Motor rapid design method based on large model

Abstract

The invention discloses a rapid motor design method based on a large model, and belongs to the field of motor design, and the method comprises the steps: carrying out the structural processing of geometric parameters, material characteristics, performance indexes and working condition data of a motor, building a nonlinear mapping relation between design parameters and the performance indexes, and constructing a physical information neural network frame, according to the method, four-step closed-loop optimization of parameter coding, neural network calculation, physical consistency verification and performance decoding is realized, an electromagnetic rule and data driving are creatively fused, dispersed engineering experience is converted into a structured knowledge base, strong support is provided for an engineer to quickly generate a physically reasonable motor scheme which meets specific requirements, and the method is suitable for popularization and application. According to the method, efficient and accurate motor intelligent design is realized, so that the motor design period is greatly shortened, the rapid iteration requirement of the industry is met, the design efficiency is remarkably improved, and a new normal form is provided for the motor intelligent design.

Description

Technical Field

[0001] This invention belongs to the field of motor design, and in particular relates to a rapid motor design method based on a large model. Background Technology

[0002] With the rapid development of new energy vehicles, industrial automation, and high-end equipment manufacturing, motor design faces unprecedented challenges in performance optimization and development efficiency improvement. Existing motor design methods primarily rely on traditional physical models such as electromagnetic theory, thermodynamics, and mechanical structure analysis, combined with finite element simulation to calculate electromagnetic fields, temperature rise, and structural strength. While these methods can accurately reflect motor performance, the design process requires multiple parameter iterations and simulation calculations under multidisciplinary coupling, resulting in significant time consumption and computational complexity. Furthermore, design results often heavily depend on engineer experience, leading to low design efficiency, large result discrepancies, and difficulty in rapidly generating solutions, thus failing to meet the industry's demands for rapid product iteration and customized development.

[0003] However, existing motor design methods still have significant shortcomings in terms of intelligence and data fusion. Current AI-assisted design largely focuses on neural network fitting based on limited datasets, lacking physical constraints on electromagnetic laws, resulting in insufficient model generalization ability and inconsistent prediction results. Furthermore, traditional models struggle to integrate multi-physics data such as electromagnetic, thermal, and structural data, failing to accurately characterize the complex nonlinear relationships between design parameters and performance indicators. Due to the lack of unified data standards and effective pre-training mechanisms, existing design tools struggle to achieve efficient mapping from raw data to high-dimensional features, and cannot quickly predict and optimize the overall performance of motors. Therefore, there is an urgent need for an intelligent motor design method that can integrate physical laws and data-driven features, possessing multi-domain information understanding and adaptive learning capabilities, to achieve a high-precision and high-efficiency motor design process. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a rapid motor design method based on a large model, comprising: The large model technique is used to obtain multi-source data for multi-source motor design, including geometric parameters, material properties, performance indicators, and operating condition data. A parameter coding model is constructed based on the multi-source data, and the multi-dimensional design parameters are converted into a high-dimensional vector representation suitable for neural network processing. A nonlinear mapping relationship between motor design parameters and performance indicators is established using a physical information neural network, and a physical constraint term is introduced into the network loss function; The residuals between the prediction results and physical laws are calculated through physical consistency verification and weight adjustments are made accordingly. Based on the optimized network output, the key performance indicators of the motor are decoded, thereby generating a motor design scheme that meets the target performance requirements.

[0005] Optionally, the process of obtaining multi-source data for multi-source motor design based on large model technology includes: extracting original parameters from simulation results of motor electromagnetics, thermal, structure and other fields, and establishing a unified data format and processing flow according to the characteristics of various types of data in order to form a standardized training dataset.

[0006] Optionally, the construction of the parameter encoding model includes: The model input is determined based on design variables such as geometric dimensions, material properties, and winding configuration. The model output is determined based on performance indicators such as torque, power, and efficiency. The input variables are standardized and vectorized to transform them into a parameterized representation with a unified dimension. A high-dimensional vector input network model is formed based on a unified format.

