Armored vehicle permanent magnet synchronous motor health monitoring model construction method and system based on reduced-order digital twin model

Abstract

The application discloses a kind of based on reduced-order digital twin model Armored vehicle permanent magnet synchronous motor health monitoring model construction method and system, method includes: based on the parameter of permanent magnet synchronous motor establishes electromagnetic-temperature coupling field finite element model, simulates the operating parameter of permanent magnet synchronous motor at preset speed;Based on the operating parameter of permanent magnet synchronous motor at preset speed constructs original feature set;Utilize principal component analysis to original feature set Dimension reduction and cumulative explained variance, obtain the matrix formed by principal component feature vector;Ordinary differential neural network is used to continuously solve ordinary differential equation by simulation obtained dimension reduction time series data, fit history state, solve network parameter's accompanying state in reverse, form permanent magnet synchronous motor digital twin reduced-order model;Remove the coupling relationship between physical system and preset initial value, avoid initial value problem to influence prediction effect, provide reliable model and experience support for permanent magnet synchronous motor health management operation.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology for motors, specifically relating to a method and system for constructing a health monitoring model for a permanent magnet synchronous motor of an armored vehicle based on a reduced-order digital twin model. Background Technology

[0002] A permanent magnet synchronous motor (PMSM) is a synchronous motor that uses permanent magnets as the magnetic field source. It has advantages such as low loss, high efficiency, high power density, good dynamic performance, low maintenance cost, small size, light weight, and reliable operation. It is widely used in industrial automation, electric vehicles, wind power generation and other fields.

[0003] The Finite Element Method (FEM) is a common method for simulating the dynamic performance of permanent magnet synchronous motors (PMSMs) under multiphysics conditions. While the FEM can provide high-precision electromagnetic and temperature field distributions, it is computationally time-consuming, memory-intensive, and resource-intensive. This disadvantage is particularly pronounced in transient or parametric studies, making it unsuitable for real-time state interaction and online state monitoring between the digital twin model and the physical entity of the PMSM.

[0004] To meet the basic requirements of virtual-real interaction between digital twin models and physical entities, it is necessary to reduce the degrees of freedom in the finite element model (FEM) and reduce the dimensionality of the high-dimensional data generated by the FEM simulation. Several reduced-order model (ROM) techniques already exist. Among them, Principal Component Analysis (PCA) is an effective ROM algorithm. PCA extracts the main dynamic features from the high-dimensional data such as current, air gap magnetic flux density, and electromagnetic torque obtained from the FEM simulation, reducing the high-dimensional system to a smaller feature space.

[0005] Neural Ordinary Differential Equations (Neural ODEs) are deep learning models that primarily solve ordinary differential equations. They can decouple the boundary conditions, initial conditions, and system state prediction processes of a physical system, making them suitable for processing continuous time-series data and serving as surrogate models for digital twin models of permanent magnet synchronous motors (PMSMs). The health state of a PMSM is characterized by state variables such as current time-frequency domain characteristics, torque ripple, vibration, and temperature. These parameters are nonlinear and dynamically changing time series. Neural ODEs possess excellent continuous modeling capabilities, exhibiting higher parameter efficiency than traditional layered neural networks, and can flexibly capture the continuous dynamic characteristics of physical systems.

