A knowledge graph-based magnetic yoke lamination process parameter simulation optimization method and system

Abstract

This invention discloses a method and system for simulating and optimizing process parameters of magnetic yoke stacking based on knowledge graphs, belonging to the field of process simulation technology. The method includes: constructing a magnetic yoke stacking database to store relevant data for magnetic yoke stacking simulation; constructing a knowledge graph library based on the data in the magnetic yoke stacking database; constructing an analysis and optimization algorithm through tensor analysis and multiphysics coupling, and combining the magnetic yoke stacking database and the knowledge graph library to provide a preliminary analysis framework for simulation design. The analysis and optimization algorithm includes a tensor analysis-multiphysics coupling mathematical model and a knowledge graph collaborative mechanism; using the data embedded in the knowledge tensor as input, performing finite element simulation, and updating the knowledge embedded tensor based on the new simulation results to form new knowledge embedded tensor data. This invention improves design quality and efficiency by applying expert experience and knowledge.

Description

Technical Field

[0001] This invention relates to the field of process simulation technology, and in particular to a method and system for simulating and optimizing process parameters of magnetic yoke stacking based on knowledge graphs. Background Technology

[0002] With the rapid development of the manufacturing industry, the optimization design of process parameters for magnetic yoke laminations, as key components, has received increasing attention. Traditional magnetic yoke lamination process parameter design mainly relies on designers' experience and trial-and-error methods, lacking scientific prediction and optimization tools. Although simulations can be performed using the finite element method or finite difference method, these methods suffer from low computational efficiency and limited accuracy, making it difficult to meet the needs of rapid iteration and optimization of complex process parameters. In existing technologies, the design of magnetic yoke lamination process parameters mainly suffers from the following problems: First, existing numerical simulation methods struggle to achieve real-time calculations of complex multiphysics fields, failing to accurately describe the mechanical behavior of magnetic yoke laminations; second, there is a lack of multi-process collaborative simulation capabilities, with each process simulation operating independently, making it difficult to reflect the cumulative effects between processes; third, the simulation model is disconnected from the actual process, unable to dynamically adjust model parameters based on deviations in actual production, resulting in insufficient prediction accuracy; fourth, there is a lack of systematic expression and utilization mechanisms for process knowledge, making it difficult to effectively inherit and apply a large amount of valuable experience; finally, existing optimization methods often target only a single process or a single objective, failing to achieve collaborative optimization of multiple objectives across the entire process. Summary of the Invention

[0003] One of the objectives of this invention is to provide a knowledge graph-based simulation optimization method for magnetic yoke stacking process parameters, in order to solve the problems of low simulation efficiency, limited accuracy, and inability to meet the needs of rapid iteration and optimization of complex process parameters in the existing technology.

[0004] This invention is achieved through the following technical solution: a simulation optimization method for magnetic yoke stacking process parameters based on knowledge graphs, comprising the following steps: S100, constructing a magnetic yoke stacking database to store relevant data for magnetic yoke stacking simulation; S200, constructing a knowledge graph library based on the data in the magnetic yoke stacking database; S300, constructing an analysis and optimization algorithm through tensor analysis and multiphysics coupling, and combining the magnetic yoke stacking database and the knowledge graph library to provide a preliminary analysis framework for simulation design. The analysis and optimization algorithm includes a tensor analysis-multiphysics coupling mathematical model and a knowledge graph collaborative mechanism. The tensor analysis and multiphysics coupling mathematical model uses high-dimensional tensors to uniformly represent multiphysics parameters. The numbers and their interactions are used to embed different physical field parameters in the simulation process of magnetic yoke stacking into each element of the tensor to construct a multi-physics tensor. By fusing the parameters of different physical fields at the tensor level, the problem in the simulation process is decomposed into different physical scales. The knowledge graph collaboration mechanism obtains a knowledge embedding tensor with embedded knowledge graph through tensor analysis and multi-physics coupling mathematical model. Data mining and analysis are performed on the knowledge embedding tensor to identify higher-order correlations and patterns. S400: The data in the knowledge embedding tensor is used as input to perform finite element simulation. Based on the new simulation results, the knowledge embedding tensor is updated to form new knowledge embedding tensor data.

[0005] Furthermore, the magnetic yoke stack database includes: design parameters, material properties, process constraints, empirical data, and standard specifications of the magnetic yoke stack; the data in the database is cleaned and parsed to construct the magnetic yoke stack database.

[0006] Furthermore, the construction of the knowledge graph library includes: transforming the data in the magnetic yoke stack database into entities, relationships, and attributes that the knowledge graph can understand; for structured data, directly mapping the structured data to entities and attributes defined in the ontology; and for unstructured data, identifying key entities in the text through named entities, extracting semantic relationships between entities through relation extraction, and identifying event information related to process flow, fault analysis, etc. through event extraction; importing the extracted and semantically represented knowledge into the knowledge graph library in batch processing or streaming mode, and performing consistency checks and data quality verification on the imported knowledge graph data to ensure the accuracy of the knowledge.