[0007] Optionally, the establishment of the physical information neural network includes: Embed Maxwell's equations, the law of electromagnetic induction, and other electromagnetic physical laws into neural networks; Design a multi-layer encoder-decoder architecture based on the training structure of a large model; Furthermore, during network training, physical constraints and data loss terms are combined to optimize model parameters.

[0008] Optionally, the physical constraints include: Basic layer constraints are used to ensure the divergence-free condition of the electromagnetic field; Application layer constraints are used to constrain motor characteristic formulas such as torque and electric power. Engineering constraints are used to limit efficiency, temperature rise, and the range of materials used. The constraint terms at each level are adaptively weighted and adjusted based on the loss convergence.

[0009] Optionally, the physical consistency check includes: The electromagnetic field residuals are calculated based on the prediction results of the trained model. The physical constraint weights are dynamically adjusted according to the magnitude of the residuals, and the degree of satisfaction of the boundary conditions is checked, including the continuity of the normal component of the magnetic induction intensity at the air gap, the continuity of the tangential component of the magnetic field intensity at the ferromagnetic material interface, and periodic boundary conditions.

[0010] Optionally, the key performance indicators of the motor based on the optimized network output include: extracting the torque density, power density and loss characteristics of the motor according to the high-dimensional feature distribution of the physical information neural network output, and performing end-to-end optimization of the model parameters through a joint loss function to obtain a motor design scheme that meets the predetermined performance indicators.

[0011] Optionally, the joint loss function includes a data loss term, a physical loss term, and a boundary loss term, and the generalization ability of the model is gradually optimized through a multi-stage progressive training strategy.

[0012] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0013] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects: This invention introduces a rapid motor design method based on a large model. It utilizes a framework combining the data learning capabilities of the large model with a Physical Information Neural Network (PINN) to unify the modeling of simulation data from multiple domains, including electromagnetics, thermal, and structural aspects of the motor, establishing a nonlinear mapping relationship between design parameters and performance indicators. This method achieves a deep fusion of electromagnetic laws and data-driven characteristics through joint optimization of physical constraints and neural networks, maintaining physical consistency while improving the model's generalization ability. Through a closed-loop mechanism of parameter encoding, physical verification, and performance decoding, the system can automatically generate motor design schemes that meet specific performance requirements, significantly reducing manual calculations and simulation iterations, improving design efficiency and accuracy, and effectively solving the problems of high computational load, long cycle time, and poor generalization in traditional motor design methods. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a training flowchart of the motor rapid design system based on a large model according to an embodiment of the present invention; Figure 2 This is a flowchart of the motor design process of the rapid motor design system based on a large model, according to an embodiment of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0018] Example 1 This embodiment provides a rapid motor design method based on a large model, including: The system model training method based on large model technology forms a complete parameter-performance mapping closed loop through the Physical Information Neural Network (PINN).

[0019] The system model training method based on large model technology: By leveraging the learning capabilities of massive data and integrating simulation data from multiple fields such as motor electromagnetics, thermal, and structure, a complex nonlinear mapping relationship between design parameters and performance indicators can be established. Pre-training technology performs comprehensive structured processing on various parameters involved in motor design, including geometric parameters, material properties, performance indicators, and actual operating condition data.

[0020] The physical information neural network part: The physical information neural network framework is constructed using large model technology, which forms four stages: parameter encoding and processing, PINN-assisted construction of the basic architecture of the large model, physical consistency verification and performance decoding, and establishes a complete parameter-performance mapping closed loop.

