An electric field analysis device and method for rare earth metal electrolysis furnaces

Abstract

This invention relates to the field of metallurgical electrochemistry and process control technology, specifically disclosing an electric field analysis device and method for rare earth metal electrolysis furnaces. The device includes: a real-time electrode morphology sensing module for non-contact acquisition of three-dimensional morphology data of the anode and cathode; a multi-source data fusion and electric field reconstruction module for dynamically reconstructing the three-dimensional electric field distribution within the furnace based on real-time data; and a process parameter dynamic control module for analyzing electric field uniformity and generating control commands. Through the above scheme, this invention achieves online real-time sensing, reconstruction, and closed-loop active control of the electric field within the electrolysis furnace, providing an effective means to improve process control accuracy and production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of metallurgical electrochemistry and process control technology, specifically relating to an electric field analysis device and method for rare earth metal electrolysis furnaces. Background Technology

[0002] Rare earth metal electrolysis is a key technological step in the field of rare earth pyrometallurgy. Its core lies in driving the directional migration of rare earth ions in the molten salt electrolyte and their deposition at the cathode using a direct current electric field. The efficiency, energy consumption, and product purity of this process are directly affected by the characteristics of the electric field distribution inside the electrolysis furnace. The uniformity and stability of the electric field distribution are crucial physical factors determining the success or failure of the electrolysis process. Electric field analysis and control technology for rare earth metal electrolysis furnaces aims to provide a theoretical basis and operational guidance for process optimization by accurately understanding the electric field distribution within the furnace. This is an important research direction for improving the technological level of the rare earth electrolysis industry.

[0003] Existing technologies mainly rely on numerical simulation methods based on finite element analysis or multiphysics coupling to perform offline simulation analysis of the electric field of electrolytic furnaces. These methods can calculate the theoretical electric field distribution cloud map based on a preset geometric model and physical property parameters, providing a reference for furnace design and preliminary selection of process parameters.

[0004] Existing technical solutions have limitations: simulation analysis relies heavily on static and idealized model assumptions, and cannot reflect in real time the dynamic changes in the distance and geometry between the anode and cathode caused by the continuous addition of raw materials, the continuous consumption of anode graphite, and the growth of cathode deposits during the production process; at the same time, the power control system can usually only macroscopically adjust the total input current or total voltage, and lacks the ability to directly perceive and actively control the microscopic and key process parameter of the electric field distribution in the furnace space.

[0005] Because the actual changes in the electric field distribution are passive and uncontrolled, and its evolution is mainly influenced by the randomness of the aforementioned physical dimensional changes, there is an inevitable gap between the theoretical electric field analysis derived from offline simulation and the actual production conditions. This disconnect between "analysis" and "reality" makes process optimization lack precise data support, hindering the realization of closed-loop control and proactive adjustment of production based on real-time electric field status, and restricting further improvements in electrolysis efficiency and reductions in energy consumption. Therefore, how to achieve real-time and accurate analysis and perception of the electric field distribution within rare earth metal electrolysis furnaces, and establish its correlation with process control, has become a pressing technical challenge in this field. Summary of the Invention

[0006] The purpose of this invention is to provide an electric field analysis device and method for rare earth metal electrolysis furnaces, so as to solve the technical contradiction in the prior art that relies on offline simulation and cannot perceive the evolution of the electric field distribution in the furnace caused by the dynamic changes of the electrodes in real time, which leads to a lack of accurate data support for process control and difficulty in achieving closed-loop control.

[0007] To achieve the above objectives, the present invention provides an electric field analysis device for a rare earth metal electrolysis furnace, comprising:

[0008] The electrode morphology real-time sensing module is used to acquire the three-dimensional geometric morphology data of the anode and cathode in the electrolytic furnace in real time without contact. The electrode morphology real-time sensing module includes at least two sets of high-resolution laser scanning units and one set of industrial vision imaging units. The electrode morphology real-time sensing module has a built-in edge computing unit for preprocessing the acquired raw point cloud data and image data. The preprocessing process includes point cloud noise reduction, coordinate system registration, and multi-frame data fusion based on feature point matching, and outputs a structured data package containing timestamps that characterizes the precise three-dimensional contours and relative spatial position relationships of the anode and cathode at the current moment.

[0009] Point cloud denoising employs a statistical outlier removal algorithm based on the 10 nearest neighbors, identifying points whose distance is greater than three standard deviations of the average distance to neighboring points as noise points. Coordinate system registration is solved using the Levenberg-Marquardt iterative algorithm to address a nonlinear least-squares optimization problem, with the objective function being... ;

[0010] These are the coordinates of the feature points in the laser point cloud / visual image coordinate system. The coordinates of the reference feature point in the world coordinate system. For rotation matrix, It is a translation vector;

[0011] The iteration termination condition is that the error change between two consecutive iterations is less than 1 × 10⁻⁶. -6 Millimeters; Multi-frame data fusion is based on the iterative nearest point principle, combined with the random sampling consensus algorithm to select matching feature point pairs, with the inlier threshold set to 0.1 millimeters, and finally outputs a structured data packet with a point cloud density ≥ 500 points / square centimeter;

[0012] The multi-source data fusion and electric field reconstruction module is used to dynamically reconstruct the three-dimensional electric field distribution within the electrolytic furnace based on the structured data package.

[0013] The process parameter dynamic control module is used to analyze the reconstructed three-dimensional electric field distribution and generate process control instructions. The process parameter dynamic control module includes an electric field uniformity evaluation submodule and a control strategy generation submodule.

[0014] Preferably, the laser scanning unit emits a scanning laser beam at a fixed frequency to acquire three-dimensional point cloud data of the anode working end face and the cathode deposit surface immersed below the molten salt surface;

[0015] The industrial vision imaging unit simultaneously acquires visible light and infrared thermal imaging video streams of the observation window area of ​​the electrolytic furnace.

[0016] Preferably, the multi-source data fusion and electric field reconstruction module first constructs an initial three-dimensional computational domain digital twin model that strictly corresponds to the geometric dimensions of the physical electrolysis furnace; the mesh division of the digital twin model adopts an adaptive unstructured tetrahedral mesh, and the mesh is refined in the preset anode and cathode regions;

[0017] The core of the multi-source data fusion and electric field reconstruction module is an embedded real-time solver.

[0018] The embedded real-time solver receives structured data packets from the real-time electrode morphology sensing module and simultaneously accesses real-time voltage and current data provided by the electrolysis power supply system.

[0019] The embedded real-time solver uses the digital twin model as the computational framework, takes the real-time acquired anode surface profile, cathode deposit surface profile and anode-cathode distance as dynamically updated boundary conditions, takes the real-time voltage as the excitation source loading condition, and uses the potential field control equation based on the finite volume method to perform real-time numerical solution based on the preset molten salt electrolyte conductivity parameters.