[0007] Furthermore, the tensor analysis-multiphysics coupling mathematical model is constructed through the following steps: S310, selecting the physical fields required in the magnetic yoke stacking simulation process, embedding the core parameters of the selected physical fields and their potential interactions into the data structure to construct the core multiphysics tensor; S320, transforming the various parameter relationships in the multiphysics tensor into operable physical field interaction operators, which include self-coupling operators and cross-coupling operators, integrating the self-coupling operators and cross-coupling operators into a matrix form, using the self-coupling operators as diagonal operators and the cross-coupling operators as off-diagonal operators, constructing the total coupling operator matrix, and combining the state vectors and external source terms of each physical field to construct a unified coupling equation; S330, decomposing the problems in the magnetic yoke stacking simulation process into different physical scales, and establishing descriptive models for different scales; S340, embedding the process parameters of the magnetic yoke stacking as constraints into the space of the multiphysics tensor to obtain the knowledge embedding tensor.

[0008] Furthermore, S310 also includes: determining the dimensions of the multiphysics tensor based on the number of selected physical fields, with each dimension corresponding to the parameter space of a physical field, and the size of the dimension depending on the degree or range of parameter discretization, thereby constructing the tensor dimensions and index space of the multiphysics tensor, and finally filling the tensor dimensions and index space of the constructed multiphysics tensor with the data stored in the database.

[0009] Furthermore, the self-coupling operator is constructed by extracting the evolution or equilibrium mechanism of each physical field and is used to describe the physical laws within that physical field; the cross-coupling operator is constructed based on the interaction mechanism between different physical fields and is used to describe the mutual response between different physical fields.

[0010] Furthermore, the unified coupling equation can be expressed by the following equation:

[0011] ,

[0012] in, The response to the electromagnetic field represents the evolution of the electromagnetic field, changes in current density and magnetic flux, etc. The response to the mechanical field represents the stress, strain, and rate of change of displacement inside the structure, or the external force / torque, representing the result of the force acting on the mechanical system. The response of the fluid field represents the response of momentum change rate, pressure gradient, and mass flux in the flow. Here, represents the input or state variable of the electromagnetic field. Dimensions in knowledge embedding tensors The parameters in the table are directly corresponding and represent the current electromagnetic parameter state of the system. The input or state variables of the mechanical field, and the dimension in the knowledge embedding tensor. The parameters in the table are directly corresponding and represent the current mechanical parameter state of the system. For the input or state variables of the fluid field, the dimension is related to the knowledge embedding tensor. The parameters in the table are directly corresponding and represent the fluid parameter state of the current system. It is a self-coupling operator for electromagnetic fields; For self-coupling operators of mechanical fields; These are self-coupling operators for fluid fields. These self-coupling operators describe the physical laws and evolution within each physical field itself, without considering direct interactions with other physical fields. The self-coupling operators for electromagnetic fields can be discretized forms of Maxwell's equations, describing how the electromagnetic field responds to its own input. Similarly, the self-coupling operators for mechanical fields can be equations of elasticity or plasticity, describing the stress-strain relationship within the mechanical field. The self-coupling operators for fluid fields can be the Navier-Stokes equations, describing the flow laws of the fluid itself. The deviation vector represents an external source term, boundary condition, or uncoupled driving term. For example, in an electromagnetic field, the deviation vector can be an external current source or voltage source; in a mechanical field, it can be an external load, gravity, or initial stress; in a fluid field, it can be an external pressure difference, mass source, or body force. This deviation vector is the part of the equation that is independent of the input response and is used to introduce the external influence on the system. For mechanical-electromagnetic cross-coupling operators, It is an electromagnetic-fluid cross-coupling operator. It is an electromagnetic-mechanical cross-coupling operator. For mechanical-fluid cross-coupling operators, For fluid-electromagnetic cross-coupling operators, It is a fluid-mechanical cross-coupling operator.

[0013] Furthermore, the different scales in S330 include: dividing the physical scale into microscale, mesoscale, and macroscale, where the microscale focuses on the constitutive relations of the magnetic yoke stack material, the mesoscale focuses on the macroscopic effects of the structural units composed of micromaterials, and the macroscale focuses on the behavior of the magnetic yoke stack as a whole; by establishing the connection rules between different scales, and ensuring that different scales maintain physical consistency in boundary conditions, conservation laws, and coupling interfaces.

[0014] Furthermore, the knowledge graph collaboration mechanism also includes: identifying and mining high-order associations and patterns embedded in knowledge tensors, transforming high-order associations and patterns into logical reasoning rules of IF-THEN form, and re-injecting the extracted reasoning rules as ontology or axioms into the knowledge graph library. The reasoning engine of the knowledge graph library then uses the injected reasoning rules to discover or predict new knowledge.

[0015] Furthermore, S400 also includes: calculating the difference between the new knowledge embedding tensor data and the predicted knowledge embedding tensor data of the current model, and iteratively updating and correcting the data in the knowledge embedding tensor based on the calculated error and in conjunction with the update rules.

[0016] Furthermore, the updated knowledge embedding tensor can be represented by the following formula:

[0017] ,in, The update quantity for the knowledge embedding tensor. The learning rate is a positive number less than 1, used to control the step size of each update, preventing the model from overfitting the new data and maintaining stability. These are the tensor eigenvalues ​​obtained from the finite element method. The preceding model is the tensor feature value of the tensor prediction embedded with existing knowledge; This is the confidence matrix. By introducing the confidence matrix, the model can selectively trust new data. For data with high confidence, the correction will be smaller; for data with low confidence, the model will make corrections based on it to a greater extent, thereby avoiding contamination of the model by erroneous or noisy data.

