Rare earth permanent magnet material production low-delay control system based on digital twinning technology

Abstract

The invention belongs to the technical field of production control, and discloses a rare earth permanent magnet material production low-delay control system based on a digital twinning technology. Comprising the steps of obtaining production sensing data of a rare earth permanent magnet material production line; constructing a digital twinborn model of the production process of the rare earth permanent magnetic material by using the production sensing data; performing simulation by using a digital twinborn model to obtain a change trend of the key production parameters; calculating a sensitivity coefficient matrix of each key production parameter according to the change trend of the key production parameters; constructing a multi-objective optimization function based on the sensitivity coefficient matrix, and solving the multi-objective optimization function by using an improved particle swarm algorithm to obtain an optimal control parameter combination; the production efficiency and the product quality are improved, the energy consumption and the raw material loss are reduced, the resource utilization rate is improved, and green production is promoted.

Description

Technical Field

[0001] The present invention relates to the field of production control technology, and more specifically, to a low-delay control system for rare earth permanent magnet material production based on digital twin technology. Background Art

[0002] The patent with application publication number CN111091944A discloses a lanthanum-rich cerium-yttrium multi-phase fine-grained rare earth permanent magnet material and a preparation method thereof. Based on the characteristics of fast heating speed, short heating time and holding time of spark plasma sintering technology, the mutual diffusion and chemical heterogeneity of elements inside the lanthanum-rich cerium-yttrium multi-phase magnet are controlled, various core-shell morphologies, magnetic hardening and pinning effects are accurately controlled, and the coercive force of the magnet is improved; while using rapid sintering to avoid abnormal grain growth, step-by-step heating is designed, temperature is accurately controlled, and heating is uniform, so as to avoid overheating of local areas during the sintering process and cause the coexistence of coarse and fine grain areas, so as to achieve uniform internal organizational structure of the magnet, fine grains, and an average grain size of less than 5μm, thereby improving the comprehensive magnetic properties of the lanthanum-rich cerium-yttrium multi-phase rare earth permanent magnet material and meeting commercial requirements. The preparation method is simple and convenient, easy to operate, and can accurately control the element distribution and organizational structure of the multi-phase magnet, reduce the preparation cycle and cost, and is suitable for large-scale production.

[0003] However, the existing production control of rare earth permanent magnet materials faces many challenges. First, the modeling and perception capabilities of complex production processes are insufficient, resulting in control accuracy and real-time performance that are difficult to meet the requirements of high-quality production; data utilization efficiency is low, and it is difficult to fully explore and utilize valuable information in the production process, which affects the scientificity and timeliness of decision-making; secondly, the parameter optimization method is relatively rough, and it is difficult to accurately identify and regulate key production parameters, resulting in poor control effects; in addition, it is often difficult to balance multiple production goals, and it is difficult to achieve comprehensive optimization of performance, efficiency, energy consumption and other aspects; at the same time, the system response speed is insufficient, and it is difficult to adjust quickly when facing production fluctuations, which increases the risk of product quality fluctuations; finally, there is a lot of room for improvement in energy management and raw material utilization, which not only affects production efficiency and product quality, but also leads to higher production costs and waste of resources.

[0004] In view of this, the present invention proposes a low-delay control system for the production of rare earth permanent magnet materials based on digital twin technology to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A low-latency control system for rare earth permanent magnet material production based on digital twin technology, comprising: a data acquisition module for acquiring production sensor data of a rare earth permanent magnet material production line;

[0006] The model building module uses production sensor data to build a digital twin model of the rare earth permanent magnet material production process;

[0007] The simulation module uses the digital twin model to simulate and obtain the changing trends of key production parameters;

[0008] Matrix fitting module, used to calculate the sensitivity coefficient matrix of each key production parameter according to the change trend of key production parameters;

[0009] The parameter optimization module constructs a multi-objective optimization function based on the sensitivity coefficient matrix, and uses the improved particle swarm algorithm to solve the multi-objective optimization function to obtain the optimal control parameter combination; each module is connected by wired and / or wireless means.

[0010] Further, the production sensing data includes temperature data, pressure data, flow data, electrical parameters, material property data, fluid related data and diffusion related data;

[0011] Temperature data includes melting furnace temperature, sintering furnace temperature, mold temperature and cooling temperature; pressure data includes press pressure, air pressure and hydraulic pressure; flow data includes raw material input flow, coolant flow and protective gas flow; electrical parameters include motor current, heating power and control system voltage; material property data includes material density, material specific heat capacity and material thermal conductivity; fluid-related data includes fluid viscosity and fluid density; diffusion-related data includes diffusion coefficient and concentration distribution of elements.

[0012] Furthermore, the digital twin model is constructed by:

[0013] The production sensor data is cleaned, denoised, interpolated and standardized, and the preprocessed production sensor data is converted into a unified format; the timestamps of all data are unified to the same frequency and starting point; and the data of different dimensions are dimensionless; the categorical data (one-hot encoding or serial number encoding) is converted into perfect sensor data;

[0014] Construct heat conduction model, fluid dynamics model and diffusion model based on complete production data; establish interfaces between heat conduction model, fluid dynamics model and diffusion model, couple the heat conduction model, fluid dynamics model and diffusion model to obtain a digital twin model.