[0020] The boundary conditions are dynamically updated using radial basis function (RBF) interpolation to map the point cloud on the electrode surface to the boundary mesh nodes of the digital twin model. The anode-cathode distance is calculated by averaging the distances between the nearest point pairs in the point cloud. The governing equation for the potential field is the Laplace equation. Discretized using the finite volume method ;

[0021] To control the surface area, The distance between adjacent nodes. The potential of adjacent nodes. This represents the potential of the current control center node. To control the conductivity of the molten salt on the control surface, the solution is first calculated using 1×10⁻⁶. -4 The local electric field perturbation is updated using an explicit scheme with a time step of seconds, and then the entire field is solved using an implicit scheme based on the conjugate gradient method. The L2 norm is used to determine the convergence of the residuals, with a threshold of 1×10⁻⁶. -5 ;

[0022] The governing equations are iteratively calculated within each computation time step until the residual converges to a preset threshold of 1×10⁻⁶. -5 The following outputs the potential value and electric field intensity vector at each computational grid node inside the electrolytic furnace under the current operating conditions.

[0023] Preferably, the electric field uniformity assessment submodule performs spatial statistical analysis on the three-dimensional electric field data output by the multi-source data fusion and electric field reconstruction module, calculates the standard deviation of electric field intensity in key process areas, and the ratio of the maximum to the minimum value as uniformity evaluation indicators, and identifies the spatial coordinates of weak field areas with electric field intensity less than 80% of the preset threshold and strong field areas with electric field intensity greater than 120% of the preset threshold.

[0024] The control strategy generation submodule makes decisions based on the electric field uniformity assessment results and preset process objectives, and the decision logic is based on a preset fuzzy rule library.

[0025] When the regulation strategy generation submodule identifies a persistent weak field in a specific cathode region, it generates an instruction and sends it to the multi-channel programmable DC power supply unit of the electrolysis power supply system through the industrial bus communication interface to fine-tune the current distribution ratio of the corresponding cathode branch. The current adjustment range is calculated proportionally and integrally based on the degree of weak field.

[0026] The formula is: ,in, For the changed current, the proportionality coefficient =0.05 (current adjustment coefficient / field strength deviation rate), integral coefficient =0.02 (current adjustment coefficient / (field strength deviation rate·second), integration time window = 3 seconds, limited output) ; This represents the deviation of the average field strength within the cluster.

[0027] The regulation strategy generation submodule generates micro-motion control commands for the anode lifting mechanism in strong field areas caused by uneven anode consumption, so as to locally adjust the anode immersion depth.

[0028] Preferably, the coordinate system registration performed by the edge computing unit is achieved by solving a least-squares optimization problem that includes rotation matrices and translation vectors, thereby aligning the laser point cloud coordinate system with the visual image coordinate system to the world coordinate system with the center of the electrolytic furnace as the origin.

[0029] Preferably, the digital twin model constructed in the multi-source data fusion and electric field reconstruction module has initial geometric parameters derived from the design drawings of the electrolytic furnace, and is corrected through an offline high-precision calibration.

[0030] The offline high-precision calibration process is as follows: under the empty and cold state of the electrolytic furnace, a high-precision three-dimensional measuring instrument is used to obtain the precise spatial coordinates of the inner wall of the furnace, the initial installation position of the anode, and the cathode guide rod. These data are then used as reference values ​​and input into the digital twin model to complete the initial spatial alignment between the model and the physical object.

[0031] Preferably, the embedded real-time solver employs an iterative algorithm that combines explicit and implicit methods to balance computational speed and stability;

[0032] Within each real-time calculation cycle, an explicit scheme is used to quickly update the electric field disturbance caused by boundary changes, and an implicit scheme is used to solve for the stable distribution of the full field potential.

[0033] Preferably, the fuzzy rule base in the process parameter dynamic control module is obtained by machine learning training using historical optimal production data and electric field reconstruction data at the corresponding time.

[0034] The training process uses the gradient boosting decision tree algorithm, with electric field uniformity index and regional distribution characteristics as input features, and actual control actions that have been verified as effective as output labels. After training with a large number of samples, an interpretable set of "if-then" rules is generated.

[0035] Typical rules include:

[0036] If a weak field region with an area ≥ 0.01 square meters, a duration ≥ 3 seconds, and an electric field strength of 70% of the average value is detected in cathode A region, then the cathode A branch current is increased by 5%;

[0037] If the field strength deviation rate of the strong field area corresponding to the anode A1 region is ≥15%, the volume is ≥0.002 cubic meters, and the duration is ≥2 seconds, then the anode A1 lifting mechanism is controlled to descend by 0.3 mm.

[0038] Preferably, the device further includes a data storage and trend analysis module;

[0039] The data storage and trend analysis module is used to continuously store the full amount of time-series data generated by the real-time electrode morphology sensing module, the multi-source data fusion and electric field reconstruction module, and the dynamic control module of process parameters.

[0040] The data storage and trend analysis module has a built-in time-series database and data analysis engine, which can perform long-term trend analysis on the stored data. By comparing the changes in electric field uniformity index over different time periods, it can predict the anode consumption trend and the evolution of cathode deposition morphology, and generate maintenance early warning reports.

[0041] This invention also provides a method for analyzing the electric field of a rare earth metal electrolysis furnace. The method utilizes the aforementioned rare earth metal electrolysis furnace electric field analysis device to perform electric field analysis on the rare earth metal electrolysis furnace. The method includes the following steps:

[0042] Step 1: Using a laser scanning and visual imaging device deployed above the electrolysis furnace, three-dimensional point cloud data and image data of the anode working end face and the cathode deposit surface are acquired in real time at a frequency greater than 10 Hz.

[0043] Step 2: Use edge computing devices to perform fusion preprocessing on the raw data collected in Step 1. Through coordinate registration and data fusion, generate a real-time electrode morphology dataset containing accurate three-dimensional spatial coordinates.

[0044] Step 3: Input the real-time electrode morphology dataset obtained in Step 2, along with the voltage and current data read in real time from the electrolysis power supply system, into the preset digital twin model embedded real-time solver.

[0045] Step 4: In the real-time solver, the boundary conditions of the computational domain are dynamically updated with the real-time electrode morphology data, the excitation source conditions are updated with the real-time voltage data, and the finite volume method solver is called to iteratively solve the potential field control equation until convergence, and the full field potential distribution and electric field intensity vector distribution in the three-dimensional space inside the electrolytic furnace at the current moment are output.

[0046] Step 5: Perform spatial statistical analysis on the three-dimensional electric field distribution data output in Step 4, calculate the electric field uniformity evaluation index, and automatically identify abnormal areas of electric field intensity.

[0047] Step 6: Based on the analysis results of Step 5, call the preset fuzzy rule base to perform decision reasoning and generate fine-tuning instructions for the current ratio of each channel of the electrolytic power supply system or the position of the anode mechanism.

[0048] Step 7: Send the control command generated in Step 6 to the corresponding actuator through the industrial communication network, monitor the execution feedback, complete a complete cycle from electric field sensing, reconstruction analysis to closed-loop control, and return to Step 1 for continuous loop operation.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] 1. This invention achieves non-contact online measurement of electrode geometry with millimeter-level precision and millisecond-level response by integrating a real-time electrode morphology sensing module that combines laser scanning and machine vision. This fundamentally solves the core contradiction of traditional offline simulation relying on static models and failing to reflect the actual dynamic changes in production, providing a real and real-time data foundation for electric field analysis.