[0018] Another aspect of the present invention provides a knowledge graph-based simulation optimization system for magnetic yoke lamination process parameters, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the knowledge graph-based simulation optimization method for magnetic yoke lamination process parameters as described above.

[0019] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0020] 1. This invention collects data related to magnetic yoke lamination simulation, constructs a magnetic yoke lamination database, and transforms the database data into entities, relationships, and attributes that can be understood by a knowledge graph. This enables the systematic expression and utilization of magnetic yoke lamination process knowledge, and improves design quality and efficiency by applying expert experience and knowledge.

[0021] 2. This invention constructs an analysis and optimization algorithm based on tensor analysis and multiphysics coupling, realizing real-time calculation of complex multiphysics fields, accurately describing the electromagnetic and mechanical behavior of magnetic yoke laminations, and using a multi-scale hierarchical modeling approach to consider the physical laws of multiple scales, realizing collaborative simulation of the entire process, effectively reflecting the cumulative effect between processes, meeting the needs of rapid iteration and optimization of complex process parameters, solving the problem of the disconnect between simulation models and actual processes in existing technologies, and improving prediction accuracy.

[0022] 3. This invention directly links simulation results with process parameters through knowledge embedding tensors, enabling the model to optimize and make decisions at the process level. This overcomes the limitations of existing technologies that only optimize a single process or single objective, achieving collaborative optimization of multiple objectives across the entire process. At the same time, by incorporating a knowledge graph collaboration mechanism, high-order associations and patterns are transformed into logical reasoning rules, realizing the explicitness and reasoning of knowledge. This provides a preliminary analysis framework for simulation design, significantly improving the scientificity and accuracy of magnetic yoke lamination process parameter design. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0024] Figure 1 This is a flowchart of the method provided in Embodiment 1 of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] Example 1

[0027] Existing technologies for designing magnetic yoke lamination process parameters primarily rely on designers' experience and trial-and-error methods, severely lacking scientific prediction and optimization techniques. This results in low design efficiency and difficulty in ensuring design quality. Although finite element analysis software now exists to simulate and calculate magnetic yoke lamination process parameters, existing numerical simulation methods struggle to perform real-time calculations of complex multiphysics fields, failing to accurately describe the mechanical behavior of the magnetic yoke laminations. Furthermore, due to the numerous process parameters and complex manufacturing steps involved in magnetic yoke lamination, there is a lack of full-process simulation capabilities that allow for coordinated process steps, making it difficult to reflect the cumulative effects between steps. To address these issues, this embodiment discloses a knowledge graph-based simulation optimization method for magnetic yoke lamination process parameters. This method utilizes knowledge graphs to make knowledge explicit, facilitate reasoning, and continuously optimize the model, providing a preliminary analytical framework for simulation design and improving the efficiency and accuracy of magnetic yoke lamination process parameter design.

[0028] Figure 1 The flowchart of the method in this embodiment is shown. As can be seen from the flowchart, this embodiment includes the following steps:

[0029] Step 1: First, collect relevant data on the design, materials, processes, and simulations of the target magnetic yoke stack. Clean and analyze the collected structured and unstructured data to construct a magnetic yoke stack database.

[0030] Specifically, the data in this database includes: design parameters: lamination shape, lamination coefficient, hot clamping amount, etc.; material properties: silicon steel sheet grade, yield strength, magnetic flux density, resistivity, thermal conductivity, etc.; process constraints: waviness, pressing force, annealing temperature, cooling rate, etc.; experience data: engineers' experience knowledge, historical production data, failure cases, etc.; standards and specifications: industry standards, national standards, etc.

[0031] Step 2: Transform the data in the magnetic yoke stack database into entities, relations, and attributes that the knowledge graph can understand, and transform the preprocessed data into triples (entity-relationship-entity / attribute-value) in the knowledge graph.

[0032] Specifically, for structured data, database tables, CSV files, and other structured data can be directly mapped to entities and attributes defined in the ontology; while for unstructured data, Named Entity Recognition (NER) can be used to identify key entities in the text (e.g., silicon steel sheet type, waviness, hot-fitting amount); then, Relation Extraction (RE) can be used to identify semantic relationships between entities (e.g., the attribute relationship between yield strength and material); and event extraction can be used to identify event information related to process flow, fault analysis, etc. Empirical knowledge is digitized by making explicit empirical rules and tacit knowledge obtained from expert interviews graph nodes or rules.

[0033] Finally, the extracted and semantically encoded knowledge is imported into the knowledge graph database in batch or streaming mode. Consistency checks and data quality verification are performed on the imported knowledge graph data to ensure the accuracy of the knowledge.

[0034] Step 3: An analysis and optimization algorithm is constructed through tensor analysis and multiphysics coupling. Combined with the magnetic yoke stack database and knowledge graph library, the algorithm realizes the explicitness of knowledge, reasoning, and continuous optimization of the model, providing a preliminary analysis framework for simulation design.

[0035] By accurately describing physical laws through tensor analysis and multiphysics coupling, the coupling behavior between physical laws can be predicted. Combined with discrete knowledge from a knowledge graph, relationships between entities are revealed and logical reasoning is performed. This enables: knowledge extracted from simulation data and expert experience to be encoded into a knowledge graph; new patterns or predictions can be discovered through the graph's reasoning capabilities; the knowledge graph provides prior knowledge, boundary conditions, or parameter ranges for mathematical models; and simulation results can, in turn, update the knowledge graph, forming a closed-loop optimization feedback mechanism. Combining model prediction and knowledge reasoning provides more intelligent and comprehensive decision support for complex engineering simulations.