[0015] Furthermore, the heat conduction model is constructed in the following manner:

[0016] Define the control function and the calculation domain of the control function, the calculation domain includes the melting furnace, the pressing mold, the sintering furnace and the cooling device;

[0017] The control function includes the heat conduction equation, boundary conditions and initial conditions; the expression of the heat conduction equation is: , where is the density, is the specific heat capacity, is the temperature, For time, is the thermal conductivity coefficient, As internal heat source, is the gradient symbol;

[0018] The boundary conditions include known temperature conditions, known heat flow conditions, and a linear combination of known convection and radiation; the initial conditions are a given initial temperature distribution;

[0019] The computational domain is discretized to generate a computational grid; the computational grid includes a structured grid and an unstructured grid; the structured grid and the unstructured grid are encrypted to obtain a model processing grid; the control function is integrated on the basis of the model processing grid to obtain a heat conduction model.

[0020] Furthermore, the encryption method includes:

[0021] Calculate the curvature distribution of the boundary of the computational domain and define the curvature sensitive weight function ; Ways to define a curvature-sensitive weight function include:

[0022] Get the curvature value of any node in structured and unstructured grids based on curvature distribution , if the curvature value of any node Greater than 0, the curvature sensitive weight function ;in, is the reference curvature, is the critical curvature value;

[0023] If the curvature value of any node Less than or equal to 0, the curvature sensitive weight function ;in, and All are adjustable weight coefficients;

[0024] According to the curvature sensitive weight function, the grid size field is constructed ;in is the desired base grid size, and They are the horizontal axis, vertical axis and vertical axis of the three-dimensional coordinate system at the grid size location;

[0025] Using the Grid Size field The structured grid and the unstructured grid are further divided to obtain the model processing grid.

[0026] Furthermore, the further division method includes:

[0027] Traverse all cells in the computational grid and, for each cell's centroid, apply the grid size field Calculate the expected size ;

[0028] like Less than , then the corresponding unit will be encrypted and subdivided; if Greater than , the corresponding units are coarsened and merged;

[0029] The method of performing encryption segmentation includes:

[0030] Based on the expected size , calculate the minimum side length required for the cell ,in is a constant determined by optimization; check all edge lengths of the unit, if there is an edge length Greater than , then the side length The corresponding edges are equally divided to generate new nodes; the original unit is re-divided using the new nodes to generate sub-units; the generated sub-units are recursively subdivided until the side lengths of all sub-units are no greater than ;

[0031] The method of performing coarsening and merging includes:

[0032] Check all the cells in the neighborhood of the cell, if there are two cells whose area difference of any surface is less than the preset difference threshold, merge them to obtain a new cell after merging;

[0033] Calculate the centroid size of the new unit after merging ,in is the area of ​​any face of the unit before merging, is the volume of the unit before merging; if Greater than , then continue to merge with other units in the neighborhood, otherwise stop;

[0034] Define a smoothing objective function, and after coarsening and merging, smooth all units by minimizing the smoothing objective function;

[0035] Smoothing objective function ;in For the The length of the strip, is the target side length, For the The angle between the two sides, is the ideal angle, is the weight coefficient;

[0036] The model processing grid is obtained by solving the minimization smoothing objective function through the optimization algorithm.

[0037] Furthermore, the fluid dynamics model is constructed in the following manner:

[0038] On the basis of the model processing grid, the fluid dynamics control function is defined. The expression of the fluid dynamics control function is: ;in, For the The mass fraction of a substance, For the The diffusion coefficient of a substance, It is The migration coefficient of the substance, For the The coupling weight factor for the material application, For the The source term of a substance, is the energy function;

[0039] Energy function ;in, is the specific heat capacity at constant pressure, For the The total derivative of the mass fraction of a substance, Thermal conductivity;

[0040] The fluid dynamics control function is integrated on the basis of the model processing grid to obtain the fluid dynamics model.

[0041] Furthermore, the diffusion model is constructed in the following manner:

[0042] Based on the model processing grid, the diffusion control function is defined. The expression of the diffusion control function is: ;in, It is The concentration of the element, It is The diffusion coefficient tensor of the element, It is The source term of the element;

[0043] ;in, It is The frequency factor of the element, It is The diffusion activation energy of the element, is the universal gas constant, is a concentration-dependent function, For the The concentration of other elements besides the is the stress-dependent function, is the stress tensor; ,in, is the activation volume, is the Boltzmann constant;

[0044] Concentration-dependent function ,in, and is the fitting parameter, is the reference concentration;

[0045] The diffusion control function is discretized to obtain a discretized diffusion function; the discretized diffusion function is integrated on the basis of the model processing grid to obtain a preliminary fluid dynamics model;

[0046] On the basis of the preliminary fluid dynamics model, diffusion boundary conditions are set to obtain a diffusion model; the diffusion boundary conditions include first-class boundary conditions, second-class boundary conditions and third-class boundary conditions;

[0047] The first type of boundary condition is the concentration value on the specified boundary, the second type of boundary condition is the concentration gradient or diffusion flux on the specified boundary; the third type of boundary condition is a linear combination of the concentration value and diffusion flux on the boundary.