[0051] 2. This invention creatively integrates real-time sensed electrode morphology data with power supply parameters to drive an embedded real-time electric field reconstruction solver based on a digital twin model, realizing a technological paradigm shift from "offline calculation" to "online reconstruction" of the three-dimensional electric field distribution inside the electrolytic furnace. This device can continuously output dynamic electric field cloud maps that are strictly synchronized with production conditions, enabling process engineers to intuitively and accurately grasp the microscopic details and evolution trends of the electric field distribution, providing unprecedented high-precision data insights for process optimization.

[0052] 3. This invention establishes a closed-loop control link from "electric field analysis" to "process execution" through a dynamic process parameter control module. The device can automatically analyze the electric field uniformity and generate precise control commands based on an intelligent rule base, directly affecting the power supply current distribution and anode mechanical pose, thus achieving proactive intervention and optimization of the key process parameter of electric field distribution. This surpasses traditional control methods that can only perform macroscopic current and voltage regulation, fundamentally improving the uniformity and stability of the electric field distribution, and providing a direct and effective technical means to improve current efficiency, reduce unit product energy consumption, and increase the purity of metal products.

[0053] 4. The device architecture of this invention possesses excellent scalability and data value mining capabilities. The data storage and trend analysis module can accumulate electric field and operational data throughout the entire production process. Through long-term trend analysis, it can not only be used for predictive maintenance but also provide rich data assets for deeper research on process mechanisms, design of new electrolytic furnaces, and training of artificial intelligence optimization algorithms, thereby continuously promoting the technological advancement of rare earth electrolysis processes. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall technical solution architecture of the rare earth metal electrolysis furnace electric field analysis device proposed in this invention;

[0055] Figure 2 This is a schematic diagram of the core principle framework of dynamic electric field reconstruction based on digital twin and real-time sensing data in this invention;

[0056] Figure 3 This is a logical flowchart of the real-time sensing and data fusion of electrode morphology in this invention.

[0057] Figure 4 This is a real-time solution logic framework diagram of the multi-source data fusion and electric field reconstruction module in this invention;

[0058] Figure 5 This is a closed-loop control logic framework diagram of the present invention, which extends from electric field analysis to dynamic adjustment of process parameters. Detailed Implementation

[0059] Example 1: This example describes in detail the specific implementation scheme of an electric field analysis device for a rare earth metal electrolysis furnace.

[0060] Please refer to the attached document. Figures 1 to 5 The device is physically deployed in a rare earth metal electrolysis production workshop and works in conjunction with one or more industrial-grade rare earth metal electrolysis furnaces.

[0061] The device forms a complete sensing, computing and control closed loop in terms of logic and function. Its core objective is to acquire the dynamic electric field distribution in the electrolytic furnace due to electrode consumption and deposition in real time, and to make precise control of process parameters based on this distribution.

[0062] The entire device consists of a real-time electrode morphology sensing module, a multi-source data fusion and electric field reconstruction module, a process parameter dynamic control module, and a data storage and trend analysis module.

[0063] These modules are interconnected with a real-time data bus via a high-speed industrial Ethernet network, ensuring millisecond-level synchronization between data flow and control flow.

[0064] The real-time electrode morphology sensing module is the front end of the device for acquiring real-time data from the field.

[0065] The main hardware of the module is installed inside a special high-temperature resistant and corrosion-resistant protective cover directly above the electrolytic furnace body.

[0066] The protective cover adopts a double-layer water-cooled structure with circulating cooling water inside to ensure that the electronic equipment inside the cover can work stably for a long time in the environment of high temperature radiation and corrosive gas continuously emitted from the furnace opening.

[0067] The bottom of the protective cover has several observation windows with precise optical design. The observation windows are made of high-temperature resistant quartz glass and coated with an anti-reflective film and an anti-fouling coating.

[0068] The placement of the observation window was rigorously calculated to ensure that its field of view could completely cover the anode working end face area below the molten salt surface in the electrolytic furnace and the cathode deposit growth area.

[0069] Please refer to the attached document. Figure 3 The real-time electrode morphology sensing module integrates at least two high-resolution laser scanning units and one industrial vision imaging unit.

[0070] The laser scanning unit uses a pulsed lidar sensor with a wavelength of 1550 nanometers.

[0071] The reason for choosing this wavelength is that it has good penetration of visible light and near-infrared interference light generated during electrolysis, and can be effectively reflected by the molten salt surface.

[0072] Each laser scanning unit includes a laser emitter, a high-speed galvanometer device, and a high-sensitivity photodetector.

[0073] The laser emitter emits a nanosecond-level short pulse laser beam at a fixed frequency of 10 Hz. The high-speed galvanometer device drives the laser beam to deflect rapidly on a two-dimensional plane according to a preset scanning mode, thereby scanning the target area point by point.

[0074] When a single laser beam reaches the surface of the anode or cathode deposit, it is reflected. The reflected light is received by a photodetector, and the distance from each scanning point to the sensor is accurately calculated by measuring the laser's time of flight.

[0075] Each laser scanning unit has a single-point ranging accuracy of 0.1 mm and an angular resolution of 0.01 degrees, and can generate three-dimensional point cloud data containing hundreds of thousands of spatial points in a single scanning cycle.

[0076] The industrial vision imaging unit and the laser scanning unit are rigidly connected to the same mounting base to ensure that the relative positional relationship between their optical axes remains constant.

[0077] The industrial vision imaging unit includes a multispectral industrial camera equipped with a switchable filter wheel. The filter wheel is fitted with narrowband filters, the center wavelengths of which correspond to the visible light band and specific infrared bands, respectively.

[0078] During the electrolysis process, the camera automatically switches filters according to a preset program or an external trigger signal.

[0079] When using a visible light narrowband filter, the camera mainly acquires images of the macroscopic morphology and relative position of the electrodes and molten salt;

[0080] When using an infrared narrowband filter, the camera switches to infrared thermal imaging mode to acquire images of the temperature distribution on the electrode surface and in the molten salt region.

[0081] The core function of narrowband filters is to filter out the intense and unstable arc light spectrum components generated during molten salt electrolysis, allowing only background light or thermal radiation of specific wavelengths to pass through, thereby significantly improving the signal-to-noise ratio and clarity of the image.

[0082] The camera acquires images at a 10 Hz frame rate synchronized with the laser scan, generating a timestamp-aligned visible light and infrared thermal imaging video stream.

[0083] The real-time electrode morphology sensing module also integrates a high-performance edge computing unit.

[0084] This edge computing unit employs an industrial-grade multi-core processor and is equipped with a dedicated graphics processor and field-programmable gate array (FPGA) chip for real-time preprocessing and fusion of massive amounts of raw data acquired by the laser scanning unit and the industrial vision imaging unit. The preprocessing process is carried out in stages.

[0085] The first stage is data cleaning and noise reduction. For laser point cloud data, the edge computing unit first applies a statistical outlier removal algorithm to eliminate false reflection points caused by molten salt splashes, bubbles, or dust.

[0086] The specific algorithm is as follows: calculate the average and standard deviation of the distances from all points in the neighborhood of each point to that point, and identify points whose distances are greater than three times the standard deviation of the average as noise points and delete them.

