A motor digital twin modeling method and device based on multi-field coupling PINN

Abstract

A digital twin modeling method and device for motors based on multi-field coupled PINN (Partial Differential Equations) belongs to the field of intelligent manufacturing and advanced control technology for motors. Its key features include: acquiring stator current, voltage, rotor position, and vibration signals of the motor to construct a full-domain multimodal sensing layer; constructing a multi-field coupled PINN model based on automatic differentiation; and embedding the multi-field partial differential equations (PDEs) inside the motor as prior knowledge into the network structure of the PINN model. By utilizing PINN technology, the physical laws of PDEs are integrated into the neural network, ensuring both millisecond-level inference speed and physical consistency in areas lacking sensor data, achieving both high fidelity and real-time performance. It enables precise virtual sensing of unmeasurable rotor permanent magnet temperature and local magnetic flux density saturation within the motor, achieving full-domain observability. Through edge-cloud collaboration, it can automatically evolve as the physical entities of the motor age (e.g., resistance changes, magnet demagnetization), maintaining model accuracy throughout its entire lifecycle.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and advanced control technology of motors, and particularly relates to a method and device for digital twin modeling of motors based on multi-field coupling PINN. Background Technology

[0003] Traditional motor monitoring methods, when employing digital twin technology, primarily rely on external sensors (such as current, voltage, and encoders), making it difficult to directly obtain crucial information such as the internal temperature field distribution, magnetic saturation state, and local stress of the motor. This includes the following aspects: 1. The contradiction between model accuracy and real-time performance: Models based on finite element analysis (FEA) have high accuracy but consume huge amounts of computation time, which cannot meet the needs of real-time interaction; although deep learning models based on pure data-driven methods have fast inference, they lack physical interpretability and have poor generalization ability.

[0004] 2. Multi-field separation: Existing twins often only focus on a single electromagnetic or thermal property, ignoring the strong coupling effect of the four fields of electricity, magnetism, heat and force.

[0005] 3. Delay in virtual-physical interaction: Under complex and changing working conditions, cloud transmission delays prevent the virtual model from providing real-time feedback of control commands to the physical entity.

[0006] It is evident that existing fault-tolerant six-phase permanent magnet synchronous motor digital twin systems suffer from problems such as high computational load leading to poor real-time performance in high-fidelity models, difficulty in modeling multi-field coupling, and unreliable predictions due to the lack of physical constraints in purely data-driven models. Summary of the Invention

[0007] This invention aims to solve the above problems and provides a method and device for digital twin modeling of motors based on multi-field coupling PINN. Based on the Physical Information Neural Network (PINN) combined with edge computing and cloud computing architecture, it realizes multi-field state monitoring, prediction and real-time virtual-real interaction of fault-tolerant six-phase permanent magnet synchronous motor throughout its entire life cycle.

[0008] In a first aspect, the present invention provides a method for digital twin modeling of motors based on multi-field coupled PINN, comprising the following steps: Step 1: Collect stator current, voltage, rotor position, and vibration signals of the motor to establish a high-dimensional time series input vector; construct a global multimodal sensing layer. Step 2: Construct a multi-field coupled PINN model based on automatic differentiation; Step 3: Embedding physical constraints and loss function design; embedding the multi-field partial differential equations (PDEs) inside the motor as prior knowledge into the network structure of the PINN model; specifically including: Introducing vector space decoupling theory; considering the complex mutual inductance effects between the six-phase windings and the nonlinear characteristics caused by magnetic circuit saturation, the stator voltage vector u is defined. 6s With current vector i 6s Dynamic mapping relationship: ; Among them, u 6s =[u A1 ,u B1 ,u C1 ,u A2 ,u B2 ,u C2 ] T R is a six-terminal voltage input vector. s =diag(R) s R s , ..., R s ) is a 6×6 stator resistor matrix; L 6s (θ) e ) represents the inductance matrix; e 6s θ represents the back electromotive force component in the harmonic subspace; e The electric angle is the motor angle.

[0009] Explicitly model the time-varying fully coupled inductor matrix in PINN. Construct a physical constraint layer of multi-subspace orthogonal projection: using the 6×6 dimensional VSD transformation matrix TVSD, the variables in the six-phase stator natural coordinate system are mapped to three orthogonal subspaces, and a hierarchical dynamic residual constraint is constructed in PINN; Improved PINN Loss Function: The total PINN physical loss function is reconstructed into a multi-subspace weighted form: ; Among them, R fund For electromechanical energy conversion subspace residuals; R harmonic For harmonic loss subspace residuals; R mech The electromagnetic torque is the coupling residual between the equation of motion and the equation of motion; R thermal λ1 is the residual of the thermal equation; λ2 is the weighting coefficient of the fundamental component of the electromagnetic field; λ3 is the weighting coefficient of the mechanical equation; and λ4 is the weighting coefficient of the heat conduction.