[0048] Furthermore, the calculation method of the sensitivity coefficient matrix includes:

[0049] The output indicators of key production parameters are defined as product performance indicators, production efficiency indicators, energy consumption indicators and raw material utilization indicators;

[0050] The data points on the time series data are taken as sample points, and the sensitivity objective function is defined; the expression of the sensitivity function is: ;in, For sample points The corresponding vector of key production parameters, To add to the sample point A disturbance of is the objective function; is a small perturbation function, is the difference function; For sample points The corresponding sensitivity coefficient;

[0051] Small perturbation function ;

[0052] Difference function ;in, is the total number of sample points, For sample points The coefficients of the corresponding wavelet basis functions are, For sample points The corresponding disturbance point is For the sample points , the constructed random perturbation sequence;

[0053] The objective function is the weighted sum of the performance index function, production efficiency function, energy consumption function and raw material utilization function; the performance index function is defined as the weighted sum of the remanence, coercivity and energy product; the production efficiency function is defined as the product of production capacity and yield rate; the energy consumption function is the sum of the energy consumption of smelting, pressing and sintering; the raw material utilization function is defined as the ratio of actual output to theoretical output;

[0054] Initialize a symmetric matrix and transform the sample points And the corresponding sensitivity coefficient As the elements of the symmetric matrix, the calculation of the sensitivity coefficient matrix is ​​completed;

[0055] The multi-objective optimization function ;in, is a vector of control parameters to be optimized; is the process constraint matrix, is the expected product performance target vector, is the sensitivity coefficient matrix, is the regularization parameter.

[0056] Furthermore, the method for obtaining the optimal control parameter combination includes:

[0057] Initialize the particle swarm and define the number of particles , maximum number of iterations , randomly generated The position of the particle and speed , is the index of the particle; initialize the individual optimal position of each particle And the global optimal position ;

[0058] The multi-objective optimization function is used as the fitness function, and the speed update formula is defined as follows: ;in, is the inertia weight, and is the acceleration factor; and for A random number between For time The position of a particle is randomly selected when For time Time Particle speed, For time Time Particle speed, For time Time Particle The individual optimal position of For time Time Particle The global optimal position of For time Time Particle location;

[0059] Define the position update formula: ;in, For time Time Particle location;

[0060] Define the search space, if If it exceeds the predefined search space, it will be pulled back to the boundary of the search space, the value of the fitness function of each particle's new position will be calculated, and the individual optimal position and the global optimal position of each particle will be compared and updated;

[0061] If the preset maximum number of iterations is reached, the process ends; otherwise, the iteration continues and the optimal solution is output. The global optimal position corresponding to the optimal solution is the optimal control parameter combination.

[0062] The technical effects and advantages of the low-delay control system for rare earth permanent magnet material production based on digital twin technology of the present invention are as follows:

[0063] By constructing an accurate digital twin model, the present invention achieves comprehensive perception and in-depth insight into complex production processes, greatly improving the accuracy and real-time performance of production control; utilizing multi-source heterogeneous data and simulation, it can accurately predict the changing trends of key production parameters, providing a reliable basis for timely adjustment of control strategies; the introduced sensitivity analysis method effectively identifies the most critical control parameters and optimizes the control focus; the multi-objective optimization function and improved particle swarm algorithm achieve a comprehensive balance of product performance, production efficiency, energy consumption and other aspects, thereby improving production efficiency and product quality; the low-latency architecture ensures that the system responds quickly to production fluctuations and effectively reduces the generation of defective products; in addition, it also reduces energy consumption and raw material loss, improves resource utilization and promotes green production. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1This is a schematic diagram of a low-delay control system for producing rare earth permanent magnet materials based on digital twin technology of the present invention;

[0065] Figure 2 This is a schematic diagram of the low-delay control method for producing rare earth permanent magnet materials based on digital twin technology of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0067] See also Figure 1 As shown, the low-latency control system for rare earth permanent magnet material production based on digital twin technology described in this embodiment includes:

[0068] Data acquisition module, used to obtain production sensor data of rare earth permanent magnet material production line;

[0069] The model building module uses production sensor data to build a digital twin model of the rare earth permanent magnet material production process;

[0070] The simulation module uses the digital twin model to simulate and obtain the changing trends of key production parameters;

[0071] Matrix fitting module, used to calculate the sensitivity coefficient matrix of each key production parameter according to the change trend of key production parameters;

[0072] The parameter optimization module constructs a multi-objective optimization function based on the sensitivity coefficient matrix, and uses the improved particle swarm algorithm to solve the multi-objective optimization function to obtain the optimal control parameter combination; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0073] Further, the production sensing data includes temperature data, pressure data, flow data, electrical parameters, material property data, fluid related data and diffusion related data;

[0074] Temperature data includes melting furnace temperature, sintering furnace temperature, mold temperature and cooling temperature; pressure data includes press pressure, air pressure and hydraulic pressure; flow data includes raw material input flow, coolant flow and protective gas flow; electrical parameters include motor current, heating power and control system voltage; material property data includes material density, material specific heat capacity and material thermal conductivity; fluid-related data includes fluid viscosity and fluid density; diffusion-related data includes diffusion coefficient and concentration distribution of elements.