[0006] In the prior art, Chinese Patent CN117491869A discloses a health status monitoring and control method and system for permanent magnet synchronous motors based on digital twins. It proposes a current prediction model based on a four-section Runge-Kutta solver, uses particle swarm optimization to optimize motor parameters, and inputs the optimal parameters into a deadbeat predictive motor model to achieve motor control. However, this scheme does not predict the fault state or operating temperature of the permanent magnet synchronous motor. Chinese Patent CN114492137A discloses a motor condition monitoring and prediction method based on finite element method and digital twins. One approach constructs a finite element coupled circuit combination mathematical model for motor state prediction. The overall technology is based on this model without any order reduction, resulting in significant computational resource consumption. Another approach, Chinese patent CN118535914A, provides a digital twin-based method for fault diagnosis of key equipment in a power supply system. This method constructs a fault behavior twin model for key equipment in the power supply system, generates a fault dataset, and builds and updates a fault state perception twin model based on the fault dataset and a physical information neural network. The updated fault state perception twin model is then used to diagnose faults in the test samples. This approach focuses on diagnosing and classifying faults that have already occurred, without addressing fault severity prediction. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a health monitoring method for permanent magnet synchronous motors based on a reduced-order digital twin model. First, a physical model of the permanent magnet synchronous motor is established based on the relevant parameters of the target motor. Then, multiphysics simulation of the permanent magnet synchronous motor is performed using the finite element method to generate an original dataset. Principal component analysis is performed on the dataset, and the number of principal components is determined using the cumulative explained variance method to obtain a low-dimensional matrix. An ordinary differential neural network is then trained using the reduced-dimensional data to track the health status of the permanent magnet synchronous motor stator windings, the degradation status of the permanent magnets, and the real-time temperature of the motor, ultimately achieving health status monitoring of the permanent magnet synchronous motor.

[0008] To achieve the above objectives, in a first aspect, the present invention provides a method for constructing a health monitoring model of a permanent magnet synchronous motor for armored vehicles based on a reduced-order digital twin model, comprising the following steps: establishing an electromagnetic-temperature coupled field finite element model based on the parameters of the permanent magnet synchronous motor, and using the electromagnetic-temperature coupled field finite element model to simulate the operating parameters of the permanent magnet synchronous motor at a preset speed;

[0009] An original feature set is constructed based on the operating parameters of the permanent magnet synchronous motor at a preset speed;

[0010] Principal component analysis is used to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors;

[0011] An ordinary differential neural network is used to continuously solve ordinary differential equations to fit historical states from the dimensionality-reduced time series data obtained from simulation, and then reversely solves the adjoint states of the network parameters to form a digital twin reduced-order model of the permanent magnet synchronous motor.

[0012] Furthermore, an electromagnetic-temperature coupled field finite element model is established based on the parameters of the permanent magnet synchronous motor. The electromagnetic-temperature coupled field finite element model is used to simulate the operating parameters of the permanent magnet synchronous motor at a preset speed, including:

[0013] Based on the design parameters of the target permanent magnet synchronous motor, a two-dimensional model of the permanent magnet synchronous motor is established in COMSOL software;

[0014] Electromagnetic field analysis of the motor was performed using the built-in AC / DC module of COMSOL software. The motor rotor parameters, stator parameters and cylindrical coordinates were set, the stator and rotor boundaries were defined, vector changes were defined, and the material properties of each part of the motor were set.

[0015] For different physical field boundaries and finite element analysis constraints, the coupling relationship is determined, a mesh is created and solved to obtain the three-phase current, back electromotive force, electromagnetic torque, motor stator temperature and motor rotor temperature at the preset speed.

[0016] Furthermore, when creating the mesh and solving the problem, the mesh is automatically created by COMSOL, the mesh density is determined according to the physical field settings, the mesh structure is manually adjusted, the automatically created mesh sequence is deleted, and a boundary layer mesh is set at the boundary between the stator and the rotor.

[0017] Furthermore, the original feature set constructed based on the operating parameters of the permanent magnet synchronous motor at a preset speed includes:

[0018] The operating parameters of a permanent magnet synchronous motor at a preset speed are simulated using the electromagnetic-temperature coupled field finite element model. These operating parameters include the three-phase current, three-phase back electromotive force, electromagnetic torque, stator temperature, and rotor temperature at the preset speed. The negative sequence current component of the simulated three-phase current signal, together with the three-phase back electromotive force, output electromagnetic torque signal, stator temperature, and rotor temperature, constitute the original feature set. The asymmetry of the negative sequence component of the three-phase current signal is used to characterize the degree of inter-turn short circuit in the stator winding, the amplitude of the three-phase back electromotive force is used to characterize the degree of demagnetization of the permanent magnets, and the electromagnetic torque characterizes the health status of the motor.