[0021] Furthermore, the geometric parameters include key dimensions such as stator outer diameter, stator inner diameter, number of slots, slot size, stator tooth width, stator yoke thickness, rotor outer diameter, permanent magnet size, number of permanent magnet pole pairs, air gap length, motor shaft length, number of conductors per slot, winding pitch, and wire diameter; Furthermore, the material properties include physical attributes such as magnetic permeability, electrical conductivity, loss characteristics, remanence, coercivity, density, thermal conductivity, and dielectric strength. Furthermore, the performance indicators include key indicators such as rated power, speed, torque, efficiency, power density, torque density, load line voltage, no-load air gap magnetic flux density, no-load back EMF, torque pulsation, cogging torque, and losses. Furthermore, the actual operating condition data includes factors such as load characteristics, control method, temperature, humidity, altitude, stall state, and unbalanced power supply. Furthermore, by establishing standardized data formats and unified processing procedures, these multi-dimensional design parameters are converted into high-dimensional vectors suitable for large-scale model processing, ultimately extracting the deep design patterns of motor design. Furthermore, during the design process, the preliminary size range of the motor can be obtained by using the relationship between the main parameters of the motor. Furthermore, the main parameters of the motor are related as follows: ; Where D is the outer diameter of the motor. For motor shaft length, To calculate power, This refers to the motor speed. To calculate the polar arc coefficient, The air gap magnetic field waveform coefficient is 1.11 (for sinusoidal distribution). For the armature winding coefficient, For line load, This represents the maximum value of the air gap magnetic flux density. This is the ratio of the induced electromotive force to the terminal voltage under rated load. For efficiency, For power factor, This represents the number of phases in the armature winding. The number of turns in series per phase. This refers to the armature winding phase current. This refers to the motor size ratio. This is the phase electromotive force of the armature winding. For the current frequency, The number of conductors in series per phase. For the fundamental winding coefficient, For each pole flux, Number of conductors per slot For the number of parallel branches, For the number of slots, The amplitude of the fundamental magnetic flux density. For polar distance, For the length of the motor stack, Rated power; Furthermore, during the design process, when the number of slots, the number of phases, and the rated speed remain constant, the induced voltage is proportional to the number of conductors in series per phase of the motor, the magnetic flux, and the motor shaft length. Under the constraint of the bus voltage, it is necessary to reasonably adjust the relationship between the number of turns, the stator outer diameter, and the motor shaft length. Furthermore, the heat dissipation capacity of the motor is related to the current density and the line load. The current density is directly proportional to the effective value of the current and the number of conductors per slot, and inversely proportional to the slot area. The variation law of the line load is directly proportional to the number of phases, the number of turns in series per phase, and the effective value of the current. Furthermore, the electromagnetic torque of the motor is positively correlated with the armature current and the magnetic flux per pole. In practice, optimal torque output and efficiency can be achieved by precisely adjusting the ratio of the direct-axis and quadrature-axis currents and the relationship between various dimensions. Furthermore, to refine the model, the effects of variations in the number of slots and conductors must be considered. Adjusting the number of slots affects parameters such as tooth width, slot opening size, and slot area, which in turn influence current density distribution and voltage characteristics, ultimately affecting the motor's output torque. Therefore, these parameters require dynamic adjustment. Furthermore, in order to achieve a deep integration of electromagnetic laws and data-driven approaches, key parameters in electromagnetic laws are mapped to nodes in the input layer of a neural network, and the deviation between the performance predictions and theoretical calculations of the network output is constrained by a physical loss term, thereby explicitly embedding electromagnetic laws into the model and improving its interpretability and generalization ability. Furthermore, the system utilizes the preliminary scheme range and, based on the complex nonlinear mapping relationship between motor parameters and performance indicators, obtains the motor design scheme; Furthermore, the aforementioned complex parameter relationships and performance optimization process can be divided into four stages and four key steps: parameter encoding processing, core calculation, physical consistency verification, and performance decoding, forming a complete parameter-performance mapping closed loop.

[0022] Furthermore, the parameter encoding process involves converting the input motor design parameters, material properties, performance indicators, and actual operating condition data into standardized vectors. This establishes a standardized data format and a unified processing flow, effectively converting these multi-dimensional, multi-physical domain design parameters into a high-dimensional vector representation suitable for neural network processing. S1. Determine the model's inputs and outputs: Based on the specific application scenario of the motor, clarify the model's inputs and outputs. The inputs include geometric dimensions, material properties, and winding configuration, while the outputs include electromagnetic torque characteristics, power output capability, and operating efficiency, which are used to guide the subsequent construction of the PINN loss function. S2. Data preprocessing includes the following steps: S2.1 Parameter Encoding Processing: This process transforms input motor design parameters (such as geometric dimensions, material properties, and winding configuration) into standardized vectors, enabling the input layer to receive parameterized representations with a unified dimension. This includes standardization of key dimensions such as stator outer diameter, motor shaft length, and stator slot dimensions; vectorized representation of material properties such as permeability, conductivity, and density; and standardized encoding of performance indicators such as rated power, speed, torque, and efficiency.