[0075] It should be noted that temperature data, pressure data, flow data and electrical parameters are obtained using corresponding sensors; material property data are obtained by referring to material manuals or databases to obtain standard values; for standard fluids, fluid-related data are obtained by referring to fluid physical property manuals; for specific fluids, they are measured using instruments such as viscometers and densitometers; the diffusion coefficient is obtained by referring to literature data, and the concentration distribution is obtained through sampling analysis or online monitoring equipment.

[0076] Furthermore, the digital twin model is constructed in the following ways:

[0077] The production sensor data is pre-processed by cleaning, denoising, interpolation and standardization to ensure the integrity and quality of the data.

[0078] And convert the production sensor data into a unified format, such as CSV, Parquet, etc.; unify the timestamps of all data to the same frequency and starting point; perform dimensionless processing on data of different dimensions, such as Z-Score standardization; perform one-hot encoding or serial number encoding on categorical data (such as equipment names); and obtain complete sensor data.

[0079] Construct heat conduction models, fluid dynamics models and diffusion models based on complete production data; establish interfaces between heat conduction models, fluid dynamics models and diffusion models to ensure that data can be transferred between different models, and couple the heat conduction model, fluid dynamics model and diffusion model to obtain a digital twin model.

[0080] The heat conduction model is constructed in the following manner:

[0081] Define the control function and the calculation domain of the control function, the calculation domain includes the melting furnace, the pressing mold, the sintering furnace and the cooling device;

[0082] The control function includes the heat conduction equation, boundary conditions and initial conditions; the expression of the heat conduction equation is: , where is the density, is the specific heat capacity, is the temperature, For time, is the thermal conductivity coefficient, is the internal heat source (heat generated inside the material or system during heat conduction), is the gradient symbol.

[0083] The boundary conditions include known temperature conditions (temperature values ​​are specified directly on the boundaries), known heat flux conditions (heat flux density is specified on the boundaries), and known linear combinations of convection and radiation (heat exchange between the boundaries and the surrounding environment, including convection and radiation); the initial condition is a given initial temperature distribution.

[0084] Discretize the computational domain to generate computational grids; computational grids include structured grids and unstructured grids;

[0085] The discretization method includes:

[0086] According to the geometric shape of the computational domain, the computational domain is divided into structured grid and unstructured grid;

[0087] If the geometric shape of the computational domain is simple and regular (such as a cuboid, cylinder, etc.), the computational domain is divided into N5 sector-shaped (2D) or ring-shaped (2D) units in polar coordinates to form a structured grid;

[0088] If the computational domain has a complex geometry (e.g., concave corners, multiply connected domains, etc.), the computational domain is divided into N6 irregular triangular (2D) or tetrahedral (3D) elements, which constitute an unstructured grid.

[0089] Encrypt structured grids and unstructured grids. The encryption methods include:

[0090] The curvature distribution of the boundary of the computational domain is calculated, and the calculation methods include analytical method, geometric fitting method, discrete differential method, etc.; the boundary of the computational domain is the outer edge surrounding the entire computational area.

[0091] Define the curvature sensitive weight function , which is used to quantify the effect of curvature on mesh size; ways to define the curvature sensitive weight function include:

[0092] Get the curvature value of any node in structured and unstructured grids based on curvature distribution , if the curvature value of any node Greater than 0, the curvature sensitive weight function ;in, is the reference curvature, is the critical curvature value; the reference curvature is a benchmark curvature value used for normalization calculation, which can be the average curvature or maximum curvature in the calculation domain; the critical curvature value is a preset threshold used to determine whether the curvature is large enough to require special treatment.

[0093] If the curvature value of any node Less than or equal to 0, the curvature sensitive weight function ;in, and All are adjustable weight coefficients.

[0094] According to the curvature sensitive weight function, the grid size field is constructed ;in is the expected base grid size (determine the expected average cell size for the entire computational domain based on the computational accuracy requirements and hardware resource limitations), and They are the horizontal axis, vertical axis and vertical axis of the three-dimensional coordinate system at the grid size location.

[0095] Using the Grid Size field The structured grid and the unstructured grid are further divided to obtain the model processing grid.

[0096] Further divisions include:

[0097] Traverse all cells in the computational grid and, for each cell's centroid, apply the grid size field Calculate the expected size ;

[0098] like Less than , then the corresponding unit will be encrypted and subdivided; if Greater than The corresponding units are coarsened and merged.

[0099] The method of performing encryption segmentation includes:

[0100] Based on the expected size , calculate the minimum side length required for the cell ,in is a constant determined by optimization; check all edge lengths of the unit, if there is an edge length Greater than, then the length of the opposite side The corresponding edges are equally divided to generate new nodes; the original unit is re-divided using the new nodes to generate sub-units; specifically, a new node is generated on the long side of the original unit, and the original unit is divided into multiple sub-units, where the sub-unit is a triangle (2D) or a tetrahedron (3D), and the quality (such as shape, size, etc.) of the newly generated sub-unit is checked; if the quality does not meet the requirements, further optimization is performed, such as adjusting the node position or re-dividing; The value of depends on the shape of the original unit and the number of newly added nodes, usually 2-4 sub-units.