[0087] The neighborhood size is set to the 10 nearest neighbors of the given point. The KD tree algorithm is used to quickly find the neighborhood point set (KD tree construction time is controlled within 5 milliseconds / frame) to avoid neighborhood selection bias caused by uneven point cloud density in molten salt environment. For infrared image data, the median filter window size is set to 3×3 pixels, the standard deviation of Gaussian filter is σ=1.2, and the electrode edge features are extracted by Sobel operator after filtering. The edge intensity threshold is set to 80 (gray value range of 0-255) to retain effective electrode contour information.

[0088] For image data, mean filtering and Gaussian filtering algorithms are applied to smooth image noise and enhance edge features.

[0089] The second stage involves the temporal and spatial alignment of multi-sensor data. The edge computing unit pre-stores a precisely calibrated multi-sensor joint calibration matrix.

[0090] This matrix describes the transformation relationships between the lidar coordinate system, the camera coordinate system, and the world coordinate system with the center of the electrolytic furnace as the origin.

[0091] For each frame of synchronously acquired point cloud and image, the edge computing unit performs a coordinate system registration operation.

[0092] This operation is mathematically equivalent to solving a nonlinear least squares optimization problem. Its goal is to find the optimal rotation matrix and translation vector so that the spatial position overlap error of the same set of feature points observed from different sensors and from different angles is minimized after transformation to the world coordinate system.

[0093] The specific optimization objective function is defined as follows:

[0094] ;

[0095] in These are the coordinates of feature points (such as the corners of electrode guide rods and furnace body markings) in the laser point cloud / visual image coordinate system. The coordinates of the reference feature points in the world coordinate system are obtained through offline calibration; the rotation matrix is ​​solved using the Levenberg-Marquardt iterative algorithm. (3×3 matrix) and translation vector (3×1 vector), the initial values ​​for iteration are set to the identity matrix and the zero vector, and the iteration terminates when the error change between two consecutive iterations is less than 1×10. -6 The accuracy is measured in millimeters, and the registration time for a single operation is controlled within 10 milliseconds, ensuring the time synchronization of data from multiple sensors.

[0096] These features include physical markers pre-set on the furnace structure and specific angles of the electrode guide rods.

[0097] After solving the problem through an iterative optimization algorithm, all point cloud data and image pixel coordinates were uniformly transformed to the world coordinate system.

[0098] The third stage is multi-frame data fusion. Since a single scan may result in incomplete data due to occlusion or reflectivity issues, the edge computing unit will fuse point cloud data that has been registered to the same coordinate system in multiple consecutive scan cycles.

[0099] The fusion algorithm is based on the principle of iterative nearest point. It stitches together multiple point clouds into a more complete 3D model with higher point density by feature matching and rigid body transformation estimation between point clouds.

[0100] The specific steps are as follows:

[0101] Step 1: Select the fused point cloud from the previous frame as the target point cloud E, and the registered point cloud from the current frame as the source point cloud S;

[0102] Step 2: Select matching feature point pairs from S and E using the Random Sample Consensus (RANSAC) algorithm (sampling times = 1000 times, inlier threshold = 0.1 mm), and remove outliers caused by molten salt bubbles;

[0103] Step 3: Calculate the rigid body transformation matrix (including rotation and translation) that minimizes the sum of squared Euclidean distances between feature points.

[0104] Step 4: After applying the transformation matrix to the source point cloud S, merge it with the target point cloud E, and use voxelgrid downsampling (voxel size = 0.2 mm) to remove redundant points;

[0105] Step 5: Repeat steps 1 to 4 until the average registration error of the point clouds in adjacent frames is less than 0.05 mm. The final output is a structured electrode morphology dataset containing timestamps, with a point cloud density of ≥500 points / square centimeter.

[0106] Finally, the edge computing unit outputs structured data packets.

[0107] The data package contains timestamps, a 3D point cloud model of the anode working face, a 3D point cloud model of the cathode deposit surface, and 2D feature information extracted from the images.

[0108] The data packet accurately characterizes the precise three-dimensional contours, surface morphology, and relative spatial relationship between the anode and cathode at the current moment, and is encapsulated in gigabit Ethernet protocol before being sent to the multi-source data fusion and electric field reconstruction module.

[0109] The multi-source data fusion and electric field reconstruction module is the core computing engine of the device, deployed in the industrial server cluster in the workshop. The core task of this module is to dynamically reconstruct the three-dimensional electric field distribution in the molten salt electrolyte within the electrolytic furnace based on real-time sensed electrode morphology data.

[0110] Please refer to the attached document. Figure 2 With appendix Figure 4 The module's operation begins with a high-fidelity digital twin model. This digital twin model is a virtual three-dimensional computational domain that strictly corresponds to the geometry and material properties of the physical electrolytic furnace.

[0111] The initial geometric parameters of the model are entirely derived from the design drawings of the electrolytic furnace, including the inner wall dimensions of the furnace, the coordinates of the initial anode installation position, the layout of the cathode guide rod, and the positions of the metal outlet and slag discharge outlet.

[0112] To ensure that the virtual model and the physical entity are completely consistent in space, the device performs an offline high-precision calibration process before its first commissioning.

[0113] Calibration is performed in the empty, cold state of the electrolytic furnace. Operators use high-precision three-dimensional measuring equipment such as laser trackers or photogrammetric devices to accurately measure the spatial coordinates of hundreds of feature points on the inner wall of the furnace, the initial installation posture and immersion depth of each anode, and the precise position of each cathode guide rod.

[0114] These measurement data are input into the device as reference values. The corresponding geometric elements in the digital twin model are fine-tuned according to these measurement values ​​to complete the initial spatial alignment between the model and the real object, with the alignment error controlled within 0.5 mm.

[0115] The computational domain of the digital twin model is discretized using an adaptive unstructured tetrahedral mesh.

[0116] The grid generation algorithm will automatically encrypt the key areas.

[0117] Specifically, the grid density is highest in the region near the anode surface, the region near the cathode surface, and the gap region between the anode and cathode. The grid size can be refined to the 1 mm level to ensure that the subtle influence of the complex morphology of the electrode surface on the electric field distribution can be accurately characterized.

[0118] In other areas of the furnace, away from the electrodes, the grid size gradually increases to balance computational accuracy with computational resource consumption. The total number of grids in the entire computational domain is in the millions.

[0119] The core of the multi-source data fusion and electric field reconstruction module is an embedded real-time solver. This solver adopts a hybrid programming architecture, with the core computational kernel written in C++ and Fortran to ensure numerical computation efficiency, and the peripheral control and data interfaces written in Python to improve the device integration flexibility.

[0120] The solver continuously monitors two data sources:

[0121] Structured electrode morphology data packets updated at a frequency of 10 Hz from the real-time electrode morphology sensing module.

[0122] Voltage and current data are read in real time from the electrolysis power supply system via an industrial data acquisition interface, with a sampling frequency of 10 Hz. Upon receiving new data, the solver immediately initiates a real-time calculation cycle.

[0123] Within each computation cycle, the solver first dynamically updates the boundary conditions of the digital twin model.