[0036] The analysis and optimization algorithms include tensor analysis-multiphysics coupled mathematical models and knowledge graph collaborative mechanisms.

[0037] The construction logic of tensor analysis and multiphysics coupling mathematical models can be summarized as follows:

[0038] By unifying the representation of multiphysics parameters and their interactions using a high-dimensional tensor, the computational process is simplified. By embedding domain knowledge (i.e., the interdependence of different physics parameters) and computational / experimental results (system state characteristics under specific parameter combinations) into each element of the tensor, the tensor itself becomes a knowledge base containing a wealth of information. By fusing parameters of different physics at the tensor level, the foundation is laid for analyzing and optimizing coupling behavior at different scales (e.g., macroscopic energy loss and microscopic material structure). This tensor itself can be data-driven (filled with extensive simulation or experimental data) or model-driven (estimated using simplified analytical models or semi-empirical formulas). This hybrid approach allows the tensor-based mathematical model to utilize both large datasets and incorporate physical laws.

[0039] Specifically, in this embodiment, the tensor analysis-multiphysics coupling mathematical model can be constructed through the following steps:

[0040] 1) First, the core multiphysics tensor is defined, which is the foundation of tensor analysis and multiphysics coupling mathematical models. The multiphysics tensor is constructed by embedding the core parameters of multiple physics used in the simulation and their potential interactions into a data structure in a structured manner.

[0041] Specifically, in this embodiment, the multiphysics tensor can identify all relevant physical fields (e.g., electromagnetism, mechanics, fluid dynamics, thermodynamics, etc.) based on the application goals of the magnetic yoke stack. Furthermore, it defines the most representative and influential core parameters for each physical field.

[0042] For example, the core parameters of electromagnetism can include parameters such as permeability, electrical conductivity, and magnetic induction; the core parameters of mechanics can include parameters such as Young's modulus, Poisson's ratio, and coefficient of thermal expansion; and the core parameters of a volume can include parameters such as viscosity, density, and flow velocity.

[0043] Based on the selected core physics field, the dimensions of the multiphysics tensor are determined. Each dimension corresponds to the parameter space of a physics field, and the size of the dimension depends on the degree or range of discretization of these parameters, thus constructing the tensor dimensions and index space of the multiphysics tensor. Then, the data stored in the database is used to populate the constructed tensor dimensions and index space of the multiphysics tensor.

[0044] For example, in this embodiment, considering that the core physical fields in the simulation of magnetic yoke stacks typically include: deformation caused by electromagnetic force, changes in electromagnetic properties by mechanical forces, and influence on fluid flow, the electromagnetic field, mechanical field, fluid field, and their main coupling relationships are analyzed. A dimension is defined for each core physical field, incorporating all key, quantifiable parameters of that physical field into that dimension. Therefore, a three-dimensional physical field tensor is constructed, which can be expressed by the following equation:

[0045] ,

[0046] in, For multiphysics tensors, For electromagnetic parameter space, For mechanical parameter space, For fluid parameter space, It is a three-dimensional tensor space composed of electromagnetic parameter space, mechanical parameter space and fluid parameter space.

[0047] 2) Transform the various parameter relationships of physical fields in the multiphysics tensor into operable physical field interaction operators. For each major physical field in the multiphysics tensor, first determine its fundamental governing equations on a macroscopic scale (e.g., Maxwell's equations in electromagnetism, equilibrium equations in solid mechanics, and Navier-Stokes equations in fluid mechanics). Extract the evolution or equilibrium mechanism of each physical field as its self-coupling operator terms. These operators are mainly used to describe the physical laws within the physical field.

[0048] Based on the constructed operator terms, and according to the interaction mechanisms between different physical fields (i.e., the coupling relationships between different physical fields), these interaction mechanisms are mathematically represented as cross-coupling operator terms.

[0049] Specifically, cross-coupling operators can be constructed based on known physical laws (such as Lorentz force, Joule heating, thermal expansion, fluid-structure interaction, etc.). If the physical mechanism to be simulated is complex or difficult to model accurately, then data embedded in the knowledge tensor can be used for fitting or parameterization to obtain cross-coupling operators. For example, by machine learning the data patterns in multiphysics tensors, we can discover how electromagnetic effects strongly influence mechanical deformation under specific parameter combinations, thereby obtaining cross-coupling operators to describe the mutual responses between different physical fields.

[0050] Finally, the self-coupling operator terms and cross-coupling operator terms are integrated into a matrix form. In the matrix form, the self-coupling operators are treated as operators on the diagonal, while the cross-coupling operators are treated as operators off-diagonal, thus forming the final total coupling operator matrix. By combining the state vectors of each physical field and the external source terms, a unified coupling equation is constructed.