[0101] The generated sub-units are recursively subdivided until the side lengths of all sub-units are no greater than ;

[0102] The method of performing coarsening and merging includes:

[0103] Check all cells in the neighborhood of all faces of the cell (obtained by defining the neighborhood radius). If the difference in area of ​​any face of two cells is less than a preset difference threshold, they are merged to obtain a new merged cell.

[0104] Calculate the centroid size of the new unit after merging ,in is the area of ​​any face of the unit before merging, is the volume of the unit before merging; if Greater than , then continue to merge with other units in the neighborhood (the merging method is the same as above), otherwise stop.

[0105] Define a smoothing objective function, and after coarsening and merging, smooth all units by minimizing the smoothing objective function.

[0106] Smoothing objective function ;in For the The length of the strip, is the target side length, For the , The angle between the two sides, is the ideal angle, is the weight coefficient.

[0107] By using an optimization algorithm (such as the Gauss-Newton method) to solve the minimization smoothing objective function, the mesh edge length and unit shape are optimized to obtain the model processing mesh;

[0108] The control function is integrated on the basis of the model processing grid to obtain the heat conduction model.

[0109] It needs to be explained that the integrated control function refers to discretizing the continuous control function on each unit; applying the control function to each unit to form a series of algebraic equations; the specific steps are: applying the discretized heat conduction equation to each unit; applying the corresponding boundary conditions to the units at the boundary of the computational domain; assembling the equations of all units to form a global equation to obtain a heat conduction model.

[0110] Furthermore, the fluid dynamics model is constructed in the following manner:

[0111] On the basis of the model processing grid, the fluid dynamics control function is defined. The expression of the fluid dynamics control function is: ;in, is the specific heat capacity at constant pressure, For the The total derivative of the mass fraction of a substance, Thermal conductivity;

[0112] The fluid dynamics control function is integrated on the basis of the model processing grid to obtain the fluid dynamics model.

[0113] Furthermore, the diffusion model is constructed in the following manner:

[0114] Based on the model processing grid, the diffusion control function is defined. The expression of the diffusion control function is: ;in, It is The concentration of the element, It is The diffusion coefficient tensor of the element, It is The source term of the element; ;in, It is frequency factor of an element (it reflects the frequency of atomic jump attempts, is related to the lattice vibration frequency, and is obtained through experimental determination or theoretical calculation), It is The diffusion activation energy of an element (indicates the energy barrier that atoms need to cross to diffuse), is the universal gas constant (8.314 J / (mol·K)), is a concentration-dependent function, For the The concentration of other elements besides the is the stress-dependent function, is the stress tensor.

[0115] ,in, is the activation volume, is the Boltzmann constant;

[0116] Concentration-dependent function ,in, and is the fitting parameter, is the reference concentration.

[0117] The diffusion control function is discretized using a finite difference method, a finite element method or a spectral method to obtain a discretized diffusion function; the discretized diffusion function is integrated on the basis of a model processing grid to obtain a preliminary fluid dynamics model.

[0118] Diffusion boundary conditions are set on the basis of the preliminary fluid dynamics model. The diffusion boundary conditions include first-type boundary conditions, second-type boundary conditions and third-type boundary conditions. The diffusion boundary conditions are set in the same way as applying corresponding diffusion boundary conditions on the units at the boundary of the computational domain. The diffusion model is obtained.

[0119] The first type of boundary condition is the concentration value on the specified boundary. Its application scenarios are rapid surface equilibrium and constant sources, such as the contact between the material surface and the gas or liquid phase with a fixed concentration; constant sources, such as continuous surface penetration or coating process.

[0120] The second type of boundary condition is the concentration gradient or diffusion flux on the specified boundary. Its application scenarios are insulating boundaries and controlled surface reactions, such as surface oxidation or corrosion processes.

[0121] The third type of boundary condition is a linear combination of the concentration value and the diffusion flux on the boundary. Its application scenario is convection boundary conditions, for example, describing the interaction between the material surface and the flow environment, and interfacial mass transfer, such as the mass transfer process at the gas-solid interface or the liquid-solid interface.

[0122] Furthermore, the method for obtaining the change trend of the key production parameters includes:

[0123] Define key production parameters including temperature distribution, phase change rate and element distribution; initialize the parameters of the digital twin model, set the simulation time step and total simulation time; for each time step, input the actual production data at the current moment (the same data type as the production sensor data), perform a one-step simulation calculation, and store the simulation results (key production parameters); and use ensemble Kalman filtering to achieve data assimilation, regularly (such as every 10 time steps) integrate the actual observation data into the model, and adjust the parameters. After the total simulation time is over, extract the time series data consisting of the key production parameters from the stored simulation results, which is the change trend of the key production parameters.

[0124] Furthermore, the calculation method of the sensitivity coefficient matrix includes:

[0125] The output indicators of key production parameters are defined as product performance indicators, production efficiency indicators, energy consumption indicators and raw material utilization indicators.

[0126] The data points on the time series data are taken as sample points, and the sensitivity objective function is defined; the expression of the sensitivity function is: ;in, For sample points The corresponding vector of key production parameters, To add to the sample point A sufficiently small perturbation of is the objective function; is a small perturbation function, is the difference function; For sample points The corresponding sensitivity coefficient.