[0010] By introducing a weighting coefficient λ2, the model's ability to learn the harmonic behavior of the xy subspace is enhanced, thus solving the technical problem of predicting motor overheating failure under non-sinusoidal power supply using traditional methods.

[0011] Step 4: Train the completed PINN model to optimize the parameters.

[0012] The digital twin modeling method for motors based on multi-field coupled PINN described in this invention, wherein the inductance matrix in step 3 consists of two self-inductance sub-matrices and two mutual inductance sub-matrices, accurately describes the leakage magnetic coupling and main magnetic flux coupling between windings: ; Among them, L 6s (θ) e ) represents a double three-phase symmetrical motor inductance matrix; L 11 For the first set of self-inductance matrix between windings; L 12 L is the mutual inductance matrix between the first set of windings and the second set of windings; 21 L is the mutual inductance matrix between the second set of windings and the first set of windings. 22 This is the self-inductance matrix between the second set of windings. Similarly, components with the same subscript represent self-inductance, and components with different subscripts represent mutual inductance from front to back.

[0013] Furthermore, in the motor digital twin modeling method based on multi-field coupled PINN described in this invention, the hierarchical dynamic residual constraint in step 3 includes: Electromechanical energy conversion subspace (dq plane) residual R fund The subspace governs the torque output and fundamental flux linkage characteristics of the dominant motor. PINN satisfies the following voltage balance equation residual minimization: ; Where, ψ f For permanent magnet flux linkage, L d L q For fundamental frequency inductance; R d R q For the electromechanical conversion residuals of the d and q axes; u d u q For d-axis and q-axis voltages; i d i q R represents the d-axis and q-axis currents. s ω is the stator resistance of the motor; e This is the angular frequency of the motor.

[0014] Harmonic loss subspace (xy plane) residual R harmonic The current component in the xy plane does not contribute to the average torque, but it generates a large amount of copper and iron losses, directly affecting the motor temperature rise. Traditional models often ignore this. By explicitly incorporating the current component in the xy plane into the physical constraints of PINN, high-precision heat source prediction can be achieved. ; Among them, L ls For stator leakage inductance; R x R y The residuals in the subspace of harmonic losses along the x and y axes; u x uy Let i be the voltage along the x and y axes. x i y R represents the current along the x and y axes. s This is the stator resistance of the motor.

[0015] By minimizing R x and R y Digital twins can accurately capture high-order harmonic losses caused by dead-zone effects and non-sinusoidal magnetic field distributions. The electromagnetic torque is coupled with the residual R of the motion equation. mech Based on VSD theory, electromagnetic torque T e Determined solely by the dq subspace variables, PINN satisfies the following kinematic constraints: ; Among them, P n Where is the pole pair number; J is the moment of inertia; T is the moment of inertia. L ω is the load torque. m Where is the mechanical angular velocity, and B is the damping coefficient.

[0016] Furthermore, the digital twin modeling method for motors based on multi-field coupled PINN described in this invention includes the following PINN model: Input variables include time t, the motor's spatial coordinates (x, y, z), and motor operating parameters such as voltage and current. Output variables, including flux linkage ψ and torque T e ; The neural network architecture includes an input layer, hidden layers, and an output layer. The input layer receives input variables. The hidden layer contains multiple hidden layers, each using a non-linear activation function (such as ReLU). The output layer outputs the motor's state variables (such as flux linkage and torque).

[0017] Furthermore, in the motor digital twin modeling method based on multi-field coupling PINN described in this invention, a composite objective function L(θ) is designed to include observation data loss, physical residual loss and boundary condition loss during model training; ; Where θ represents the parameters of the neural network, which are the objects to be optimized; N u The number of known observation data; The spatiotemporal coordinates of the known data; The physical field values ​​predicted by the neural network at a given spatiotemporal point; In order to be in The actual measured value at N; f The number of configuration points used for physical constraints; Spatiotemporal coordinates of points configured for physical constraints; For the physical residual term; λ bc These are the weighting coefficients for the boundary condition loss term; The first term of the composite objective function is the data-driven loss, based on N. u sparse observation points of sensors (t) u x u The mean square error of ) The second term is physical information loss, which is determined by randomly sampling N samples across the entire spatiotemporal domain. f Each collocation point forces the network to satisfy the above f at these sensorless data points. thermal and f mag Physical constraints equal to zero; The third item L BC For Dirichlet or Neumann boundary condition constraints, the specific form of the boundary condition loss is usually written as: ; in, For points on the boundary, For the given boundary values.