[0023] S2.2 Processing actual operating data, including parameterized representations of environmental factors such as temperature, humidity, and load changes: By establishing standardized data formats and unified processing procedures, these multi-dimensional, multi-physical domain design parameters are effectively converted into high-dimensional vector representations suitable for neural network processing.

[0024] S3. Construct the basic architecture of a large model using PINN assistance, including the following steps: S3.1 Establishing the PINN physical knowledge base: Learning physical laws such as Maxwell's equations and the law of electromagnetic induction through the PINN network, generating a large number of motor design parameter-performance data pairs that conform to physical constraints; S3.2 Physical Law Labeling: The trained PINN network is used to systematically sample the motor design space, and the physical feasibility, boundary condition satisfaction degree and performance prediction results of each parameter combination are labeled. S3.3 High-quality training data generation: The data generated by PINN meets physical constraints, providing richer and more accurate training samples for large models than traditional simulation data; S3.4PINN loss function design: The degree of physical constraint violation calculated by PINN is used as an additional loss term for training large models, guiding large models to learn physically reasonable parameter generation patterns; S3.5 Learning Strategy: First, let the large model learn the simple physical relationships verified by PINN, and then gradually transition to complex multiphysics coupling problems to ensure the stability and effectiveness of the learning. S3.6 Comparative Learning Mechanism: PINN is used to distinguish between physically reasonable and unreasonable design schemes, training large models to have the ability to make physical judgments and avoiding the generation of parameter combinations that violate physical laws.

[0025] S4. The establishment of the physical consistency verification process includes the following steps: S4.1 Residual Verification: Based on the fundamental equations of electromagnetic fields, the residuals between the predicted results and physical laws are calculated to ensure that the network output conforms to actual physical laws. By directly encoding electromagnetic physical laws such as Maxwell's equations and heat conduction equations into the loss function of the neural network, a multi-objective constraint structure containing data loss terms, physical loss terms, and boundary loss terms is formed. Verification mechanism of physical laws S4.2: Whether the magnetic field distribution predicted by the system calculation network satisfies... The divergence-free condition is used to verify whether the electric and magnetic fields conform to Faraday's law of electromagnetic induction. Check whether the magnetic field strength and current density obey Ampere's circuital law. ; S4.3 Boundary condition verification: Verify the degree to which the boundary conditions are met, including the continuity of the normal component of the magnetic induction intensity at the air gap, the continuity of the tangential component of the magnetic field intensity at the ferromagnetic material interface, the sinusoidality of the no-load back EMF, and the saturation of the stator and rotor magnetic flux density. S4.4 Adaptive Weight Adjustment Mechanism: During model training, the loss coefficients are dynamically adjusted based on the actual convergence of each loss, ensuring that the network strictly follows the basic physical laws of electromagnetic fields while learning data features, and guaranteeing that the network prediction results fit the training data mathematically while maintaining rationality and consistency in a physical sense.

[0026] S5. Performance decoding and optimization output includes the following steps: S5.1 Key Performance Indicator Extraction: Key performance indicators such as torque density, torque ripple, and power are extracted from the optimized physical field distribution and end-to-end optimization is performed through a joint loss function that includes data loss and physical constraints. S5.2 Multi-stage progressive training strategy: A multi-stage progressive training strategy is adopted, which combines independent validation sets to continuously evaluate the generalization performance of the model and achieve fine-tuning of model parameters.

[0027] S6. Comprehensive Performance Prediction and Solution Output: The system achieves accurate prediction from electromagnetic field distribution to comprehensive motor performance indicators, including electromagnetic torque characteristics, power output capability, and operating efficiency, and finally outputs a motor solution that meets design requirements.

[0028] S7. Results Integration and Report Generation: Integrate all calculation results and generate relevant technical reports. The large-model-based rapid motor design system enables rapid output of motor solutions, reducing the complexity of motor design and shortening the motor design cycle.