[0127] Small perturbation function ;

[0128] Difference function ;in, is the total number of sample points, For sample points The coefficients of the corresponding wavelet basis functions are, For sample points The corresponding disturbance point is Satisfy a certain specific wavelet basis function, such as Daubechies wavelet basis; For the sample point , the constructed random perturbation sequence, Obey some probability distribution, such as uniform distribution or Gaussian distribution.

[0129] The objective function is the weighted sum of the performance index function, production efficiency function, energy consumption function and raw material utilization function; the performance index function is defined as the weighted sum of remanence, coercivity and energy product; the production efficiency function is defined as the product of production capacity and yield rate; the energy consumption function is the sum of the energy consumption of smelting, pressing and sintering; the raw material utilization function is defined as the ratio of actual output to theoretical output.

[0130] Initialize a symmetric matrix and transform the sample points And the corresponding sensitivity coefficient As the elements of the symmetric matrix, the calculation of the sensitivity coefficient matrix is ​​completed.

[0131] Furthermore, the multi-objective optimization function ;in, is a vector of control parameters to be optimized; is the process constraint matrix, is the expected product performance target vector, is the sensitivity coefficient matrix, is the regularization parameter.

[0132] It should be noted that the control parameters to be optimized include melting temperature, melting time, pressing pressure, holding time, sintering temperature, sintering time, heat treatment temperature, heat treatment time and cooling rate.

[0133] The process constraint matrix represents the relationship between control parameters and process variables. A mathematical model is established based on physical and chemical principles to obtain the relationship equation between control parameters and process variables. The mapping relationship between control parameters and process variables is established using historical production data and machine learning algorithms (such as multivariate linear regression, neural networks, etc.). Through orthogonal experiments or response surface methodology, the influence coefficients of control parameters on process variables are obtained, thus completing the process constraint matrix.

[0134] The expected product performance target vector includes the product performance indicators that are expected to be achieved, and determines the key performance indicators including remanence, coercive force and maximum energy product; sets a target value for each product performance indicator, normalizes different product performance indicators, and assigns weights to each product performance indicator according to its importance, and obtains the expected product performance target vector through weighted calculation.

[0135] Furthermore, the method for obtaining the optimal control parameter combination includes:

[0136] Initialize the particle swarm and define the number of particles , maximum number of iterations , randomly generated The position of the particle and speed , is the index of the particle; initialize the individual optimal position of each particle And the global optimal position ; It needs to be explained that each particle represents a set of possible control parameter combinations; for example, a particle contains specific values ​​of melting temperature, pressing pressure, and sintering time; the particle's is the coordinate of the solution it represents in the search space; each dimension corresponds to a control parameter; for example, there are 5 control parameters, then the position of each particle is a 5-dimensional vector.

[0137] The speed of a particle indicates the direction and size of its movement in the search space, which determines how the particle updates its position. For each particle, the individual optimal position is the position of the best solution found by the corresponding particle so far. The global optimal position is the position of the best solution found by all particles in the entire particle swarm. That is, the individual optimal position is the best production result produced historically by each specific parameter combination. The global optimal position is the best production result produced by all tried parameter combinations.

[0138] The multi-objective optimization function is used as the fitness function, and the speed update formula is defined as follows: ;in, is the inertia weight, and is the acceleration factor; and for A random number between For time The position of a particle is randomly selected when For time Time Particle speed, For time Time Particle speed, For time Time Particle The individual optimal position of For time Time Particle The global optimal position of For time Time Particle location.

[0139] Define the position update formula: ;in, For time Time Particle location;

[0140] Define the search space, if If it exceeds the preset search space, it will be pulled back to the boundary of the search space, the value of the fitness function of each particle's new position will be calculated, and the individual optimal position and the global optimal position of each particle will be compared and updated.

[0141] If the preset maximum number of iterations is reached, the process ends; otherwise, the iteration continues and the optimal solution is output. The global optimal position corresponding to the optimal solution is the optimal control parameter combination.

[0142] In this embodiment, by constructing an accurate digital twin model, comprehensive perception and in-depth insight into the complex production process is achieved, which greatly improves the accuracy and real-time performance of production control; by utilizing multi-source heterogeneous data and simulation, it can accurately predict the changing trends of key production parameters, providing a reliable basis for timely adjustment of control strategies; the introduced sensitivity analysis method effectively identifies the most critical control parameters and optimizes the control focus; the multi-objective optimization function and the improved particle swarm algorithm achieve a comprehensive balance in product performance, production efficiency, energy consumption and other aspects, significantly improving production efficiency and product quality; the low-latency architecture ensures that the system responds quickly to production fluctuations and effectively reduces the generation of defective products; in addition, it also reduces energy consumption and raw material loss, improves resource utilization, and promotes green production. Example

[0143] See also Figure 2As shown, the part not described in detail in this embodiment is described in Example 1, and a low-delay control method for rare earth permanent magnet material production based on digital twin technology is provided, including:

[0144] S1. Obtain production sensor data of rare earth permanent magnet material production line;

[0145] S2. Use production sensor data to build a digital twin model of the rare earth permanent magnet material production process;

[0146] S3. Use the digital twin model to simulate and obtain the changing trend of key production parameters;

[0147] S4. Calculate the sensitivity coefficient matrix of each key production parameter according to the change trend of the key production parameters;

[0148] S5. Construct a multi-objective optimization function based on the sensitivity coefficient matrix, use the improved particle swarm algorithm to solve the multi-objective optimization function, and obtain the optimal control parameter combination. Example

[0149] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the low-latency control method for the production of rare earth permanent magnet materials based on digital twin technology provided above is implemented.