[0051] For example, in this embodiment, the core physical fields in the simulation of magnetic yoke stacks are electromagnetic field, thermal field, and fluid field. These typically include: electromagnetic force causing deformation, heat changing electromagnetic properties, and affecting fluid flow. By analyzing these fields and their main coupling relationships, a unified coupling equation can be constructed as shown in the following equation:

[0052] ,

[0053] in, The response to the electromagnetic field represents the evolution of the electromagnetic field, changes in current density and magnetic flux, etc. The response to the mechanical field represents the stress, strain, and rate of change of displacement inside the structure, or the external force / torque, representing the result of the force acting on the mechanical system. The response of the fluid field represents the response of momentum change rate, pressure gradient, and mass flux in the flow. Here, represents the input or state variable of the electromagnetic field. Dimensions in knowledge embedding tensors The parameters in the table are directly corresponding and represent the current electromagnetic parameter state of the system. The input or state variables of the mechanical field, and the dimension in the knowledge embedding tensor. The parameters in the table are directly corresponding and represent the current mechanical parameter state of the system. For the input or state variables of the fluid field, the dimension is related to the knowledge embedding tensor. The parameters in the table are directly corresponding and represent the fluid parameter state of the current system. It is a self-coupling operator for electromagnetic fields; For self-coupling operators of mechanical fields; These are self-coupling operators for fluid fields. These self-coupling operators describe the physical laws and evolution within each physical field itself, without considering direct interactions with other physical fields. The self-coupling operators for electromagnetic fields can be discretized forms of Maxwell's equations, describing how the electromagnetic field responds to its own input. Similarly, the self-coupling operators for mechanical fields can be equations of elasticity or plasticity, describing the stress-strain relationship within the mechanical field. The self-coupling operators for fluid fields can be the Navier-Stokes equations, describing the flow laws of the fluid itself. The deviation vector represents an external source term, boundary condition, or uncoupled driving term. For example, in an electromagnetic field, the deviation vector can be an external current source or voltage source; in a mechanical field, it can be an external load, gravity, or initial stress; in a fluid field, it can be an external pressure difference, mass source, or body force. This deviation vector is the part of the equation that is independent of the input response and is used to introduce the external influence on the system.

[0054] For mechanical-electromagnetic cross-coupling operators, It is an electromagnetic-fluid cross-coupling operator. It is an electromagnetic-mechanical cross-coupling operator. For mechanical-fluid cross-coupling operators, For fluid-electromagnetic cross-coupling operators, It is a fluid-mechanical cross-coupling operator.

[0055] These cross-coupling operators collectively describe how the input of one physical field affects the response of another. For example, to aid understanding of the specific forms of the cross-coupling operators, the electromagnetic-mechanical cross-coupling operator and the mechanical-fluid cross-coupling operator are shown here, with the expressions for the two specific cross-coupling terms as follows:

[0056] ,in, For the integral volume, For current density, It represents the magnetic flux density. The symbol for the cross product of vectors. The symbol for tensor product is... For the Laplace operator, Let be the displacement vector. The above equation indicates that this integral is applied to the volume. The above indicates the effect of the mechanical effect (Lorentz force) generated by the interaction of current and magnetic field on the displacement field.

[0057] ,in, The symbol is for partial differentials. For fluid density, For the fluid velocity field, The density of the volume force; Let be the stress tensor. The above equation shows how mechanical deformation affects the momentum and flow field changes of a fluid. As can be seen from these two cross-coupled terms in the example, these coupling terms allow the interaction between different physical fields to be described. For example, in the two equations shown, the Lorentz force generated by the electromagnetic field affects the deformation of the mechanical field, which in turn affects the motion of the fluid.

[0058] It should be noted that in the above formula, the matrix This represents the output response, that is, the set of response quantities or output quantities in each physical field. Specifically, it can be the evolution rate of physical field variables, the residuals of equilibrium equations, or the driving force / flux of the system in the current state. This vector usually originates from the discretized form of the governing equations of each physical field (such as Maxwell's equations, elasticity equations, and Navier-Stokes equations), and is the result that the model attempts to solve or predict. Matrix The input response represents the input quantity or state variable of each physical field, which is the key variable describing the current state of the physical field. The goal of this unified coupling equation is to map these input variables to the output response through coupling operators.

[0059] Coupled Tensor This is a 3×3 matrix where each element is an operator (which can be a differential operator, integral operator, matrix, or a more complex nonlinear mapping) that defines how the input response is transformed into the output response and how different physical fields influence each other. The main diagonal of the matrix contains self-coupling operators, describing the internal physical laws and evolution of each physical field without considering direct interactions with other physical fields; the off-diagonal terms are cross-coupling operators, describing how the input of one physical field affects the response of another. Compared to traditional coupled problems that require iterative solutions to the equations of different physical fields, leading to convergence and computational inefficiencies, this implementation, through unified coupling operators, can effectively construct and solve the problem, thereby improving computational efficiency and stability, especially important for data simulation in strongly coupled problems.

[0060] 3) Decompose the problems in the simulation process into different physical scales and establish models at each scale. Specifically, the physical scales can be divided into microscopic, mesoscopic, and macroscopic scales. The microscopic scale mainly focuses on the constitutive relations of the magnetic yoke lamination material, such as magnetostriction, thermal expansion, and elastoplasticity. The mesoscopic scale mainly focuses on the macroscopic effects of the structural units composed of microscopic materials, such as eddy current losses, magnetic domain motion, and local stress concentration in the magnetic yoke lamination. The macroscopic scale mainly focuses on the behavior of the magnetic yoke lamination as a whole, which can be described using theories such as continuum mechanics and macroscopic electromagnetics.