[0018] Furthermore, the digital twin modeling method for motors based on multi-field coupled PINN described in this invention is characterized in that the model training includes the following process: Data acquisition involves collecting operational data from the motor's sensors and then preprocessing the data (normalization, noise reduction, etc.) to improve training effectiveness. Network initialization: Initialize the weights and biases of PINN, and set the learning rate; Forward propagation involves inputting the input variables into PINN and calculating the output variables. Loss calculation: Calculate the loss based on the loss function; Backpropagation calculates the gradient of the loss function with respect to the network parameters using automatic differentiation techniques, and updates the network parameters through an optimizer. Iterative training involves repeating the forward propagation, loss calculation, and back propagation processes until the loss function converges. Model validation involves using data outside the training set to validate the accuracy of PINN and check whether the model meets the physical constraints. Hyperparameter tuning involves adjusting the network structure (number of hidden layers, number of neurons, etc.) and training parameters (learning rate, weight parameters, etc.) to optimize model performance.

[0019] Secondly, the present invention provides a digital twin modeling system for motors based on multi-field coupling PINN, including a multi-modal information acquisition module, a multi-field coupling modeling module, a model constraint module, and a model training module; The multimodal information acquisition module is used to acquire motor stator current, voltage, rotor position, and vibration signals, establish a high-dimensional time series input vector, and construct a global multimodal sensing layer. The multi-field coupling modeling module is used to construct a multi-field coupling PINN model based on automatic differentiation; The model constraint module is used to embed physical constraints and loss function design; it embeds the multi-field partial differential equations (PDEs) inside the motor as prior knowledge into the network structure of the PINN model; specifically, it includes: Introducing vector space decoupling theory; defining stator voltage vector u 6s With current vector i 6s Dynamic mapping relationship: ; Among them, u 6s =[u A1 ,u B1 ,u C1 ,u A2 ,u B2 ,u C2 ] T R is a six-terminal voltage input vector. s =diag(R) s R s , ..., R s ) is a 6×6 stator resistor matrix; L 6s (θ) e ) represents the inductance matrix; e 6s θ represents the back electromotive force component in the harmonic subspace; e For the electric angle of the motor; Explicitly model the time-varying fully coupled inductor matrix in PINN; Constructing a physical constraint layer for orthogonal projection of multiple subspaces: Using the 6×6 dimensional VSD transformation matrix TVSD, the variables in the natural coordinate system of the six-phase stator are mapped to three orthogonal subspaces, and hierarchical dynamic residual constraints are constructed in PINN; Improved PINN Loss Function: The total PINN physical loss function is reconstructed into a multi-subspace weighted form: ; Among them, R fund For electromechanical energy conversion subspace residuals; R harmonic For harmonic loss subspace residuals; R mech The electromagnetic torque is the coupling residual between the equation of motion and the equation of motion; R thermal λ1 is the residual of the thermal equation; λ2 is the weighting coefficient of the fundamental component of the electromagnetic field; λ3 is the weighting coefficient of the mechanical equation; and λ4 is the weighting coefficient of the heat conduction.

[0020] By introducing a weighting coefficient λ2, the model's ability to learn the harmonic behavior of the xy subspace is enhanced, thereby solving the technical problem of predicting motor overheating failure under non-sinusoidal power supply using traditional methods. The model training module is used to train the designed PINN model.

[0021] Furthermore, in the multi-field coupled PINN-based digital twin modeling system for motors described in this invention, the inductance matrix is ​​composed of two self-inductance sub-matrices and two mutual inductance sub-matrices, accurately describing the leakage magnetic coupling and main magnetic flux coupling between windings: ; Among them, L 6s (θ) e ) represents a double three-phase symmetrical motor inductance matrix; L 11 For the first set of self-inductance matrix between windings; L 12 L is the mutual inductance matrix between the first set of windings and the second set of windings; 21 L is the mutual inductance matrix between the second set of windings and the first set of windings. 22 This is the self-inductance matrix between the second set of windings. Similarly, components with the same subscript represent self-inductance, and components with different subscripts represent mutual inductance from front to back.