[0124] It maps the received anode surface point cloud model and cathode surface point cloud model onto the corresponding boundary nodes of the digital twin model mesh using a spatial interpolation algorithm.

[0125] Spatial interpolation employs the radial basis function (RBF) interpolation method, selecting a Gaussian function. As a basis function Spatial radial distance, shape parameter =5 mm, adapted to the micro-morphology of electrode surface; firstly, the point cloud data is projected onto the boundary grid surface of the digital twin model, and the correspondence between the point cloud and the grid node is determined by the parametric mapping of the grid surface. Then, the displacement of each boundary node is calculated by interpolation to realize the dynamic update of the anode surface profile, cathode deposit profile and anode-cathode spacing; for the anode-cathode spacing update, the distance between the nearest point pair of the anode surface point cloud and the cathode surface point cloud is calculated, and the average value is taken as the real-time spacing of the grid area. The update frequency is consistent with the point cloud acquisition frequency (10Hz).

[0126] This means that the boundary of the computational domain is no longer a fixed geometry, but dynamically deforms as the electrodes are consumed and deposited in real time.

[0127] Meanwhile, the real-time voltage value read from the electrolysis power supply system is used as the excitation source loading condition and applied to the anode boundary node, while the cathode boundary node is set as a zero potential reference point.

[0128] The conductivity parameter of the electrolyte is pre-assigned as a material property to all mesh elements within the computational domain.

[0129] Considering the influence of molten salt composition and temperature on conductivity, the device maintains a conductivity-temperature-composition lookup table and dynamically interpolates and corrects the conductivity values ​​based on real-time molten salt temperature data.

[0130] After the boundary conditions and excitation sources are ready, the solver calls the potential field solver based on the finite volume method to perform numerical calculations.

[0131] The governing equations to be solved are the Laplace equations describing the steady current field distribution.

[0132] The specific process is as follows:

[0133] The three-dimensional computational domain of the digital twin model is discretized using an adaptive unstructured tetrahedral mesh (total mesh count = 1.2 × 10⁻⁶). 6 (Electrode surface grid size = 1 mm, grid size in areas away from the electrode = 5 mm);

[0134] Laplace equation Integrating over each control volume (centered on a grid node), and applying the Gaussian divergence theorem, the volume integral is converted into a surface integral over the control surface, yielding the discrete algebraic equation:

[0135] ;

[0136] in To control the conductivity of the molten salt surface (obtained by temperature interpolation), Initial molten salt conductivity =8S / m, coefficient =0.02, For real-time molten salt temperature, (Initial molten salt temperature) To control the surface area, The distance between adjacent nodes. The potential of adjacent nodes. This represents the potential of the current control body's central node.

[0137] Substitute the real-time voltage (anode boundary) and zero potential (cathode boundary) as boundary conditions into the discrete equations to assemble a sparse linear system of equations (the proportion of non-zero elements in the coefficient matrix is ​​≤0.5%).

[0138] The solution is obtained by combining explicit and implicit iterative algorithms: First, the local electric field disturbance (such as the field strength change caused by local anode consumption) is updated quickly through the explicit scheme (time step = 10-4 seconds). Then, the full field potential distribution is solved through the implicit scheme (using the conjugate gradient method, with preconditioner = ILU(0)). The iteration terminates when the residual norm R < 1e-5 (the residual is calculated using the L2 norm).

[0139] via potential gradient Calculate the electric field intensity vector The gradient is calculated using the central difference scheme to ensure that the direction and magnitude accuracy of E meet the process requirements (field strength calculation error ≤ 2%).

[0140] The solver employs an iterative algorithm that combines explicit and implicit methods.

[0141] Within each computation time step, it first uses an explicit format to quickly process the local electric field disturbances caused by changes in the shape of the electrode boundary. This step is fast but has slightly poor stability.

[0142] Next, an implicit scheme is used to solve the potential distribution across the entire field in a stable manner. Although the implicit scheme involves a large amount of computation per step, it is unconditionally stable and can guarantee the convergence of the solution.

[0143] The two formats are iterated alternately until the residual norm calculated for the entire field converges to a preset threshold of 1×10⁻⁶. -5 The convergence criterion is as follows:

[0144] ;

[0145] Represents the residual. To calculate the total number of grid nodes within the domain, Indicates the first After the nth iteration The potential value of each node, Indicates the first The potential value of the node after the next iteration.

[0146] When the residual Less than When the iteration is considered to have converged.

[0147] The solver then terminates the calculation and outputs the three-dimensional electric field distribution results under the current operating conditions.

[0148] The output includes the potential scalar value at each grid node in the computational domain, as well as the electric field intensity vector value calculated by the potential gradient.

[0149] These data constitute a dynamic electric field cloud map of the internal space of the electrolytic furnace, clearly revealing the spatial distribution, direction, and strong and weak regions of the electric field intensity.

[0150] The entire real-time solution process, from data reception to result output, takes less than 100 milliseconds, ensuring real-time synchronization with the production process.

[0151] The process parameter dynamic control module receives three-dimensional electric field distribution data from the multi-source data fusion and electric field reconstruction module, and generates precise process control commands accordingly.

[0152] Please refer to the attached document. Figure 5 This module contains two core sub-modules: an electric field uniformity assessment sub-module and a control strategy generation sub-module.

[0153] The electric field uniformity assessment submodule first performs spatial statistical analysis on the three-dimensional electric field data.

[0154] It does not analyze the entire furnace space, but focuses on the key process areas that play a decisive role in the electrolytic deposition process. These areas are usually defined as a spatial layer within a specific distance above the cathode deposit surface.

[0155] The key process area is specifically defined as: a spatial region extending 50 mm upwards from the cathode deposit surface to 10 mm below the molten salt surface (this region is the core area for rare earth ion reduction deposition), using a spatial trimming algorithm (based on grid nodes). Coordinate filtering Extract grid node data for this region from the overall electric field data; the electric field uniformity evaluation index is calculated as follows:

[0156] coefficient of variation , where the average Standard deviation , This represents the total number of grid nodes in the region. For the first The electric field strength modulus of each node;

[0157] Extreme value ratio ( This is the maximum field strength in the region. (Minimum field strength within the region); process requirements ≤15%, If the value is ≤2.0, it is considered to be an uneven electric field.

[0158] Within this region, the evaluation submodule calculates the statistical characteristics of the electric field intensity magnitudes on all grid nodes:

[0159] This includes the mean, standard deviation, maximum, and minimum values. The electric field uniformity evaluation index consists of two quantitative parameters:

[0160] The first is the coefficient of variation of the electric field strength, which is the ratio of the standard deviation to the mean. The smaller the value, the better the uniformity.

[0161] The second is the extreme value ratio, which is the ratio of the maximum value to the minimum value.

[0162] Meanwhile, the module presets two thresholds based on process knowledge: the weak field threshold is set to 80% of the average electric field strength of the region, and the strong field threshold is set to 120% of the average electric field strength of the region.

[0163] The evaluation submodule uses a spatial query algorithm to quickly identify all sets of grid cells with electric field strength less than the weak field threshold and all sets of grid cells with electric field strength greater than the strong field threshold.