[0029] Example 2 This embodiment provides a rapid motor design method based on a large model, including: In this embodiment, the core of the rapid motor design system lies in constructing a physical information neural network framework using large model technology. By directly encoding electromagnetic physical laws such as Maxwell's equations into the loss function of the neural network, multiple objective constraints are formed, including data loss terms, boundary loss terms, and physical loss terms. During model training, an adaptive weight adjustment mechanism is adopted to dynamically adjust the loss coefficients based on the actual convergence of each loss term. Simultaneously, a multi-stage progressive training strategy is implemented, and the generalization performance of the model is continuously evaluated using a validation dataset independent of the training set, thereby achieving fine-tuning of the model parameters.

[0030] See Figure 1In this embodiment, at the data processing level, with the help of pre-training technology, the system performs comprehensive structured processing on various parameters involved in motor design, including geometric parameters (stator outer diameter, stator inner diameter, number of slots, slot size, stator tooth width, stator yoke thickness, rotor outer diameter, permanent magnet size, number of permanent magnet pole pairs, air gap length, motor shaft length, number of conductors per slot, winding pitch, wire diameter, and other key dimensions), material properties (permeability, conductivity, loss characteristics, remanence, coercivity, density, thermal conductivity, dielectric strength, and other physical properties), and performance indicators (rated power, speed, torque, efficiency, power density, torque density, load line voltage, etc.). Key indicators such as no-load air gap magnetic flux density, no-load back EMF, torque pulsation, cogging torque, and losses, as well as actual operating condition data (load characteristics, control method, temperature, humidity, altitude, stall state, unbalanced power supply, etc.), are transformed into high-dimensional vector representations suitable for large-scale model processing by establishing standardized data formats and unified processing procedures. Ultimately, this enables the large-scale model to accurately learn and establish the complex nonlinear mapping relationship between motor design parameters and performance indicators, achieving efficient and accurate intelligent motor design.

[0031] See Figure 2 In this embodiment, during the design process, when the system receives the main performance indicators of the designed motor (such as rated power, rated speed, rated torque, bus voltage, motor external dimensions, etc.) as input, it generates an initial scheme through the electromagnetic relationship of the motor.

[0032] The main parameters of the motor are related as follows: ; Where D is the outer diameter of the motor. For motor shaft length, To calculate power, n is the motor speed. To calculate the polar arc coefficient, The air gap magnetic field waveform coefficient is 1.11 (for sinusoidal distribution). Where is the armature winding factor, and A is the line load. This represents the maximum value of the air gap magnetic flux density. This is the ratio of the induced electromotive force to the terminal voltage under rated load. For efficiency, Here, m is the power factor, N is the number of phases in the armature winding, I is the number of turns in series per phase, and m is the phase current in the armature winding. Where E is the motor size ratio, E is the armature winding phase electromotive force, and f is the current frequency. The number of conductors in series per phase. For the fundamental winding coefficient, For each pole flux, Number of conductors per slot For the number of parallel branches, For the number of slots, The amplitude of the fundamental magnetic flux density. For polar distance, For the length of the motor stack, This is the rated power.

[0033] See Figure 2 In this embodiment, based on the above linear relationship, the preliminary size range of the motor can be obtained. The system uses the preliminary range of the scheme to obtain the design scheme of the motor based on the complex nonlinear mapping relationship between the motor parameters and performance indicators.