[0019] Furthermore, principal component analysis is used to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors, including:

[0020] The time series of the simulation parameters are organized into a matrix, with each column representing a feature and each row representing a time point.

[0021] For each feature, the mean is normalized to zero and the variance is normalized to obtain the standardized matrix X'.

[0022] Calculate the covariance matrix of the standardized matrix X', and perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​and eigenvectors;

[0023] For different monitoring targets of permanent magnet synchronous motors, the cumulative explained variance method is used to select an appropriate number of principal components for each monitoring target, calculate the variance contribution rate of each principal component, sum the variance contribution rates, and select the features with the largest contribution as the principal components of the corresponding monitoring targets; retain the eigenvectors corresponding to the principal components, project the original data onto the principal components, and obtain the matrix composed of the eigenvectors of the principal components.

[0024] Furthermore, an ordinary differential neural network is used to continuously solve ordinary differential equations to fit historical states from the reduced-dimensional time series data obtained from simulation, and then inversely solves the adjoint states of the network parameters to form a digital twin reduced-order model of the permanent magnet synchronous motor, including:

[0025] The differential form of the ordinary differential neural network is obtained by subdividing the network layers, and the hidden state of each layer is solved based on the initial state.

[0026] For different monitoring targets, the dimensionality-reduced data is organized into a form suitable for time series modeling, and input data is constructed according to the time window;

[0027] The ordinary differential neural network is trained using the input data. By solving the ordinary differential equations and fitting the measured data of the permanent magnet synchronous motor, the state variables at the termination time are solved. Then, the loss function calculated in the forward calculation is solved, and the adjoint state method is used to update the model parameters of the ordinary differential neural network to obtain the adjoint state with respect to the hidden state. Similarly, the adjoint state with respect to the parameters of the ordinary differential neural network and time is obtained. The global parameter gradient is obtained by iteratively solving in the reverse direction, and the parameters of the ordinary differential neural network are updated to form a digital twin reduced-order model of the permanent magnet synchronous motor.

[0028] Secondly, this invention provides a health monitoring method for permanent magnet synchronous motors based on a reduced-order digital twin model. The reduced-order digital twin model of the permanent magnet synchronous motor obtained by the model construction method specifically includes: acquiring three-phase current, three-phase back electromotive force, electromagnetic torque, and temperature data of the permanent magnet synchronous motor; inputting these data into an ordinary differential neural network model obtained by reducing the order of the multi-physics coupled finite element model of the permanent magnet synchronous motor, i.e., the reduced-order digital twin model of the permanent magnet synchronous motor; monitoring the health status within a set time period to determine whether there is a potential fault; setting an error threshold; when the error between the current or back electromotive force and the prediction of the ordinary differential neural network model exceeds the threshold, issuing an early warning of stator winding inter-turn short circuit fault, permanent magnet demagnetization fault, and overheating phenomenon; and displaying the degree of short circuit fault, permanent magnet demagnetization rate, and overheating degree according to the error magnitude, further forming an adjustment strategy.

[0029] On the other hand, the present invention provides a method for constructing a health monitoring model of a permanent magnet synchronous motor for armored vehicles based on a reduced-order digital twin model, including a finite element simulation module, a feature set acquisition module, a feature set processing module, and a fitting and solving module.

[0030] The finite element simulation module establishes an electromagnetic-temperature coupled field finite element model based on the parameters of the permanent magnet synchronous motor, and uses the electromagnetic-temperature coupled field finite element model to simulate the operating parameters of the permanent magnet synchronous motor at a preset speed.

[0031] The feature set acquisition module constructs an original feature set based on the operating parameters of the permanent magnet synchronous motor at a preset speed;

[0032] The feature set processing module uses principal component analysis to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors.