[0022] Furthermore, in the multi-field coupled PINN-based digital twin modeling system for motors described in this invention, the hierarchical dynamic residual constraints include: Electromechanical energy conversion subspace (dq plane) residual R fund The subspace governs the torque output and fundamental flux linkage characteristics of the dominant motor. PINN satisfies the following voltage balance equation residual minimization: ; Where, ψ f For permanent magnet flux linkage, L d L q For fundamental frequency inductance; R d R q For the electromechanical conversion residuals of the d and q axes; u d u q For d-axis and q-axis voltages; i d i q R represents the d-axis and q-axis currents. s ω is the stator resistance of the motor; e This is the angular frequency of the motor.

[0023] Harmonic loss subspace (xy plane) residual R harmonicThe current component in the xy plane does not contribute to the average torque, but it generates a large amount of copper and iron losses, directly affecting the motor temperature rise. Traditional models often ignore this. By explicitly incorporating the current component in the xy plane into the physical constraints of PINN, high-precision heat source prediction can be achieved. ; Among them, L ls For stator leakage inductance; R x R y The residuals in the subspace of harmonic losses along the x and y axes; u x u y Let i be the voltage along the x and y axes. x i y R represents the current along the x and y axes. s This is the stator resistance of the motor.

[0024] Electromagnetic torque and the coupling residual R of the equation of motion mech Based on VSD theory, electromagnetic torque T e Determined solely by the dq subspace variables, PINN satisfies the following kinematic constraints: ; Among them, P n Let J be the pole logarithm, J be the moment of inertia, and T be the moment of inertia. L ω is the load torque. m Where is the mechanical angular velocity, and B is the damping coefficient.

[0025] Thirdly, the present invention provides a motor digital twin modeling device based on multi-field coupling PINN, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to implement the motor digital twin modeling method based on multi-field coupling PINN as described in the first aspect when the computer program is executed.

[0026] The digital twin modeling method and device for motors based on multi-field coupled PINN described in this invention integrates the physical laws of partial differential equations (PDEs) into a neural network using PINN technology. This ensures both millisecond-level inference speed and physical consistency in areas lacking sensor data, achieving both high fidelity and real-time performance. It enables precise virtual sensing of unmeasurable rotor permanent magnet temperature and local magnetic flux density saturation within the motor, achieving full-domain observability. Through edge-cloud collaboration, it can automatically evolve as the physical entities of the motor age (e.g., resistance changes, magnet demagnetization), maintaining model accuracy throughout its entire lifecycle. Detailed Implementation

[0027] The following examples provide a detailed description of the motor digital twin modeling method and apparatus based on multi-field coupling PINN described in this invention.

[0028] PINN is a neural network model that combines physical laws with deep learning. It embeds physical laws (such as partial differential equations and conservation laws) into the loss function of the neural network, enabling the model to learn not only from data but also from known physical laws. The core idea of ​​PINN is to use physical constraints to guide the training of the neural network, thereby improving the model's accuracy and generalization ability, especially in situations where data is scarce or noisy.

[0029] Example 1 This embodiment discloses a digital twin modeling method for motors based on multi-field coupled PINN, including the following steps: Step S11: Collect motor stator current, voltage, rotor position, and vibration signals to establish a high-dimensional time series input vector; construct a global multimodal sensing layer. Step S12: Construct a multi-field coupled PINN model based on automatic differentiation; in this embodiment, the PINN model includes: Input variables include time t, the motor's spatial coordinates (x, y, z), and motor operating parameters such as voltage and current. Output variables, including flux linkage ψ and torque T e ; The neural network architecture includes an input layer, hidden layers, and an output layer; the input layer is used to receive input variables; the hidden layer contains multiple hidden layers, each using a non-linear activation function; and the output layer is used to output the state variables of the motor.

[0030] Step 13: Embedding physical constraints and loss function design; embedding the multi-field partial differential equations (PDEs) inside the motor as prior knowledge into the network structure of the PINN model; specifically including: Introducing vector space decoupling theory; considering the complex mutual inductance effects between the six-phase windings and the nonlinear characteristics caused by magnetic circuit saturation, the stator voltage vector u is defined. 6s With current vector i 6s Dynamic mapping relationship: ; Among them, u 6s =[u A1 ,u B1 ,u C1 ,u A2 ,u B2 ,u C2 ] T R is a six-terminal voltage input vector. s =diag(R) s R s , ..., R s ) is a 6×6 stator resistor matrix; L 6s (θ)e () represents the inductance matrix; In PINN, a time-varying fully coupled inductor matrix is ​​explicitly modeled. In this embodiment, the inductor matrix consists of two self-inductance submatrices and two mutual inductance submatrices, accurately describing the leakage magnetic coupling and main flux coupling between the windings. .

