to generate a specific waveform of pulsating water flow by adjusting the backwash flow regulating valve and the operating frequency of the backwash water pump during the backwash stratification stage. Data transmission and preprocessing module: performs timestamp alignment and cleaning on the acquired visual data and process parameters in the DCS system; Deep learning interface prediction module: It uses a hybrid architecture combining U-Net semantic segmentation network and long short-term memory network to process the spatiotemporal sequence data of resin sedimentation, and combines historical regeneration data and real-time image data to predict the final stable position of the cation and cation resin interface and the interface descent rate during the transportation process; Real-time decision support module: Based on the output of the deep learning prediction model, it automatically assesses the current resin separation risk level and calculates the optimal transportation strategy. Manual confirmation mechanism module: When preparing to open a key valve or when determining the end of the stratification, a confirmation request is sent to the main control room operator; Adaptive backwash stratification module: dynamically controls the backwash water pump in a closed loop based on the real-time expansion rate of the resin; Intelligent delivery and interface locking module: achieves zero cross-contamination control during the delivery stage; Feedback and optimization mechanism module: continuously optimizes the control strategy by correlating the effluent quality after each regeneration with the resin separation process data and using reinforcement learning algorithms.

2. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Multispectral panoramic visual acquisition module: acquires chromatographic characteristics, particle packing density, and dynamic interface changes of cation exchange resin, anion exchange resin, inert resin, and mixed resin layers inside the resin separation tower.

3. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 2, characterized in that: Multispectral panoramic vision acquisition module: By deploying a high-resolution industrial camera array at 9 standard viewing mirrors on the side wall of the separation tower, covering the visible and near-infrared bands, it supports real-time acquisition of chromatographic characteristics, particle packing density and interface dynamic changes of cation resin, anion resin, inert resin and mixed lipid layer in the tower at a frame rate of 30fps or higher.

4. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 3, characterized in that: Multispectral panoramic vision acquisition module: Equipped with a ring polarization light source and adaptive exposure control algorithm, it ensures high-quality image data acquisition even at night or under conditions of slight window contamination; it is equipped with a backup imaging unit for critical windows, and if a node fails or a window is severely contaminated, it automatically switches to the interpolation algorithm mode of the adjacent window and records the fault information in the log; it uses adaptive exposure and bandpass filtering algorithms to eliminate interference from changes in ambient light or reflective interference, preserving the key spectral characteristics of the resin layer.

5. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Fluid dynamics pulse control module: With a built-in fluid dynamics model, it can automatically calculate the required shear force based on the degree of resin agglomeration. Through flow fluctuations in the form of sine waves or square waves, it breaks the van der Waals forces and electrostatic adsorption between aged resins, achieving complete depolymerization and stratification of the resin.

6. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Data transmission and preprocessing module: Process parameters include backwash flow rate, tower temperature, and pressure; Gaussian filtering and histogram equalization algorithms are used to enhance image contrast and eliminate bubble occlusion interference, and edge detection algorithms are used to initially extract the resin layer contours. A panoramic thermal map of the tower is constructed through image stitching technology.

7. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: The deep learning interface prediction module, U-Net, is used for pixel-level classification of single-frame images, identifying the distribution of light-colored regions in anion resin, dark-colored regions in cation resin, and intermediate-colored regions in inert resin. U-Net is also used to extract the spatial distribution features of the resin; its basic convolution operation formula is as follows: These are the pixel values ​​of the output feature map, representing the probability of resin category. I It is the input image. K It is a convolution kernel.

8. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Deep learning interface prediction module: Long Short-Term Memory network is used to capture the dynamic trend of interface changes over time. The interface position prediction model is represented as: The predicted interface height for the next moment. The height of the current screen. To transmit flow, The resin aging characteristic coefficient is color / particle size. These are the model weight parameters.

9. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Real-time decision support module: Risk level R is calculated using the following formula: For interface clarity probability, This is the normalized value of the distance from the current interface to the target outlet. To reduce the interface lag, , , These are the weighting coefficients.

10. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 9, characterized in that: Real-time decision support module: When R When the speed is below the set threshold, it is considered a safe operating condition, and full-speed conveying is maintained; when R When the load is raised, the deceleration logic is automatically triggered or the conveying is paused, waiting for the interface to reset.

11. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Manual confirmation mechanism module: The operation interface displays the enhanced resin interface image and AI confidence score in real time. After the operator confirms via physical button or touch screen, the next operation will be executed. If no confirmation is made within the specified time, the current safe state will be maintained or an alarm will be triggered.

12. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Adaptive backwashing stratification module: The expansion rate E is calculated using the following formula: The dynamic resin layer height for visual recognition. The static resin layer height is determined by adjusting the backwash flow rate using a PID algorithm to precisely maintain the expansion rate within the specified range of 50%-80%, and automatically executes a linear deceleration strategy at the end of the layering process to prevent resin disturbance caused by abrupt pump stoppage.

13. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Intelligent conveying and interface locking module: During the process of conveying anion resin to the anion regeneration tower, the position of the anion-inert interface is tracked in real time. When the interface drops to a preset safe distance from the middle outlet, a millisecond-level interlock is automatically triggered, the conveying valve is immediately closed, and the backwash water replenishment valve is opened at the same time to compensate for the liquid level, ensuring that the cation resin never enters the anion regeneration tower.

14. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Feedback and optimization mechanism module: Effluent water quality includes sulfate and sodium ion data.

15. The intelligent control system for a mixed-bed resin separation tower in a nuclear power plant according to claim 1, characterized in that: Feedback and optimization mechanism module: If the effluent indicators are abnormal after a regeneration, the system automatically reviews the video recordings and control parameters of that separation, adjusts the interface judgment threshold or backwash intensity for the next regeneration, and optimizes the model parameters. θ Online updates: η For learning rate, This is a loss function based on water quality deviation.

Images