[0033] The fitting and solution module uses an ordinary differential neural network to continuously solve the ordinary differential equations to fit the historical state of the reduced-dimensional time series data obtained from the simulation, and then solves the adjoint state of the network parameters in reverse to form a digital twin reduced-order model of the permanent magnet synchronous motor.

[0034] The present invention can also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can realize the method for constructing a health monitoring model of an armored vehicle permanent magnet synchronous motor based on a reduced-order digital twin model and the method for monitoring the health of a permanent magnet synchronous motor based on a reduced-order digital twin model as described in the present invention.

[0035] Simultaneously, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, it can realize the method for constructing a health monitoring model of an armored vehicle permanent magnet synchronous motor based on a reduced-order digital twin model and the method for monitoring the health of a permanent magnet synchronous motor based on a reduced-order digital twin model, as described in this invention.

[0036] Compared with the prior art, the present invention has at least the following beneficial effects: It performs data dimensionality reduction on the established refined multi-physics coupled finite element model of the permanent magnet synchronous motor, trains an ordinary differential neural network using dimensionality-reduced simulation data, and uses this as a digital twin reduced-order model for the permanent magnet synchronous motor. This decouples the physical system from the preset initial values, avoiding the influence of initial value problems on the prediction effect. The digital twin reduced-order model based on the ordinary differential neural network is used to predict the degree of stator inter-turn short circuit, monitor the degree of permanent magnet demagnetization, detect temperature, and provide over-temperature warnings during the operation of the permanent magnet synchronous motor, providing reliable models and experience support for the health management and maintenance of the permanent magnet synchronous motor. Attached Figure Description

[0037] Figure 1 A flowchart illustrating a health monitoring method for permanent magnet synchronous motors based on digital twins;

[0038] Figure 2 Flowchart for finite element modeling;

[0039] Figure 3 This is a diagram illustrating the cumulative interpretation method.

[0040] Figure 4 This is a schematic diagram of the permanent magnet synchronous health monitoring process. Detailed Implementation

[0041] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] The performance of permanent magnet synchronous motors is affected by two main factors: electromagnetic and temperature. This invention establishes a finite element model coupling these two factors, which can more accurately simulate the behavior of a motor under different operating conditions, especially the operating parameters at a preset speed. This helps to understand the complex interactions within the motor. The simulated operating parameters are used to construct an original feature set, providing basic data for subsequent analysis and modeling. These original features contain key information about motor operation. Principal component analysis is used for dimensionality reduction and cumulative explained variance, which helps to screen out the key features that have the greatest impact on motor performance. This not only reduces computational complexity but also improves the model's predictive accuracy, allowing the model to focus more on the most representative features. Using ordinary differential neural networks, the model can capture the dynamic characteristics of the data by continuously solving ordinary differential equations, thereby more accurately simulating the actual operation of the motor. This allows the model to learn the laws governing the change of motor state over time, enabling advanced applications such as state prediction and fault diagnosis. The final digital twin reduced-order model combines the physical characteristics of the electromagnetic-temperature coupled field with the learning results from the motor operating data. This reduces model complexity while maintaining prediction accuracy, making it more suitable for applications such as real-time monitoring and predictive maintenance. It helps to detect potential problems in advance and intervene in a timely manner, thereby improving the reliability and service life of the motor.

[0043] Based on the design parameters of the target permanent magnet synchronous motor, a two-dimensional model of the permanent magnet synchronous motor is established in COMSOL software. Ignoring the axial variation of the internal electromagnetic field, the two-dimensional model of the permanent magnet synchronous motor is used as the analysis object to effectively analyze the motor's operating state. The finite element modeling process is as follows: Figure 2 As shown.

[0044] The AC / DC module built into the COMSOL software is selected to perform electromagnetic field analysis on the motor. The transient solution of the AC / DC module is used to obtain the time sequence information of the two-dimensional electromagnetic field during motor operation, and to simulate multiple electromagnetic field quantities that are inconvenient to measure (such as the magnetic field strength in the air gap and the loss density in the magnet). First, the steady-state study of the motor is carried out, and then the transient study is carried out on the basis of this.