[0031] A physical constraint layer for multi-subspace orthogonal projection is constructed: using a 6×6 dimensional VSD transformation matrix TVSD, the variables in the six-phase stator natural coordinate system are mapped to three orthogonal subspaces, and a hierarchical dynamic residual constraint is constructed in PINN; in this embodiment, the hierarchical dynamic residual constraint includes: Electromechanical energy conversion subspace (dq plane) residual R fund The subspace governs the torque output and fundamental flux linkage characteristics of the dominant motor. PINN satisfies the following voltage balance equation residual minimization: ; Where, ψ f For permanent magnet flux linkage, L d L q For fundamental frequency inductance; Harmonic loss subspace (xy plane) residual R harmonic The current component in the xy plane does not contribute to the average torque, but it generates a large amount of copper and iron losses, directly affecting the motor temperature rise. Traditional models often ignore this. By explicitly incorporating the current component in the xy plane into the physical constraints of PINN, high-precision heat source prediction can be achieved. ; Among them, L ls For stator leakage inductance; by minimizing R x and R y Digital twins can accurately capture high-order harmonic losses caused by dead zone effects and non-sinusoidal magnetic field distribution.

[0032] Electromagnetic torque and the coupling residual R of the equation of motion mech Based on VSD theory, electromagnetic torque T e Determined solely by the dq subspace variables, PINN satisfies the following kinematic constraints: ; Among them, P n Let J be the pole logarithm, J be the moment of inertia, and T be the moment of inertia. L This represents the load torque.

[0033] Improved PINN Loss Function: The total PINN physical loss function is reconstructed into a multi-subspace weighted form: ; Among them, R fund For electromechanical energy conversion subspace residuals; R harmonic For harmonic loss subspace residuals; R mech λ1 represents the coupling residual between electromagnetic torque and the equation of motion; λ2 is the weighting coefficient. By introducing the weighting coefficient λ2, the model's ability to learn the harmonic behavior of the xy subspace is enhanced, thus solving the technical problem of predicting motor overheating failure under non-sinusoidal power supply using traditional methods.

[0034] Step S14: Train the designed PINN model to optimize the parameters. In this embodiment, a composite objective function L(θ) is designed to include observation data loss, physical residual loss, and boundary condition loss for model training. ; Where θ represents the parameters of the neural network, which are the objects to be optimized; N u The number of known observation data; The spatiotemporal coordinates of the known data; The physical field values ​​predicted by the neural network at a given spatiotemporal point; In order to be in The actual measured value at N; f The number of configuration points used for physical constraints; Spatiotemporal coordinates of points configured for physical constraints; For the physical residual term; λ bc These are the weighting coefficients for the boundary condition loss term; The first term of the composite objective function is the data-driven loss, based on N. u sparse observation points of sensors (t) u x u The mean square error of ) The second term is physical information loss, which is determined by randomly sampling N samples across the entire spatiotemporal domain. f Each collocation point forces the network to satisfy the above f at these sensorless data points. thermal and f mag Physical constraints equal to zero; The third item L BC For Dirichlet or Neumann boundary condition constraints, the specific form of the boundary condition loss is usually written as: ; in, For points on the boundary, For the given boundary values.

[0035] The model training includes the following process: Data acquisition involves collecting operational data from the motor's sensors and then preprocessing the data (normalization, noise reduction, etc.) to improve training effectiveness. Network initialization: Initialize the weights and biases of PINN, and set the learning rate; Forward propagation involves inputting the input variables into PINN and calculating the output variables. Loss calculation: Calculate the loss based on the loss function; Backpropagation calculates the gradient of the loss function with respect to the network parameters using automatic differentiation techniques, and updates the network parameters through an optimizer. Iterative training involves repeating the forward propagation, loss calculation, and back propagation processes until the loss function converges. Model validation involves using data outside the training set to validate the accuracy of PINN and check whether the model meets the physical constraints. Hyperparameter tuning involves adjusting the network structure (number of hidden layers, number of neurons, etc.) and training parameters (learning rate, weight parameters, etc.) to optimize model performance.