[0061] By establishing rules governing connections between different scales and ensuring physical consistency across boundary conditions, conservation laws, and coupling interfaces, the parameters and constitutive relations of microscopic materials (such as magnetostriction coefficients) can be applied upwards to the mesoscopic scale through homogenization techniques or equivalent medium theory, thereby influencing macroscopic properties (e.g., affecting stress intensification parameters in the mesoscopic eddy current loss equation). Information such as local losses and stress distributions calculated at the mesoscopic scale can then serve as macroscopic source terms or effective material parameters, reflecting them at the entire system level. Macroscopic field variables (such as macroscopic magnetic fields and stress fields) can act as boundary conditions or driving forces, driving downwards to local analyses at the microscopic or mesoscopic scales to obtain more refined details.

[0062] For example, in this embodiment, the microscale can be determined by focusing on the behavior of the grains, the most basic building blocks of the magnetic yoke laminate material. The magnetostrictive coupling equation is constructed through the constitutive relations of the material, i.e., its intrinsic response to force, magnetic field, and heat, as shown in the following equation:

[0063] ,

[0064] in, For the total strain tensor, the subscripts i and j indicate the components in the Cartesian coordinate system, that is, this is a second-order tensor that describes the degree and direction of material deformation. The superscript M emphasizes that this is the micro-strain at the material level. For elastic stiffness tensor, which is a fourth-order tensor, it describes the elastic properties of the material and is used to capture the anisotropy of the material (i.e. the different elastic responses of the material in different directions). For isotropic materials, it can be simplified to Young's modulus and Poisson's ratio. For stress tensor, it is used to describe the force per unit area inside a material; For the magnetostriction coefficient tensor, which is a fourth-order tensor that quantifies how magnetic field strength is converted into strain; and These are the two components of the magnetic flux density, because the magnetostrictive effect is usually related to the square of the magnetic field. The coefficient of thermal expansion; This refers to the change in temperature. The symbol for Kronecker indicates that thermal expansion is generally an isotropic volumetric change (unless the material itself is anisotropic in thermal expansion).

[0065] At the mesoscale, we can focus on the behavior of mesostructures composed of micromaterials. At this scale, microscopic constitutive relations still play a role, but more macroscopic effects, such as eddy current losses and stress intensification, begin to become important. This leads to the construction of the eddy current loss-stress coupling equation, as shown below:

[0066] ,

[0067] in, This represents the total eddy current loss, which is one of the main forms in which electromagnetic energy is converted into heat energy in conductive materials. For electrical conductivity, The magnetic field strength vector. The curl of the magnetic field strength, according to Ampere's circuital law, is actually equal to the current density (in steady state). Coupling factor This is static pressure, which is the uniform compressive stress acting on the material; The equivalent plastic strain rate is the rate at which the material undergoes plastic deformation. This equation describes how, at the mesoscale, total energy loss (specifically eddy current loss) is affected by both electromagnetic and mechanical effects.

[0068] The macroscopic scale can be achieved by focusing on the overall behavior of the entire magnetic yoke stack. At this scale, methods from continuum mechanics and macroscopic electromagnetics are typically employed, ignoring microscopic and mesoscopic details and describing the system through equivalent macroscopic parameters. By coupling the governing equations of electromagnetic, mechanical, and fluid fields together, the dynamic equilibrium of the entire macroscopic system under the influence of different physical fields is described, resulting in a unified multi-field governing equation, as shown in the following equation:

[0069] ,

[0070] in, It is a nonlinear differential operator that acts on the state vector of a multiphysics field and contains the partial differential equations of each physical field. This is a multiphysics state vector, which can directly correspond to the input-output response in the unified coupling equation, except that the specific field variables are specified here. Resistivity It is a magnetic vector potential. This is the elastic constant matrix, which contains information such as the material's Young's modulus and Poisson's ratio. It is related to the aforementioned microscale elastic stiffness tensor, but is usually macroscopically equivalent. It is the displacement vector; For fluid density; This is a coupling source term.

[0071] The coupled source term represents the cross-coupling effect between different physical fields and the external load / source term, and can be expressed by the following equation:

[0072] ,

[0073] in, The gradient of the electric potential generates electric field components; For mass density, It is the acceleration due to gravity. Here is the coupling stiffness matrix; The coefficient of thermal expansion of the fluid. The displacement vector is the displacement of the coupled body.

[0074] The electromagnetic source term is the current density term caused by induced current and potential gradient, and can be used as a source term in Maxwell's equations. This represents the mechanical source term, which includes gravity and contact / connection forces. This term can be considered as part of the response of the mechanical field, including external forces and interactions within the solid. This usually indicates that the buoyancy term is the manifestation of the change in fluid density caused by temperature changes in the gravitational field. This term can correspond to the response of the fluid field, thus coupling temperature (which may come from heating caused by electromagnetic fields or external heat sources) into the fluid dynamics.

[0075] 4) Finally, the process parameters of the target magnetic yoke stack are embedded as constraints into the space of the multiphysics tensor, resulting in the knowledge embedding tensor as shown in the following equation:

[0076] , ,

[0077] in, Embedding tensors for knowledge This represents the tensor multiplication by pattern. This is the core parameter matrix of the process parameters. This indicates the specific core process parameters. This is the transpose symbol for a matrix.