[0045] First, define the parameters, including defining variables, setting functions, and setting material parameters. Define the number of poles, rotor diameter, and magnet height of the motor rotor; define the number of stator slots and slot winding type; define cylindrical coordinates for later defining the motor speed; set the stator and rotor boundary division; and define vector transformations for later viewing state values ​​in a specific direction. Define the material properties of each part of the motor, adding Air, Soft Iron (Without Losses), N50, Aluminum, Structural Steel, and Copper materials to the model tree. Then, assign corresponding materials to the air gap, permanent magnet, coil, and iron core, and set relevant parameters such as conductivity, relative permittivity, permeability, magnetic coercivity, loss coefficient, and specific heat capacity to complete the material settings.

[0046] After importing the model and defining the parameters, perform geometry and physics settings. First, set up the electromagnetic field by applying finite element analysis constraints and setting magnetic flux conservation conditions for air and iron core. Define the magnetization direction and intensity of the permanent magnet. Set Ampere's law conditions for the stator core. Determine the number of turns of the stator winding coil based on the motor parameters, define the current density and voltage excitation for the winding, and set loss calculations for the coil and permanent magnet. Perform global torque calculation and define the motor speed.

[0047] After completing the electromagnetic field setup, the temperature field setup is performed. Material parameters such as thermal conductivity, specific heat capacity, and density are set in the heat conduction module; electromagnetic losses are coupled into the heat conduction module as a heat source, including copper losses in the windings, iron core losses, and permanent magnet losses; natural convection and radiation boundary conditions are set for the model to simulate the actual operating conditions of the motor.

[0048] After completing the geometric model and physics settings, mesh generation is performed. First, the mesh is automatically created using COMSOL, and the mesh density is determined based on the physics settings. Then, the mesh structure is manually adjusted, automatically created mesh sequences are deleted, and a boundary layer mesh is set at the stator-rotor interface to increase the boundary mesh density for calculating the electromagnetic effects within the air gap at the stator-rotor junction. After mesh generation, a solver is set up to solve for the state parameters, simulating the three-phase current I of the motor at a preset speed. a I b I c Three opposite electromotive forces e a e b e c Rotational speed n, electromagnetic torque T e The motor stator temperature Temp1 and the motor rotor temperature Temp2 are state variables.

[0049] The operating state of a permanent magnet synchronous motor can be characterized by state values ​​obtained through measurement and simulation, as well as time and frequency domain characteristics. The negative sequence current component of the three-phase current signal obtained from the simulation, together with the three-phase back electromotive force, the output electromagnetic torque signal, the stator temperature, and the rotor temperature, constitutes the original feature set, providing original data support for the next step of using principal component analysis to reduce the data dimensionality. The asymmetry of the negative sequence component of the three-phase current signal is mainly used to characterize the degree of inter-turn short circuit in the stator winding, the amplitude of the three-phase back electromotive force is mainly used to characterize the degree of demagnetization of the permanent magnets, and the output electromagnetic torque mainly characterizes the health status of the motor from the side.

[0050] The time series of different physical quantities obtained from the simulation are organized into a matrix form for principal component analysis. All data are organized into a matrix X, where each column represents a feature and each row represents a time point. Assuming the sample size is N and the number of features is M = 9 (three-phase current negative sequence component, three-phase back electromotive force, electromagnetic torque, stator temperature, rotor temperature), then:

[0051] First, the data is standardized by normalizing the mean and variance of each feature according to the following formula:

[0052]

[0053] Where μ is the mean of the feature and σ is the standard deviation of the feature.

[0054] The standardized data matrix X' has a mean of 0 and a variance of 1 for each column.