[0036] Step S15, Deployment and Optimization: Deploy the trained PINN model into the digital twin system of the six-phase permanent magnet synchronous motor, and continuously optimize the model based on real-time operating data to improve model accuracy and robustness.

[0037] Example 2 This embodiment discloses a digital twin modeling system for motors based on multi-field coupling PINN, including a multi-modal information acquisition module, a multi-field coupling modeling module, a model constraint module, and a model training module; The multimodal information acquisition module is used to acquire motor stator current, voltage, rotor position, and vibration signals, establish a high-dimensional time series input vector, and construct a global multimodal sensing layer. The multi-field coupling modeling module is used to construct a multi-field coupling PINN model based on automatic differentiation; The model constraint module is used to embed physical constraints and loss function design; it embeds the multi-field partial differential equations (PDEs) inside the motor as prior knowledge into the network structure of the PINN model; specifically, it includes: Introducing vector space decoupling theory; defining stator voltage vector u 6s With current vector i 6s Dynamic mapping relationship: ; Among them, u 6s =[u A1 ,u B1 ,u C1 ,u A2 ,u B2 ,uC2 ] T R is a six-terminal voltage input vector. s =diag(R) s R s , ..., R s () is a 6×6 stator resistor matrix; Explicitly model the time-varying fully coupled inductor matrix in PINN; Constructing a physical constraint layer for orthogonal projection of multiple subspaces: Using the 6×6 dimensional VSD transformation matrix TVSD, the variables in the natural coordinate system of the six-phase stator are mapped to three orthogonal subspaces, and hierarchical dynamic residual constraints are constructed in PINN; Improved PINN Loss Function: The total PINN physical loss function is reconstructed into a multi-subspace weighted form: ; Wherein, λ2 is the weighting coefficient; by introducing the weighting coefficient λ2, the model's ability to learn the harmonic behavior of the xy subspace is enhanced, thereby solving the technical problem of predicting motor overheating failure under non-sinusoidal power supply using traditional methods. The model training module is used to train the designed PINN model.

[0038] The specific operation steps of the motor digital twin modeling system based on multi-field coupling PINN described in this embodiment are the same as those of the motor digital twin modeling method based on multi-field coupling PINN described in Embodiment 1 above, and will not be repeated here.

[0039] Example 3 This embodiment discloses a motor digital twin modeling device based on multi-field coupling PINN, including a memory and a processor; the memory is used to store a computer program; the processor is used to implement the motor digital twin modeling method based on multi-field coupling PINN as described in Embodiment 1 when the computer program is executed. The specific modeling method steps are the same as those in Embodiment 1, and will not be repeated here.

[0040] The computer described in this application embodiment can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. The computer-readable storage medium can be any usable medium that a computer can read, or a data storage device such as a server or data center that integrates one or more usable media. The usable medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital versatile optical disc (DVD)), or a semiconductor medium (e.g., solid-state drive (SSD)). The software formed by the computer's stored code can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other storage media that are mature in the art.

[0041] In the various embodiments of this application, the functional modules can be integrated into one processing unit or module, or each module can exist physically separately, or two or more modules can be integrated into one unit or module. In the above embodiments, they can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0042] The above description is merely a specific embodiment 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 method for digital twin modeling of motors based on multi-field coupled PINN, characterized in that... Includes the following steps: Step 1: Collect stator current, voltage, rotor position, and vibration signals of the motor to establish a high-dimensional time series input vector; construct a global multimodal sensing layer. Step 2: Construct a multi-field coupled PINN model based on automatic differentiation; Step 3: Embedding physical constraints and loss function design; embedding the multi-field partial differential equations inside the motor as prior knowledge into the network structure of the PINN model; specifically... include: Introducing vector space decoupling theory; defining stator voltage vector u 6s With current vector i 6s Dynamic mapping relationship: ； in, u 6s =[ u A1 , u B1 , u C1 , u A2 , u B2 , u C2 ] T For a six-terminal voltage input vector, R s = diag ( R s , R s , ..., R s () is a 6×6 stator resistor matrix; L 6s ( θ e () represents the inductance matrix; e 6s This is the back electromotive force component in the harmonic subspace; θ e For the electric angle of the motor; Explicitly model the time-varying fully coupled inductor matrix in PINN; Constructing a physical constraint layer for orthogonal projection of multiple subspaces: utilizing a 6×6 dimensional VSD transformation matrix T VSD The variables in the six-phase stator natural coordinate system are mapped to three orthogonal subspaces, and hierarchical dynamic residual constraints are constructed in PINN; Improved PINN Loss Function: The total PINN physical loss function is reconstructed into a multi-subspace weighted form: ； Among them, R fund For electromechanical energy conversion subspace residuals; R harmonic For harmonic loss subspace residuals; R mech The electromagnetic torque is the coupling residual between the equation of motion and the equation of motion; R thermal The residuals of the thermal equations; λ 1 represents the weighting coefficient of the fundamental component of the electromagnetic field; λ 2 represents the weighting coefficient for higher harmonics; λ 3 represents the weighting coefficients of the mechanical equations; λ 4 represents the heat conduction weighting coefficient; Step 4: Train the completed PINN model.