[0164] The specific steps for identifying abnormal regions are as follows:

[0165] Set weak field threshold Strong field threshold ( (meaning the average field strength in the critical process area), traverse all mesh nodes in this area and mark them. The node is a weak field point. The nodes are strong field points; For the first The field strength value corresponding to each grid node;

[0166] K-means clustering algorithm was used to cluster weak field points and strong field points separately, and the number of clusters was [not specified]. Determined by the elbow rule ( The initial cluster centers were selected using the K-means++ algorithm to avoid local optima.

[0167] For each cluster, calculate the coordinates of the geometric center of its smallest circumscribed cuboid. (Based on the world coordinate system of the electrolytic furnace), volume of the cuboid (Unit: cubic meters) and deviation of average field strength within the cluster ( (mean field strength of nodes within the cluster).

[0168] Filtering valid anomaly regions: Weak field regions must meet the following requirements cubic meters and Strong field area must meet cubic meters and This process eliminates transient, small-area anomalies caused by molten salt flow.

[0169] It not only calculates the total area ratio of these anomalous regions, but more importantly, it uses clustering analysis algorithms to aggregate these discrete anomalous grid units into several continuous "weak field regions" and "strong field regions", and calculates the geometric center coordinates, area size, and average field strength deviation of each anomalous region.

[0170] The regulation strategy generation submodule makes decisions based on the detailed evaluation results described above. Its decision-making logic is based on a pre-defined, interpretable fuzzy rule base.

[0171] This rule base was not written directly from human experience, but was generated through machine learning training.

[0172] The training data comes from the historical production database. The device extracts a large amount of electric field reconstruction data from the database at historical moments and its corresponding process operation records that were later proven to be optimal.

[0173] The gradient boosting decision tree algorithm is used for training, with electric field uniformity index, the number, area, and location distribution of weak and strong field regions as input features, and actual and effective control actions as output labels.

[0174] After training, the algorithm generates a series of fuzzy rules in the form of "if-then".

[0175] For example, a typical rule might be: "If" a weak field region with an area greater than 0.01 square meters and a duration greater than 3 seconds is detected in cathode A region, and "the electric field strength in that region is only 70% of the average value, then" the current setting value of cathode A branch is increased by 5% of its current value.

[0176] The core rules of the fuzzy rule base also include:

[0177] If the geometric center of the strong field region is located in the anode A1 region, and ΔE=15%, V=0.002 cubic meters, and the duration is ≥2 seconds, then the anode A1 lifting mechanism is controlled to descend by 0.3 mm.

[0178] If multiple weak field regions are scattered (cluster number is 4), and the average Then, the current in all corresponding cathode branches will be increased by 3%-5% simultaneously (according to...). (Linear distribution); For the proportional-integral (PI) calculation of current regulation, the formula is: ,in, For the changed current, the proportionality coefficient =0.05 (current adjustment coefficient / field strength deviation rate), integral coefficient =0.02 (current adjustment coefficient / (field strength deviation rate·second), integration time window = 3 seconds, limited output) (To avoid arc instability caused by sudden current changes); for anode lifting and lowering adjustments, the displacement... , =0.02 mm / (field strength deviation rate), displacement accuracy controlled within 0.1 mm, lifting speed = 0.5 mm / s.

[0179] These rules form the knowledge base for regulatory strategies.

[0180] When the real-time evaluation results trigger one or more rules, the control strategy generation submodule starts the instruction generation process.

[0181] For current regulation, the device is designed for a multi-channel programmable DC power supply. The electrolytic power supply system equips each group of cathodes or anodes with an independent silicon controlled rectifier or transistor switching power supply module.

[0182] The regulation strategy generation submodule calculates the target current value of the cathode branch or anode branch that needs to be adjusted based on the rule reasoning results.

[0183] The current adjustment is not done in one step, but is calculated using a closed-loop proportional-integral control algorithm.

[0184] The algorithm calculates the adjustment amount in real time based on the current degree of electric field anomaly and the dynamic change trend of the anomaly region.

[0185] The proportional coefficient and integral time constant in proportional-integral calculations are not fixed values, but are adaptively switched according to different stages of the electrolysis process.

[0186] For example, in the early stages of electrolysis, when the cathode deposits are just beginning to grow, the device uses more conservative control parameters with a smaller proportional coefficient and a longer integral time to avoid over-adjustment that could cause process fluctuations. In the stable production period, the device uses more aggressive control parameters with a larger proportional coefficient and a shorter integral time to quickly smooth out electric field disturbances.

[0187] The calculated current adjustment command is sent to the corresponding power module via a high-speed fieldbus, such as PROFINET or EtherCAT, at millisecond intervals.

[0188] After the power module executes the command, it feeds back the actual output current value to the device, forming a current closed loop.

[0189] For strong field regions caused by excessively rapid local consumption of the anode, the control strategy generation submodule will generate micro-motion control commands for the anode lifting mechanism.

[0190] The anode of an electrolytic furnace is usually composed of multiple sets of anode blocks, each set of anodes is equipped with an independent electric or hydraulic lifting mechanism.

[0191] The module maps the spatial coordinates of the strong field zone to the specific anode block number, and then generates the displacement command for the anode block to be raised or lowered, with the displacement accuracy controlled within 0.1 mm.

[0192] The command is also sent to the corresponding anode lifting controller via the fieldbus. After all control commands are sent, the device continuously monitors the feedback from the electric field reconstruction module, that is, observes whether the electric field strength in the target area changes in the expected direction.

[0193] If the expected results are not achieved within a certain time window, the strategy generation submodule will re-evaluate and generate new adjustment instructions until the electric field anomaly is eliminated or reduced to an acceptable range.

[0194] The verification standard for the effect of regulation feedback is:

[0195] After weak field region regulation, the average field strength of nodes within the cluster And the duration is ≥1 second;

[0196] After strong field region regulation, the average field strength of nodes within the cluster ,and cubic meter;

[0197] Electric field uniformity index , The feedback monitoring time window is 2 seconds. If the above standard is not met within 2 seconds, the adjustment amount will be recalculated until the requirements are met. If the standard is not met after 5 consecutive adjustments, an early warning signal will be triggered, prompting the operator to check the electrode mechanical structure or molten salt composition.

[0198] The data storage and trend analysis module serves as the device's data hub and value mining engine, operating continuously.

[0199] This module is equipped with a high-performance time-series database, which is used to receive and store all time-series data generated by all the aforementioned modules.

[0200] The stored data sequence includes:

[0201] The data includes raw point cloud and image data summaries sampled at 10 Hz, structured data packages of electrode morphology, three-dimensional mesh data and field strength distribution summaries output from each electric field reconstruction, all issued process control commands and their actuator feedback status, and voltage and current time series of the electrolysis power supply system.

[0202] All of these data are timestamped to the millisecond level, ensuring that the data across the entire chain is traceable and correlated.

[0203] The data analytics engine periodically performs in-depth mining and trend analysis on the massive amounts of stored data.

[0204] One of the core analytical tasks is tracking the long-term trend of electric field uniformity.

[0205] The engine calculates the average electric field uniformity index daily, weekly, and monthly, and plots its curves over time.

[0206] By analyzing the trend of the curve, it can be determined whether the overall stability of the electrolysis process is improving or deteriorating.