[0055] After standardization, calculate the covariance matrix C of the standardized matrix X':

[0056]

[0057] Each element of the covariance matrix represents the correlation between features.

[0058] Performing eigenvalue decomposition on the covariance matrix C yields eigenvalues ​​and eigenvectors:

[0059] C·v i =λ i ·v i

[0060] Where λ i These are eigenvalues, representing the magnitude of the variance explained by each principal component; v i It is an eigenvector, representing the direction of each principal component.

[0061] For different monitoring targets of permanent magnet synchronous motors (winding short-circuit degree, permanent magnet demagnetization degree, real-time temperature curve), the cumulative explained variance method is used to select an appropriate number of principal components for each monitoring target, normalize the eigenvalues, calculate the variance contribution rate corresponding to each principal component, and sum the variance contribution rates. Typically, the number of principal components that cumulatively explain the variance is selected to reach 85%. A schematic diagram of the cumulative explained variance method is shown below. Figure 3 As shown in the figure, there are 9 candidate features. By calculating the contribution of each feature, it can be seen from the figure that the cumulative contribution of the first four features exceeds 85%. Therefore, the first 4 features are selected as the principal components of a certain monitoring target.

[0062] By retaining the eigenvectors corresponding to the principal components, the original data is projected onto these principal components to obtain a dimensionality-reduced dataset. The dimensionality-reduced data matrix X is then calculated using the following formula. reduced :

[0063] X reduced =X'·P

[0064] Where P is a matrix composed of the eigenvectors of the first four principal components.

[0065] Ordinary differential neural networks (ODNs) continuously solve ordinary differential equations to fit historical states from dimensionality-reduced time series data obtained from simulations, and then solve the adjoint states of network parameters θ in reverse order to obtain the global gradient, thereby updating the ODN parameters. After that, the trained ODN is used to predict the state of permanent magnet synchronous motors and achieve health status monitoring.

[0066] The number of layers in an ordinary differential neural network can be considered as a continuous variable, which can be infinitely subdivided to obtain its differential form:

[0067]

[0068] Where, x t θ represents the state variables of the permanent magnet synchronous motor; θ represents the parameters to be learned in the network; and t represents time.

[0069] By pre-setting the initial states x0, t0, and t1, the hidden states of each layer can be solved:

[0070]

[0071] First, the dimensionality-reduced data needs to be organized into a form suitable for time series modeling, based on different monitoring targets. Input data is constructed according to a time window; assuming the window size is T, the input can be represented as:

[0072] X input =[X(tT),X(t-T+1),...,X(t)]

[0073] The output is:

[0074] Y output =X(t+1)

[0075] Where X(t) represents the dimensionality-reduced data.

[0076] The ordinary differential neural network is trained using the input data. After solving the ordinary differential equations to fit the measured data of the permanent magnet synchronous motor and completing the solution of the state variables at the termination time, the loss function L calculated in the forward computation is solved.

[0077]

[0078] The adjoint state method is used to update the model parameters of the ordinary differential neural network, where the adjoint state is defined as:

[0079]

[0080] In the formula, a(t) is the adjoint state of variable t with respect to the hidden state x.

[0081] Similarly, the adjoint state of the ordinary differential neural network with respect to the parameters θ and time t under variable t can be expressed as:

[0082]

[0083] Since the state quantity at the termination time is only related to the state θ of the ordinary differential neural network parameters at the previous time... n-1 Related, therefore set a θ When a(t) = 0, solve for a by reverse iteration. θ (t) and a t (t), thus obtaining the global parameter gradient:

[0084]

[0085] This updates the parameters θ of the ordinary differential neural network, forming a digital twin reduced-order model of the permanent magnet synchronous motor, and uses the updated reduced-order model to predict the trend of the permanent magnet synchronous motor's health status over a subsequent period.