2. The method for digital twin modeling of motors based on multi-field coupled PINN according to claim 1, characterized in that: The inductance matrix described in step 3 consists of two self-inductance sub-matrices and two mutual inductance sub-matrices. ； in, L 6s ( θ e () is a double three-phase symmetrical motor inductance matrix; L 11 This is the self-inductance matrix between the first set of windings; L 12 This is the mutual inductance matrix between the first set of windings and the second set of windings; L 21 This is the mutual inductance matrix between the second set of windings and the first set of windings; L 22 This is the self-inductance matrix between the second set of windings.

3. The method for digital twin modeling of motors based on multi-field coupled PINN according to claim 1, characterized in that, The hierarchical dynamic residual constraints mentioned in step 3 include: Electromechanical energy conversion subspace residual R fund PINN satisfies the following voltage balance equation: ； in, ψ f It is a permanent magnet flux linkage. L d , L q For fundamental frequency inductance; R d 、R q for d, q Shaft electromechanical conversion residual; u d 、u q for d, q Shaft voltage; i d , i q for d, q shaft current; R s This refers to the stator resistance of the motor. ω e This refers to the angular frequency of the motor. Harmonic loss subspace residual R harmonic :Will xy The planar current component is explicitly incorporated into the physical constraints of PINN: ； in, L ls For stator leakage inductance; R x R y for x , y Axial harmonic loss subspace residual; u x , u y for x, y shaft voltage, i x , i y for x、 y shaft current, R s This refers to the stator resistance of the motor. Electromagnetic torque and the coupling residual of the equation of motion R mech PINN satisfies the following kinematic constraints: ； in, P n It is the extreme logarithm; J It is the moment of inertia; T L This is the load torque; ω m For mechanical angular velocity, B is the damping coefficient.

4. The method for digital twin modeling of motors based on multi-field coupled PINN according to claim 1, 2, or 3, characterized in that: The PINN model includes: Input variables, including time t Spatial coordinates of the motor ( x, y, z ), motor operating parameters; Output variables, including magnet links ψ Torque T e ; The neural network architecture includes an input layer, hidden layers, and an output layer; the input layer is used to receive input variables; the hidden layer contains multiple hidden layers, each using a non-linear activation function; and the output layer is used to output the state variables of the motor.

5. The method for digital twin modeling of motors based on multi-field coupled PINN according to claim 4, characterized in that: During model training, a composite objective function L( ) is designed, which includes observation data loss, physical residual loss, and boundary condition loss. θ ); ； in, θ The parameters of the neural network are the objects that need to be optimized. N u The number of known observation data; The spatiotemporal coordinates of the known data; The physical field values ​​predicted by the neural network at a given spatiotemporal point; In order to be in The actual measured value at the location; N f The number of configuration points used for physical constraints; Spatiotemporal coordinates of points configured for physical constraints; For physical residuals; λ bc These are the weighting coefficients for the boundary condition loss term; The first term of the composite objective function is the data-driven loss, based on N u Sparse observation points of sensors ( t u , x u The mean square error of ) The second term is physical information loss, which is determined by random sampling across the entire spatiotemporal domain. N f Each point is assigned a coordinate point, which forces the network to satisfy the above conditions at these sensorless data points. f thermal and f mag Physical constraints equal to zero; The third item L BC The boundary conditions are Dirichlet or Neumann constraints.