[0207] Another key task is electrode consumption and deposition prediction.

[0208] The engine analyzes the long-term variation of the electric field intensity above a specific anode region by comparing it with historical data from the same period.

[0209] If the average electric field strength above a certain anode shows a continuous and slow upward trend, it often indicates that the anode is thinning due to consumption, resulting in a decrease in the resistance between it and the cathode.

[0210] The device can predict the remaining lifespan of the anode based on this information and generate a maintenance warning report in advance, indicating that the anode needs to be replaced at a planned downtime in the future.

[0211] Similarly, for the cathode, by analyzing the historical patterns of electric field distribution corresponding to the morphology of sediments in different regions of the cathode, the growth trend of cathode nodules can be predicted, and the cathode cleaning operation that may need to be performed can be warned in advance to avoid short circuits caused by uneven deposition.

[0212] In addition, all the stored structured data constitutes a rich process data asset library, which can be used to support research on more advanced process optimization algorithms, such as finding the global optimal process parameters based on deep learning, or to provide real-world verification data for the electromagnetic field design of new energy-saving electrolytic furnaces.

[0213] The device described in this embodiment achieves millimeter-level precision dynamic geometric measurement through a real-time electrode morphology sensing module, realizes online three-dimensional electric field calculation 10 times per second through a multi-source data fusion and electric field reconstruction module, achieves millisecond-level intelligent closed-loop control through a process parameter dynamic adjustment module, and realizes continuous accumulation and mining of process knowledge through a data storage and trend analysis module.

[0214] These four modules are interconnected, forming a complete digital twin closed loop from physical world perception to virtual world reconstruction and analysis, and then feedback control of the physical world. This enables real-time insight and proactive optimization of the electric field inside the rare earth metal electrolysis furnace, a key physical field, fundamentally improving the accuracy, efficiency, and stability of the electrolysis process.

[0215] Example 2: This example provides an alternative implementation scheme for an electric field analysis device for a rare earth metal electrolysis furnace. It is consistent with Example 1 in its core principle, but the data fusion method of the real-time electrode morphology sensing module has been specifically optimized to adapt to application scenarios with higher precision requirements or more complex working conditions.

[0216] In the real-time electrode morphology sensing module, this embodiment employs an assisted positioning system based on ultra-wideband wireless positioning technology to further enhance the robustness and accuracy of multi-sensor data fusion.

[0217] The environment of an electrolytic furnace is harsh; high temperatures, corrosive gases, and metal vapors can occasionally cause temporary interference with laser scanning or visual imaging, leading to missing or degraded data from individual sensors. Therefore, miniature ultra-wideband positioning tags are pre-installed on the furnace structure and on specific non-contact locations such as the hangers of each anode group and the cathode guide rods.

[0218] These tags are small, heat-resistant, and emit ultra-wideband pulse signals at a fixed frequency.

[0219] The ultra-wideband positioning base station array installed inside the protective cover receives these signals and calculates the three-dimensional spatial coordinates of each tag in real time at a frequency of hundreds of times per second by measuring the time difference of arrival of the signals, with a positioning accuracy of up to the centimeter level.

[0220] When performing multi-sensor data registration, the edge computing unit not only relies on image feature points, but also introduces the real-time coordinates of these ultra-wideband labels as an additional, high-frequency constraint into the coordinate system-registration optimization model.

[0221] When laser or visual data is temporarily lost due to interference, the device can rely on ultra-wideband positioning data to maintain basic tracking of the macroscopic position of the electrodes.

[0222] When all sensor data are normal, the ultra-wideband data, as a strong constraint, can reduce the uncertainty of the registration optimization problem. Especially when the electrodes undergo rapid and minute deformation, it can provide a more stable and reliable spatial reference, thus making the output electrode morphology structured data package more available and accurate even under extreme conditions.

[0223] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0224] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An electric field analysis device for a rare earth metal electrolysis furnace, characterized in that, The application relates to a real-time sensing and dynamic control system for a molten salt electrolysis furnace, which comprises the following modules: an electrode shape real-time sensing module for non-contact real-time acquisition of three-dimensional geometric shape data of an anode and a cathode in a molten salt electrolysis furnace, the electrode shape real-time sensing module comprising at least two groups of high-resolution laser scanning units and one group of industrial visual imaging units; the electrode shape real-time sensing module is internally provided with an edge computing unit for preprocessing of collected original point cloud data and image data, the preprocessing process comprising point cloud noise reduction, coordinate system registration and multi-frame data fusion based on feature point matching, and outputting a structured data package comprising a time stamp and representing the accurate three-dimensional profile and relative spatial position relationship of the anode and the cathode at the current moment; In the formula, the point cloud noise reduction adopts a statistical outlier removal algorithm based on 10 nearest neighbor points, and a point with a distance greater than 3 times the standard deviation of the average distance of the neighborhood points is determined as a noise point; the coordinate system registration solves a nonlinear least squares optimization problem by using a Levenberg-Marquardt iteration algorithm, and the objective function is . are feature point coordinates in a laser point cloud / visual image coordinate system, are reference feature point coordinates in a world coordinate system, is a rotation matrix, is a translation vector; The iteration termination condition is that the error variation of two adjacent iterations is less than 1x10 -6 The multi-frame data fusion is based on the iterative nearest point principle, the matching feature point pairs are screened by combining the random sample consensus algorithm, the inlier threshold is set to 0.1 millimeter, and the structured data package with a point cloud density of greater than or equal to 500 points / square centimeter is finally output. a multi-source data fusion and electric field reconstruction module for dynamic reconstruction of the three-dimensional electric field distribution in the molten salt electrolysis furnace based on the structured data package, a process parameter dynamic control module for analyzing the reconstructed three-dimensional electric field distribution and generating process control instructions, the process parameter dynamic control module comprising an electric field uniformity evaluation submodule and a control strategy generation submodule; the multi-source data fusion and electric field reconstruction module firstly constructs an initial three-dimensional calculation domain digital twin model strictly corresponding to the geometric size of a physical molten salt electrolysis furnace; the digital twin model adopts self-adaptive unstructured tetrahedral mesh, and the mesh is encrypted in the preset anode and cathode area; the core of the multi-source data fusion and electric field reconstruction module is an embedded real-time solver; the embedded real-time solver receives the structured data package from the electrode shape real-time sensing module and synchronously accesses real-time voltage and current data provided by an electrolysis power supply system; the embedded real-time solver takes the digital twin model as a calculation framework, takes the real-time acquired anode surface profile, cathode deposit surface profile and anode-cathode spacing as dynamic updated boundary conditions, takes real-time voltage as an excitation source loading condition, and performs real-time numerical solving on the potential field control equation based on the finite volume method according to preset molten salt electrolyte conductivity parameters; The boundary condition dynamic updating adopts a radial basis function (RBF) interpolation method to map the electrode surface point cloud to the boundary grid nodes of the digital twin model, and the distance between the cathode and the anode is calculated by the average value of the nearest point pair distance of the point cloud; the potential field control equation is the Laplace equation , which is discretized by the finite volume method ; for the control surface area, for the distance between adjacent nodes, for the potential of adjacent nodes, for the potential of the central node of the current control body, for the molten salt conductivity of the control surface, the solution is first taken as 1 x 10 -4 seconds time step of the explicit format to update the local electric field disturbance, and then the implicit format of the conjugate gradient method is used for full field solution, and the L2 norm is used for residual convergence judgment, and the threshold is 1 x 10 -5 ; The control equation is iteratively calculated in each calculation time step until the residual converges to a preset threshold 1 x 10 -5 Thus, the potential value and the electric field intensity vector at each calculation grid node in the internal space of the electrolytic furnace under the current working condition are output.