[0086] A diagram illustrating the health monitoring process is shown below. Figure 4The three-phase current, three-phase back electromotive force, electromagnetic torque, and temperature data of the permanent magnet synchronous motor (PMSM) are obtained through testing on a test bench. These data are then input into an ordinary differential neural network model derived from a multi-physics coupled finite element model of the PMSM to monitor the short-circuit degree, permanent magnet demagnetization degree, and real-time temperature. By analyzing the health status over a period of time, the model determines whether there are potential faults. An error threshold is set. When the error between the current or back electromotive force and the prediction of the ordinary differential neural network model exceeds the threshold, it provides an early warning of stator winding inter-turn short-circuit faults, permanent magnet demagnetization faults, and overheating phenomena. The model displays the degree of short-circuit faults, permanent magnet demagnetization rates, and overheating degrees based on the magnitude of the error, in order to further formulate relevant adjustment strategies.

[0087] Based on the concept of the method, the present invention also provides a method for constructing a health monitoring model of a permanent magnet synchronous motor for armored vehicles based on a reduced-order digital twin model, including a finite element simulation module, a feature set acquisition module, a feature set processing module, and a fitting and solving module.

[0088] The finite element simulation module establishes an electromagnetic-temperature coupled field finite element model based on the parameters of the permanent magnet synchronous motor, and uses the electromagnetic-temperature coupled field finite element model to simulate the operating parameters of the permanent magnet synchronous motor at a preset speed.

[0089] The feature set acquisition module constructs an original feature set based on the operating parameters of the permanent magnet synchronous motor at a preset speed;

[0090] The feature set processing module uses principal component analysis to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors.

[0091] The fitting and solution module uses an ordinary differential neural network to continuously solve the ordinary differential equations to fit the historical state of the reduced-dimensional time series data obtained from the simulation, and then solves the adjoint state of the network parameters in reverse to form a digital twin reduced-order model of the permanent magnet synchronous motor.

[0092] On the other hand, the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement the method for constructing a health monitoring model of an armored vehicle permanent magnet synchronous motor based on a reduced-order digital twin model as described in the present invention, and can also implement a method for monitoring the health of a permanent magnet synchronous motor based on a reduced-order digital twin model.

[0093] The present invention can also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can realize the method for constructing a health monitoring model of an armored vehicle permanent magnet synchronous motor based on a reduced-order digital twin model as described in the present invention, and can also realize a method for monitoring the health of a permanent magnet synchronous motor based on a reduced-order digital twin model.

[0094] The computer device may be a laptop, a desktop computer, or a workstation.

[0095] The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or an off-the-shelf programmable gate array (FPGA).

[0096] The memory described in this invention can be an internal storage unit of a laptop, desktop computer, or workstation, such as memory or hard disk; or it can be an external storage unit, such as a portable hard disk or flash memory card.

[0097] Computer-readable storage media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer-readable storage media can include: read-only memory (ROM), random access memory (RAM), solid-state drives (SSDs), or optical discs, etc. Random access memory can include resistive random access memory (ReRAM) and dynamic random access memory (DRAM).

[0098] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model, characterized in that, Includes the following steps: An electromagnetic-temperature coupled field finite element model is established based on the parameters of the permanent magnet synchronous motor, and the operating parameters of the permanent magnet synchronous motor at a preset speed are simulated using the electromagnetic-temperature coupled field finite element model. An original feature set is constructed based on the operating parameters of the permanent magnet synchronous motor at a preset speed; Principal component analysis is used to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors; An ordinary differential neural network is used to continuously solve ordinary differential equations to fit historical states from the dimensionality-reduced time series data obtained from simulation, and then reversely solves the adjoint states of the network parameters to form a digital twin reduced-order model of permanent magnet synchronous motor. The three-phase current, three-phase back electromotive force, electromagnetic torque and temperature data of permanent magnet synchronous motor are obtained and input into the ordinary differential neural network model obtained by reducing the order of the multi-physics field coupled finite element model of permanent magnet synchronous motor, that is, the digital twin reduced order model of permanent magnet synchronous motor, and the health status is monitored within a set time to determine whether there is a potential fault. An error threshold is set. When the error between the current or back electromotive force and the prediction of the ordinary differential neural network model exceeds the threshold, an early warning is issued for stator winding inter-turn short circuit faults, permanent magnet demagnetization faults, and overheating phenomena. The degree of short circuit fault, permanent magnet demagnetization rate, and overheating degree are displayed according to the error magnitude, and an adjustment strategy is further formed.