6. The method for digital twin modeling of motors based on multi-field coupled PINN according to claim 5, characterized in that, The model training includes the following process: Data acquisition involves collecting operating data from the motor's sensors and then preprocessing the data. Network initialization: Initialize the weights and biases of PINN, and set the learning rate; Forward propagation involves inputting the input variables into PINN and calculating the output variables. Loss calculation: Calculate the loss based on the loss function; Backpropagation calculates the gradient of the loss function with respect to the network parameters using automatic differentiation techniques, and updates the network parameters through an optimizer. Iterative training involves repeating the forward propagation, loss calculation, and back propagation processes until the loss function converges. Model validation involves using data outside the training set to validate the accuracy of PINN and check whether the model meets the physical constraints. Hyperparameter tuning involves adjusting the network structure and training parameters to optimize model performance.

7. A digital twin modeling system for motors based on multi-field coupled PINN, characterized in that... It includes a multimodal information acquisition module, a multi-field coupling modeling module, a model constraint module, and a model training module; The multimodal information acquisition module is used to acquire motor stator current, voltage, rotor position, and vibration signals, establish a high-dimensional time series input vector, and construct a global multimodal sensing layer. The multi-field coupling modeling module is used to construct a multi-field coupling PINN model based on automatic differentiation; The model constraint module is used to embed physical constraints and loss function design; it embeds the multi-field partial differential equations inside the motor as prior knowledge into the network structure of the PINN model; specifically... include: Introducing vector space decoupling theory; defining stator voltage vector u 6s With current vector i 6s Dynamic mapping relationship: ； in, u 6s =[ u A1 , u B1 , u C1 , u A2 , u B2 , u C2 ] T For a six-terminal voltage input vector, R s = diag ( R s , R s , ..., R s () is a 6×6 stator resistor matrix; L 6s ( θ e () represents the inductance matrix; e 6s This is the back electromotive force component in the harmonic subspace; θ e For the electric angle of the motor; Explicitly model the time-varying fully coupled inductor matrix in PINN; Constructing a physical constraint layer for orthogonal projection of multiple subspaces: utilizing a 6×6 dimensional VSD transformation matrix T VSD The variables in the six-phase stator natural coordinate system are mapped to three orthogonal subspaces, and hierarchical dynamic residual constraints are constructed in PINN; Improved PINN Loss Function: The total PINN physical loss function is reconstructed into a multi-subspace weighted form: ； Among them, R fund For electromechanical energy conversion subspace residuals; R harmonic For harmonic loss subspace residuals; R mech The electromagnetic torque is the coupling residual between the equation of motion and the equation of motion; R thermal The residuals of the thermal equations; λ 1 represents the weighting coefficient of the fundamental component of the electromagnetic field; λ 2 represents the weighting coefficient for higher harmonics; λ 3 represents the weighting coefficients of the mechanical equations; λ 4 represents the heat conduction weighting coefficient; The model training module is used to train the designed PINN model.

8. The motor digital twin modeling system based on multi-field coupled PINN according to claim 7, characterized in that: The inductance matrix consists of two self-inductance sub-matrices and two mutual inductance sub-matrices. ； in, L 6s ( θ e () is a double three-phase symmetrical motor inductance matrix; L 11 This is the self-inductance matrix between the first set of windings; L 12 This is the mutual inductance matrix between the first set of windings and the second set of windings; L 21 This is the mutual inductance matrix between the second set of windings and the first set of windings; L 22 This is the self-inductance matrix between the second set of windings.

9. The motor digital twin modeling system based on multi-field coupled PINN according to claim 7, characterized in that, The hierarchical dynamic residual constraints include: Electromechanical energy conversion subspace residual R fund PINN satisfies the following voltage balance equation: ； in, ψ f It is a permanent magnet flux linkage. L d , L q For fundamental frequency inductance; R d 、R q for d, q Shaft electromechanical conversion residual; u d 、u q for d, q Shaft voltage; i d , i q for d, q shaft current; R s This refers to the stator resistance of the motor. ω e This refers to the angular frequency of the motor. Harmonic loss subspace residual R harmonic :Will xy The planar current component is explicitly incorporated into the physical constraints of PINN: ； in, L ls For stator leakage inductance; R x R y for x , y Axial harmonic loss subspace residual; u x , u y for x, y shaft voltage, i x , i y for x、 y shaft current, R s This refers to the stator resistance of the motor. Electromagnetic torque and the coupling residual of the equation of motion R mech PINN satisfies the following kinematic constraints: ； in, P n For extreme logarithms, J For rotational inertia, T L This is the load torque; ω m For mechanical angular velocity, B is the damping coefficient.

10. A digital twin modeling device for motors based on multi-field coupled PINN, characterized in that: It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the motor digital twin modeling method based on multi-field coupling PINN as described in any one of claims 1-6 when the computer program is executed.

Images