2. The electric field analysis device for a rare earth metal electrolysis cell according to claim 1, characterized by the laser scanning unit emits a scanning laser beam at a fixed frequency to collect three-dimensional point cloud data of the working end surface of the anode immersed below the molten salt liquid surface and the surface of the cathode deposit; the industrial visual imaging unit synchronously collects visible light and infrared thermal imaging video streams of the observation window area of the molten salt electrolysis furnace.

3. The electric field analysis device for a rare earth metal electrolysis cell according to claim 2, characterized by the electric field uniformity evaluation submodule performs spatial statistical analysis on the three-dimensional electric field data output by the multi-source data fusion and electric field reconstruction module, calculates the ratio of the standard deviation, the maximum value and the minimum value of the electric field intensity in the key process area as the uniformity evaluation index, and identifies the spatial coordinates of the weak field area with an electric field intensity less than 80% of the preset threshold value and the strong field area with an electric field intensity greater than 120% of the preset threshold value; the control strategy generation submodule makes decisions based on the electric field uniformity evaluation result and preset process targets, and the decision logic is based on a preset fuzzy rule base. The control strategy generation submodule generates instructions and sends them to the multi-channel programmable DC power supply unit of the electrolysis power supply system through the industrial bus communication interface when it identifies that there is a persistent weak field in the cathode area, to fine-tune the current distribution ratio of the corresponding cathode branch, and the current adjustment amplitude is calculated in proportion to the weak field degree. The formula is: wherein, is the changed current, the proportional coefficient is 0.05 (current adjustment coefficient / field strength deviation rate), the integral coefficient is 0.02 (current adjustment coefficient / (field strength deviation rate·sec), the integral time window = 3 seconds, the amplitude output ; is the average field strength deviation value within the cluster; The control strategy generation submodule generates micro-motion control instructions for the anode lifting mechanism to locally adjust the anode immersion depth for the strong field area caused by uneven anode consumption.

4. The electric field analysis device for a rare earth metal electrolysis cell according to claim 3, characterized by The coordinate system registration performed by the edge computing unit is realized by solving a least squares optimization problem containing a rotation matrix and a translation vector, and the laser point cloud coordinate system and the visual image coordinate system are unified to the world coordinate system with the electrolytic furnace center as the origin.

5. A device for analyzing the electric field of a rare earth metal electrolysis cell according to claim 4, characterized in that, The digital twin model constructed in the multi-source data fusion and electric field reconstruction module has initial geometric parameters derived from the design drawings of the electrolytic furnace and corrected through offline high-precision calibration. The offline high-precision calibration process is as follows: in the cold state of the empty electrolytic furnace, a high-precision three-dimensional measuring instrument is used to obtain the accurate spatial coordinates of the inner wall of the furnace, the initial installation position of the anode, and the cathode guide rod, and these data are input as reference values into the digital twin model to complete the initial spatial alignment of the model and the actual object.

6. A device for analyzing the electric field of a rare earth metal electrolysis cell according to claim 5, characterized in that, The embedded real-time solver uses an iterative algorithm combining explicit and implicit methods to balance the calculation speed and stability. In each real-time calculation period, the electric field disturbance caused by boundary changes is quickly updated using the explicit format, and then the implicit format is used for stable solving of the full-field potential distribution.

7. The electric field analysis device for a rare earth metal electrolysis cell according to claim 6, characterized in that The fuzzy rule base in the process parameter dynamic control module is trained by machine learning using historical optimal production data and electric field reconstruction data at the corresponding time; The training process uses the gradient boosting decision tree algorithm, takes the electric field uniformity index and regional distribution characteristics as input features, and takes the actual control actions verified as effective as output labels. After a large number of sample training, an interpretable "if-then" rule set is generated; Typical rules include: If a weak field area with an area of ≥0.01 square meters, a duration of ≥3 seconds, and a field strength of 70% of the average value is detected in the cathode A region, increase the cathode A branch current by 5%; If the field strength deviation rate of the corresponding strong field area in the anode A1 region is ≥15%, the volume is ≥0.002 cubic meters, and the duration is ≥2 seconds, control the anode A1 lifting mechanism to lower 0.3 millimeters.

8. The electric field analysis device for a rare earth metal electrolysis cell according to claim 7, characterized in that The device also includes a data storage and trend analysis module; The data storage and trend analysis module is used to continuously store the full amount of time series data generated by the electrode morphology real-time sensing module, the multi-source data fusion and electric field reconstruction module, and the process parameter dynamic control module. The data storage and trend analysis module has a time series database and a data analysis engine built-in, which can perform long-term trend analysis on the stored data, predict the anode consumption trend and the evolution law of the cathode deposition morphology by comparing the changes in the electric field uniformity index in different time periods, and generate maintenance warning reports.

9. A method of analyzing an electric field of a rare earth metal electrolysis cell, characterized by, The rare earth metal electrolytic furnace electric field analysis device of any one of claims 1-8 is used to realize rare earth metal electrolytic furnace electric field analysis, and the method comprises the following steps: Step 1, through the laser scanning and visual imaging equipment deployed above the electrolytic furnace, three-dimensional point cloud data and image data of the anode working end face and the cathode deposit surface are collected in real time at a frequency greater than 10 Hz; Step 2, the original data collected in step 1 are fused and preprocessed by an edge computing device, and through coordinate registration and data fusion, an electrode real-time shape data set containing accurate three-dimensional space coordinates is generated; Step 3, the electrode real-time shape data set obtained in step 2 is input into a preset digital twin model embedded real-time solver together with voltage and current data read from the electrolysis power supply system in real time; Step 4, in the real-time solver, the boundary conditions of the calculation domain are dynamically updated by the electrode real-time shape data, the excitation source conditions are updated by the real-time voltage data, the finite volume method solver is called to solve the potential field control equation iteratively until convergence, and the full-field potential distribution and electric field intensity vector distribution of the three-dimensional space inside the electrolytic furnace at the current time are output; Step 5, the three-dimensional electric field distribution data output in step 4 are subjected to spatial statistical analysis, the electric field uniformity evaluation index is calculated, and the abnormal area of the electric field intensity is automatically identified; Step 6, based on the analysis result of step 5, a preset fuzzy rule base is called to make decision reasoning, and fine-tuning instructions for the current ratio of the electrolysis power supply system or the position of the anode mechanism are generated; Step 7, the control instructions generated in step 6 are sent to the corresponding executing mechanism through an industrial communication network, and the execution feedback is monitored, a complete cycle from electric field perception, reconstruction analysis to closed-loop control is completed, and the process returns to step 1 for continuous cyclic operation.

Images