[0078] In this embodiment, a unified coupling equation is used to accurately simulate the detailed physical field variable distribution of the entire system under given input and boundary conditions, and this detailed distribution data serves as the foundation. To elevate this low-level physical information to a more meaningful "process-performance" correlation level for engineering simulation, the calculated detailed physical field results are refined and summarized using knowledge embedding tensors. Furthermore, by injecting process parameter vectors, the physical results are directly correlated with actual production process parameters, enabling the model to directly optimize and make decisions at the process level.

[0079] It should be noted that the mathematical model disclosed in this embodiment, through knowledge embedding tensors, enables the model to store and represent system characteristics under different combinations of parameters from different physical domains in a compact and multi-dimensional manner, realizing the embedding of physical knowledge, much like a multi-dimensional lookup table, linking complex parameter relationships and system responses. Combined with coupling tensors, the self-coupling and cross-coupling operators of various physical fields are uniformly represented in matrix form, making the model structure clear, easy to understand and extend, and systematically capturing all possible interactions between physical fields. Finally, combined with multi-scale hierarchical modeling, the model follows different dominant laws at different scales. Although each scale has independent equations, there is implicit or explicit information transmission between the equations. For example, in the above formula, the accumulated microscopic magnetostriction effect affects the mesoscopic stress state, thus affecting eddy current losses; while mesoscopic losses and stresses serve as source terms or parameters for macroscopic equations. This hierarchical processing allows the model to simultaneously consider both details and the overall picture.

[0080] In this embodiment, the knowledge graph collaboration mechanism includes the following:

[0081] 1) Based on the constructed knowledge embedding tensor, perform data mining and analysis on the knowledge embedding tensor to identify the higher-order associations and patterns contained therein. Higher-order associations and patterns in the knowledge embedding tensor can be identified through machine learning algorithms (such as decision trees and association rule mining) or human expert experience, and these higher-order associations and patterns can be transformed into logical reasoning rules in IF-THEN form. For example, when simulating with magnetic induction intensity as the main parameter, its higher-order associations and patterns might be that when the magnetic induction intensity is within a certain range and the superposition coefficient is within a certain range, a specific performance indicator exhibits outliers.

[0082] 2) The extracted reasoning rules are used as ontology or axioms and re-injected into the knowledge graph database. The reasoning engine of the knowledge graph uses these rules to discover or predict new knowledge.

[0083] For example, the numerical knowledge embedded in a tensor can be transformed into reasoning rules that a knowledge graph can understand and utilize, as expressed by the following formula:

[0084] ,

[0085] KG reasoning rules are logical reasoning rules that knowledge graphs can execute. In this embodiment, they are represented in the form of "IF-THEN". Knowledge graphs use these rules to infer the relationships or attributes between entities.

[0086] This represents a value extracted or computed from the knowledge embedding tensor. This represents a tensor slice or eigenvalue at a magnetic flux density of B = 1.62T. This represents a tensor slice or eigenvalue at a superposition factor λ = 0.93. These are tensor operation symbols (such as tensor product, Hadamard product, or more complex tensor decomposition / contraction) used to fuse tensor information under different parameter conditions to represent the system state characteristics under a specific combination of process parameters. This is a critical threshold, which may be set based on experience, design specifications, or hazard thresholds. This means that if the aforementioned conditions are met, the knowledge graph will predict a certain physical quantity, which in this case is expressed as the deformation width of the magnetic yoke laminations exceeding 0.2 mm.

[0087] Step 4: Use the data in the knowledge embedding tensor as input to perform finite element simulation. Based on the new simulation results, extract the feature values ​​of the dimensions corresponding to the knowledge embedding tensor, update the knowledge embedding tensor, and form new knowledge embedding tensor data.

[0088] The difference between the new knowledge embedding tensor data and the predicted knowledge embedding tensor data of the current model is calculated. Based on the calculated error, the data in the knowledge embedding tensor is iteratively updated and corrected according to the defined update rules.

[0089] For example, updating the knowledge embedding tensor based on simulation results can be expressed as follows:

[0090] ,

[0091] in, The update quantity for the knowledge embedding tensor. The learning rate is a positive number less than 1, used to control the step size of each update, preventing the model from overfitting the new data and maintaining stability. These are the tensor eigenvalues ​​obtained from the finite element method. The preceding model is the tensor feature value of the tensor prediction embedded with existing knowledge; This is the confidence matrix. By introducing the confidence matrix, the model can selectively trust new data. For data with high confidence, the correction will be smaller; for data with low confidence, the model will make corrections based on it to a greater extent, thereby avoiding contamination of the model by erroneous or noisy data.

[0092] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A knowledge graph-based magnetic yoke lamination process parameter simulation optimization method, characterized in that, The simulation optimization method includes: S100. Construct a magnetic yoke stack database to store relevant data for magnetic yoke stack simulation; S200. Construct a knowledge graph database based on the data in the magnetic yoke stack database; S300 utilizes tensor analysis and multiphysics coupling to construct an analysis and optimization algorithm, and combines it with a magnetic yoke stack database and knowledge graph library to provide a preliminary analysis framework for simulation design. The analysis and optimization algorithm includes tensor analysis-multiphysics coupled mathematical model and knowledge graph collaborative mechanism. Tensor analysis and multiphysics coupling mathematical models are used to represent multiphysics parameters and their interactions in a unified manner through high-dimensional tensors. Different physical field parameters in the magnetic yoke stacking simulation process are embedded into each element of the tensor to construct a multiphysics tensor. By integrating the parameters of different physical fields at the tensor level, the problems in the simulation process are decomposed into different physical scales. The knowledge graph collaboration mechanism obtains a knowledge embedding tensor embedded in the knowledge graph through tensor analysis and multiphysics coupling mathematical model, and performs data mining and analysis on the knowledge embedding tensor to identify higher-order associations and patterns. S400. Take the data in the knowledge embedding tensor as input, perform finite element simulation, and update the knowledge embedding tensor according to the new simulation results to form new knowledge embedding tensor data.