[0034] See Figure 2 In this embodiment, during the motor design process, when the number of slots, number of phases, and rated speed remain constant, the induced voltage is proportional to the number of conductors in series per phase, the magnetic flux, and the motor shaft length. Under the constraint of the bus voltage, the relationship between the number of turns, stator outer diameter, and motor shaft length needs to be reasonably adjusted. Simultaneously, the motor's heat dissipation capacity is related to the current density and line load. The current density is directly proportional to the effective current value and the number of conductors per slot, and inversely proportional to the slot area. The variation of the line load is directly proportional to the number of phases, the number of turns in series per phase, and the effective current value. The motor's electromagnetic torque is positively correlated with the armature current and the magnetic flux per pole. For the most common surface-mounted permanent magnet motor, since the direct-axis and quadrature-axis inductances are equal, the torque is proportional to the quadrature-axis current, and the proportionality constant is determined by the number of pole pairs and the permanent magnet flux linkage. However, for internal permanent magnet motors, due to magnetic circuit asymmetry, the torque is also affected by the direct-axis current, generating an additional reluctance torque component, making the relationship between torque and current more complex. For motors with more complex structures, their electromagnetic torque is related to the slot size, stator tooth width, and their special structure that differs from ordinary motors. In practice, optimal torque output and efficiency can be achieved by precisely adjusting the ratio of direct-axis and quadrature-axis currents and the relationship between various dimensions.

[0035] To refine the model, the impact of variations in the number of slots and conductors needs to be considered. Adjusting the number of slots directly affects parameters such as tooth width and slot opening size. To ensure the rationality of the magnetic circuit design, these parameters need to be dynamically adjusted inversely to the number of slots. Simultaneously, increasing the number of slots changes the slot area, thus affecting the current density distribution and voltage characteristics, ultimately impacting the motor's output torque. Therefore, the number of conductors per slot needs to be adjusted accordingly. Furthermore, changes in the number of conductors affect the induced voltage. Adjusting the number of winding turns ensures the stability of the bus voltage, guaranteeing voltage consistency under varying parameters. To achieve a deep integration of electromagnetic laws and data-driven approaches, key parameters from the aforementioned electromagnetic laws can be mapped to neural network input layer nodes. A physical loss term constrains the deviation between the network's output performance predictions and theoretical calculations, thereby explicitly embedding electromagnetic laws into the model and enhancing its interpretability and generalization ability.

[0036] The specific implementation plan for the above complex parameter relationships and performance optimization process is as follows: 1. Implementation methods for the core architecture of the large model; The system employs an advanced Transformer architecture as its backbone network, constructing a deep learning model through multi-layer encoders. The network design fully considers the specificities of motor design problems, ensuring both model expressive power and computational efficiency.

[0037] By directly embedding physical laws such as Maxwell's equations and the law of electromagnetic induction into the loss function of a neural network, a learning framework combining data-driven and physical constraints is formed. The loss function consists of three main components: a data loss term, a physical loss term, and a boundary loss term, which are responsible for data fitting, adherence to physical laws, and satisfaction of boundary conditions, respectively.

[0038] The system implements an intelligent adaptive weight adjustment strategy, which can dynamically balance the relationship between data learning and physical constraints based on the convergence of various losses during training. When the data fitting effect is good, the system automatically increases the weight of physical constraints; when the physical constraints are difficult to meet, the system appropriately relaxes the data fitting requirements to ensure that the generated design scheme is physically reasonable and feasible.

[0039] 2. The meticulous design of the PINN loss function; The data loss term employs an adaptive weighting strategy, assigning different weights to different data points based on their importance and reliability. Key performance points (such as rated operating points) are assigned higher weights, boundary operating points are assigned moderate weights, and interpolation points are assigned lower weights.

[0040] Hierarchical construction of physical loss terms: The physical loss term is constructed using a hierarchical approach. The foundational layer contains strict constraints from Maxwell's equations, ensuring the fundamental physical laws of electromagnetic fields; the application layer contains physical constraints specific to motors, such as torque and power formulas; and the engineering layer contains engineering constraints relevant to practical applications, such as efficiency requirements and temperature rise limits. Different weights and penalty mechanisms are used for the physical constraints at different levels.

[0041] Accurate modeling of the boundary loss term: The boundary loss term pays special attention to various boundary conditions in the motor. The air gap boundary is constrained by the continuity of the normal component of magnetic induction intensity, the ferromagnetic interface by the continuity of the tangential component of magnetic field intensity, and the winding boundary by the current density boundary condition. The system also considers periodic boundary conditions to ensure that the periodic symmetry characteristics of the motor are accurately expressed.