[0150] Since the electronic device introduced in this embodiment is an electronic device used to implement the low-latency control method for producing rare earth permanent magnet materials based on digital twin technology in the embodiment of this application, based on the low-latency control method for producing rare earth permanent magnet materials based on digital twin technology introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation method of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not introduced in detail here. As long as the technical personnel of this field implement the electronic device used by the low-latency control method for producing rare earth permanent magnet materials based on digital twin technology in the embodiment of this application, it belongs to the scope of protection of this application.

[0151] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0152] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. Low-delay control system for rare earth permanent magnet material production based on digital twin technology, characterized by: include: Data acquisition module, used to obtain production sensor data of rare earth permanent magnet material production line; The model building module uses production sensor data to build a digital twin model of the rare earth permanent magnet material production process; The simulation module uses the digital twin model to simulate and obtain the changing trends of key production parameters; Matrix fitting module, used to calculate the sensitivity coefficient matrix of each key production parameter according to the change trend of key production parameters; The parameter optimization module constructs a multi-objective optimization function based on the sensitivity coefficient matrix, and uses the improved particle swarm algorithm to solve the multi-objective optimization function to obtain the optimal control parameter combination; each module is connected by wired and / or wireless means.

2. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 1 is characterized in that: The production sensing data includes temperature data, pressure data, flow data, electrical parameters, material property data, fluid related data and diffusion related data; Temperature data includes melting furnace temperature, sintering furnace temperature, mold temperature and cooling temperature; pressure data includes press pressure, air pressure and hydraulic pressure; flow data includes raw material input flow, coolant flow and protective gas flow; Electrical parameters include motor current, heating power and control system voltage; material property data include material density, material specific heat capacity and material thermal conductivity; fluid-related data include fluid viscosity and fluid density; diffusion-related data include diffusion coefficient and concentration distribution of elements.

3. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 2 is characterized in that: The digital twin model is constructed in the following ways: The production sensor data is cleaned, denoised, interpolated and standardized, and the preprocessed production sensor data is converted into a unified format; the timestamps of all data are unified to the same frequency and starting point; and the data of different dimensions are dimensionless; the categorical data (one-hot encoding or serial number encoding) is converted into perfect sensor data; Construct heat conduction model, fluid dynamics model and diffusion model based on complete production data; establish interfaces between heat conduction model, fluid dynamics model and diffusion model, couple the heat conduction model, fluid dynamics model and diffusion model to obtain a digital twin model.

4. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 3 is characterized in that: The heat conduction model is constructed in the following manner: Define the control function and the calculation domain of the control function, the calculation domain includes the melting furnace, the pressing mold, the sintering furnace and the cooling device; The control function includes the heat conduction equation, boundary conditions and initial conditions; the expression of the heat conduction equation is: , where is the density, is the specific heat capacity, is the temperature, For time, is the thermal conductivity coefficient, As internal heat source, is the gradient symbol; The boundary conditions include known temperature conditions, known heat flow conditions, and a linear combination of known convection and radiation; the initial conditions are a given initial temperature distribution; The computational domain is discretized to generate a computational grid; the computational grid includes a structured grid and an unstructured grid; the structured grid and the unstructured grid are encrypted to obtain a model processing grid; the control function is integrated on the basis of the model processing grid to obtain a heat conduction model.

5. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 4 is characterized in that: The encryption method includes: Calculate the curvature distribution of the boundary of the computational domain and define the curvature sensitive weight function ; Ways to define a curvature-sensitive weight function include: Get the curvature value of any node in structured and unstructured grids based on curvature distribution , if the curvature value of any node Greater than 0, the curvature sensitive weight function ;in, is the reference curvature, is the critical curvature value; If the curvature value of any node Less than or equal to 0, the curvature sensitive weight function ;in, and All are adjustable weight coefficients; According to the curvature sensitive weight function, the grid size field is constructed in is the desired base grid size, and They are the horizontal axis, vertical axis and vertical axis of the three-dimensional coordinate system at the grid size location; Using the Grid Size field The structured grid and the unstructured grid are further divided to obtain the model processing grid.

6. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 5 is characterized in that: The further division method includes: Traverse all cells in the computational grid and, for each cell's centroid, apply the grid size field Calculate the expected size ; like Less than , then the corresponding unit will be encrypted and subdivided; if Greater than , the corresponding units are coarsened and merged; The method of performing encryption segmentation includes: Based on the expected size , calculate the minimum side length required for the cell ,in is a constant determined by optimization; check all edge lengths of the unit, if there is an edge length Greater than , then the side length The corresponding edges are equally divided to generate new nodes; the original unit is re-divided using the new nodes to generate sub-units; the generated sub-units are recursively subdivided until the side lengths of all sub-units are no greater than ; The method of performing coarsening and merging includes: Check all the cells in the neighborhood of the cell, if there are two cells whose area difference of any surface is less than the preset difference threshold, merge them to obtain a new cell after merging; Calculate the centroid size of the new unit after merging ,in is the area of ​​any face of the unit before merging, is the volume of the unit before merging; if Greater than , then continue to merge with other units in the neighborhood, otherwise stop; Define a smoothing objective function, and after coarsening and merging, smooth all units by minimizing the smoothing objective function; Smoothing objective function ;in For the The length of the strip, is the target side length, For the , The angle between the two sides, is the ideal angle, is the weight coefficient; The model processing grid is obtained by solving the minimization smoothing objective function through the optimization algorithm.

7. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 6 is characterized in that: The fluid dynamics model is constructed in the following manner: On the basis of the model processing grid, the fluid dynamics control function is defined. The expression of the fluid dynamics control function is: ;in, For the The mass fraction of a substance, For the The diffusion coefficient of a substance, It is The migration coefficient of a substance, For the The coupling weight factor for the material application, For the The source term of a substance, is the energy function; Energy function ;in, is the specific heat capacity at constant pressure, For the The total derivative of the mass fraction of a substance, Thermal conductivity; The fluid dynamics control function is integrated on the basis of the model processing grid to obtain the fluid dynamics model.

8. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 7, characterized in that: The diffusion model is constructed in the following manner: Based on the model processing grid, the diffusion control function is defined. The expression of the diffusion control function is: ;in, It is The concentration of the element, It is The diffusion coefficient tensor of the element, It is The source term of the element; ;in, It is The frequency factor of the element, It is The diffusion activation energy of the element, is the universal gas constant, is a concentration-dependent function, For the The concentration of other elements besides the is the stress-dependent function, is the stress tensor; ,in, is the activation volume, is the Boltzmann constant; Concentration-dependent function ,in, and is the fitting parameter, is the reference concentration; The diffusion control function is discretized to obtain a discretized diffusion function; the discretized diffusion function is integrated on the basis of the model processing grid to obtain a preliminary fluid dynamics model; On the basis of the preliminary fluid dynamics model, diffusion boundary conditions are set to obtain a diffusion model; the diffusion boundary conditions include first-class boundary conditions, second-class boundary conditions and third-class boundary conditions; The first type of boundary condition is the concentration value on the specified boundary, the second type of boundary condition is the concentration gradient or diffusion flux on the specified boundary; the third type of boundary condition is a linear combination of the concentration value and diffusion flux on the boundary.

9. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 8, characterized in that: The calculation method of the sensitivity coefficient matrix includes: The output indicators of key production parameters are defined as product performance indicators, production efficiency indicators, energy consumption indicators and raw material utilization indicators; The data points on the time series data are taken as sample points, and the sensitivity objective function is defined; the expression of the sensitivity function is: ;in, For sample points The corresponding vector of key production parameters, To add to the sample point A disturbance of is the objective function; is a small perturbation function, is the difference function; For sample points The corresponding sensitivity coefficient; Small perturbation function Difference function ;in, is the total number of sample points, For sample points The coefficients of the corresponding wavelet basis functions are, For sample points The corresponding disturbance point is For the sample point , the constructed random perturbation sequence; The objective function is the weighted sum of the performance index function, production efficiency function, energy consumption function and raw material utilization function; the performance index function is defined as the weighted sum of the remanence, coercivity and energy product; the production efficiency function is defined as the product of production capacity and yield rate; the energy consumption function is the sum of the energy consumption of smelting, pressing and sintering; the raw material utilization function is defined as the ratio of actual output to theoretical output; Initialize a symmetric matrix and transform the sample points And the corresponding sensitivity coefficient As the elements of the symmetric matrix, the calculation of the sensitivity coefficient matrix is ​​completed; The multi-objective optimization function ;in, is a vector of control parameters to be optimized; is the process constraint matrix, is the expected product performance target vector, is the sensitivity coefficient matrix, is the regularization parameter.

10. The low-delay control system for rare earth permanent magnet material production based on digital twin technology according to claim 9, characterized in that: The method for obtaining the optimal control parameter combination includes: Initialize the particle swarm and define the number of particles , maximum number of iterations , randomly generated The position of the particle and speed , is the index of the particle; initialize the individual optimal position of each particle And the global optimal position ; The multi-objective optimization function is used as the fitness function, and the speed update formula is defined as follows: ;in, is the inertia weight, and is the acceleration factor; and for A random number between For time The position of a particle is randomly selected when For time Time Particle speed, For time Time Particle speed, For time Time Particle The individual optimal position of For time Time Particle The global optimal position of For time Time Particle location; Define the position update formula: ;in, For time Time Particle location; Define the search space, if If it exceeds the predefined search space, it will be pulled back to the boundary of the search space, the value of the fitness function of each particle's new position will be calculated, and the individual optimal position and the global optimal position of each particle will be compared and updated; If the preset maximum number of iterations is reached, the process ends; otherwise, the iteration continues and the optimal solution is output. The global optimal position corresponding to the optimal solution is the optimal control parameter combination.

Images