2. The knowledge graph based magnetic yoke lamination process parameter simulation optimization method according to claim 1, characterized in that, The magnetic yoke stack database includes: Design parameters, material properties, process constraints, empirical data, and standard specifications for magnetic yoke laminations; The data in the database is cleaned and parsed to construct the magnetic yoke stack database.

3. The knowledge graph based magnetic yoke lamination process parameter simulation optimization method according to claim 1, characterized in that, The construction of the knowledge graph library includes: Transform the data in the magnetic yoke stack database into entities, relationships, and attributes that can be understood by the knowledge graph. For structured data, the structured data is directly mapped to the entities and attributes defined in the ontology; For unstructured data, key entities in the text are identified by named entities, semantic relationships between entities are extracted by relation extraction, and event information related to process flow and fault analysis is identified by event extraction. The extracted and semantically encoded knowledge is imported into the knowledge graph database in batch or streaming mode. Consistency checks and data quality verification are performed on the imported knowledge graph data to ensure the accuracy of the knowledge.

4. The knowledge graph based simulation optimization method of process parameters for magnetic yoke laminations according to claim 1, characterized in that, The tensor analysis-multiphysics coupling mathematical model is constructed through the following steps: S310. Select the physical fields required in the simulation of magnetic yoke stacking, and embed the core parameters of the selected physical fields and their potential interactions into the data structure to construct the core multiphysics tensor. S320. Transform the various parameter relationships in the multiphysics tensor into operable physical field interaction operators. The operable physical field interaction operators include self-coupling operators and cross-coupling operators. The self-coupling operator and the cross-coupling operator are integrated into a matrix form, with the self-coupling operator as the operator terms on the diagonal and the cross-coupling operator as the operator terms off-diagonal, to construct the total coupling operator matrix. By combining the state vectors of each physical field and the external source terms, a unified coupling equation is constructed. S330. Decompose the problems in the simulation process of magnetic yoke stacking into different physical scales, and establish descriptive models for different scales. S340. The process parameters of the magnetic yoke laminations are embedded as constraints into the space of the multiphysics tensor to obtain the knowledge embedding tensor.

5. The knowledge graph based magnetic yoke lamination process parameter simulation optimization method according to claim 4, characterized in that, S310 further includes: determining the dimension of the multiphysics tensor based on the number of selected physical fields, where each dimension corresponds to the parameter space of a physical field, and the size of the dimension depends on the degree or range of parameter discretization. This allows us to construct the tensor dimension and index space of the multiphysics tensor. Finally, the data stored in the database is populated into the tensor dimension and index space of the constructed multiphysics tensor.

6. The knowledge graph based magnetic yoke lamination process parameter simulation optimization method according to claim 4, characterized in that, The self-coupling operator is constructed by extracting the evolution or equilibrium mechanism of each physical field and is used to describe the physical laws inside the physical field. Cross-coupling operators are constructed based on the interaction mechanisms between different physical fields and are used to describe the mutual responses between different physical fields.

7. The knowledge graph based simulation optimization method of process parameters for magnetic yoke laminations according to claim 4, characterized in that, The different scales in S330 include: dividing the physical scale into microscopic scale, mesoscopic scale, and macroscopic scale, wherein... At the microscale, we focus on the constitutive relations of magnetic yoke laminate materials. The mesoscale focuses on the macroscopic effects of structural units composed of microscopic materials. At the macro scale, attention is paid to the overall behavior of the magnetic yoke stack; By establishing connection rules between different scales, and ensuring that different scales maintain physical consistency in boundary conditions, conservation laws, and coupling interfaces.

8. The knowledge graph based magnetic yoke lamination process parameter simulation optimization method according to claim 1, characterized in that, The knowledge graph collaboration mechanism also includes: Identify and mine higher-order associations and patterns in knowledge embedding tensors, and transform these associations and patterns into logical reasoning rules of IF-THEN form. The extracted reasoning rules are used as ontology or axioms and re-injected into the knowledge graph database. The reasoning engine of the knowledge graph database then uses the injected reasoning rules to discover or predict new knowledge.

9. The knowledge graph based simulation optimization method of process parameters for magnetic yoke laminations according to claim 1, characterized in that, The S400 also includes: Calculate the difference between the new knowledge embedding tensor data and the predicted knowledge embedding tensor data of the current model. Based on the calculated error, the data in the knowledge-embedded tensor is iteratively updated and corrected according to the update rules.

10. A knowledge graph based magnetic yoke lamination process parameter simulation optimization system, characterized in that, The simulation optimization system includes: processor; The memory stores a computer program that, when executed by a processor, implements the knowledge graph-based simulation optimization method for magnetic yoke stacking process parameters as described in any one of claims 1 to 9.

Images