[0042] 3. Implementation of progressive learning strategies; Prioritize learning simple physical relationships: The system employs a progressive learning strategy, moving from simple to complex. Initially, it focuses on learning linear electromagnetic relationships, such as the magnetic field distribution under no-load conditions and the current distribution under simple loads. The physical constraints at this stage are relatively simple, making model convergence easier and laying the foundation for subsequent complex learning.

[0043] The gradual introduction of complex coupling problems: After mastering the basic physical relationships, the system gradually introduces complex coupling problems. First, it introduces nonlinear effects of the electromagnetic field, such as the saturation effect of ferromagnetic materials and the demagnetization effect of permanent magnets. Then, it introduces multiphysics coupling effects, such as electromagnetic-thermal coupling and electromagnetic-structural coupling. Finally, it addresses complex coupling problems across the entire operating range.

[0044] Monitoring mechanism for learning stability: The system establishes a monitoring mechanism for learning stability, monitoring the convergence of various losses in real time. When a loss exhibits oscillations or divergence, the system automatically adjusts the learning strategy, such as reducing the learning rate, adjusting weight allocation, or adding regularization constraints. This ensures the stability and effectiveness of the entire learning process.

[0045] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0046] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0047] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A rapid motor design method based on a large model, characterized in that, include: The large model technique is used to obtain multi-source data for multi-source motor design, including geometric parameters, material properties, performance indicators, and operating condition data. A parameter coding model is constructed based on the multi-source data, and the multi-dimensional design parameters are converted into a high-dimensional vector representation suitable for neural network processing. A nonlinear mapping relationship between motor design parameters and performance indicators is established using a physical information neural network, and a physical constraint term is introduced into the network loss function; The residuals between the prediction results and physical laws are calculated through physical consistency verification and weight adjustments are made accordingly. Based on the optimized network output, decode the key performance indicators of the motor to generate a motor design scheme that meets the target performance requirements; The physical consistency check includes: The electromagnetic field residuals are calculated based on the prediction results of the trained model. The physical constraint weights are dynamically adjusted according to the magnitude of the residuals, and the degree of satisfaction of the boundary conditions is checked, including the continuity of the normal component of the magnetic induction intensity at the air gap, the continuity of the tangential component of the magnetic field intensity at the ferromagnetic material interface, and periodic boundary conditions.

2. The method according to claim 1, characterized in that, The process of obtaining multi-source data for multi-source motor design based on large model technology includes: extracting original parameters from simulation results of motor electromagnetics, thermal, structure and other fields, and establishing a unified data format and processing flow according to the characteristics of various data in order to form a standardized training dataset.

3. The method according to claim 1, characterized in that, The construction of the parameter encoding model includes: The model input is determined based on design variables such as geometric dimensions, material properties, and winding configuration. The model output is determined based on performance indicators such as torque, power, and efficiency. The input variables are standardized and vectorized to transform them into a parameterized representation with a unified dimension. A high-dimensional vector input network model is formed based on a unified format.

4. The method according to claim 1, characterized in that, The establishment of the physical information neural network includes: Embed Maxwell's equations, the law of electromagnetic induction, and other electromagnetic physical laws into neural networks; Design a multi-layer encoder-decoder architecture based on the training structure of a large model; Furthermore, during network training, physical constraints and data loss terms are combined to optimize model parameters.

5. The method according to claim 4, characterized in that, The physical constraints include: Basic layer constraints are used to ensure the divergence-free condition of the electromagnetic field; Application layer constraints are used to constrain motor characteristic formulas such as torque and electric power. Engineering constraints are used to limit efficiency, temperature rise, and the range of materials used. The constraint terms at each level are adaptively weighted and adjusted based on the loss convergence.

6. The method according to claim 1, characterized in that, The key performance indicators of the motor based on the optimized network output decoding include: extracting the torque density, power density and loss characteristics of the motor from the high-dimensional feature distribution of the physical information neural network output, and optimizing the model parameters end-to-end through the joint loss function to obtain a motor design scheme that meets the predetermined performance indicators.

7. The method according to claim 6, characterized in that, The joint loss function includes data loss, physical loss, and boundary loss terms, and the generalization ability of the model is gradually optimized through a multi-stage progressive training strategy.

8. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-7.

Images