2. The method for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model according to claim 1, characterized in that, An electromagnetic-temperature coupled field finite element model is established based on the parameters of the permanent magnet synchronous motor. The electromagnetic-temperature coupled field finite element model is used to simulate the operating parameters of the permanent magnet synchronous motor at a preset speed, including: Based on the design parameters of the target permanent magnet synchronous motor, a two-dimensional model of the permanent magnet synchronous motor is established in COMSOL software; Electromagnetic field analysis of the motor was performed using the built-in AC / DC module of COMSOL software. The motor rotor parameters, stator parameters and cylindrical coordinates were set, the stator and rotor boundaries were defined, vector changes were defined, and the material properties of each part of the motor were set. For different physical field boundaries and finite element analysis constraints, the coupling relationship is determined, a mesh is created and solved to obtain the three-phase current, back electromotive force, electromagnetic torque, motor stator temperature and motor rotor temperature at the preset speed.

3. The method for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model according to claim 2, characterized in that, When creating the mesh and solving the problem, the mesh is automatically created by COMSOL. After determining the mesh density according to the physical field settings, the mesh structure is manually adjusted, the automatically created mesh sequence is deleted, and a boundary layer mesh is set at the boundary between the stator and the rotor.

4. The method for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model according to claim 1, characterized in that, The original feature set constructed based on the operating parameters of the permanent magnet synchronous motor at a preset speed includes: The operating parameters of a permanent magnet synchronous motor at a preset speed are simulated using the electromagnetic-temperature coupled field finite element model. These operating parameters include the three-phase current, three-phase back electromotive force, electromagnetic torque, stator temperature, and rotor temperature at the preset speed. The negative sequence current component of the simulated three-phase current signal, together with the three-phase back electromotive force, output electromagnetic torque signal, stator temperature, and rotor temperature, constitute the original feature set. The asymmetry of the negative sequence component of the three-phase current signal is used to characterize the degree of inter-turn short circuit in the stator winding, the amplitude of the three-phase back electromotive force is used to characterize the degree of demagnetization of the permanent magnets, and the electromagnetic torque characterizes the health status of the motor.

5. The method for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model according to claim 1, characterized in that, Principal component analysis (PCA) is used to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors, including: The time series of the simulation parameters are organized into a matrix, with each column representing a feature and each row representing a time point. For each feature, the mean is normalized to zero and the variance is normalized to obtain the standardized matrix. ; Calculate the normalized matrix The covariance matrix, for the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues ​​and eigenvectors; For different monitoring targets of permanent magnet synchronous motors, the cumulative explained variance method is used to select the number of principal components for each monitoring target, calculate the variance contribution rate corresponding to each principal component, sum the variance contribution rates, and select the features with the largest contribution as the principal components of the corresponding monitoring targets; retain the eigenvectors corresponding to the principal components, project the original data onto the principal components, and obtain the matrix composed of the eigenvectors of the principal components.

6. The method for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model according to claim 1, characterized in that, An ordinary differential neural network is used to continuously solve ordinary differential equations to fit historical states from dimensionality-reduced time series data obtained from simulation, and then inversely solves the adjoint states of the network parameters to form a digital twin reduced-order model of the permanent magnet synchronous motor, including: The differential form of the ordinary differential neural network is obtained by subdividing the network layers, and the hidden state of each layer is solved based on the initial state. For different monitoring targets, the dimensionality-reduced data is organized into a form suitable for time series modeling, and input data is constructed according to the time window; The ordinary differential neural network is trained using the input data. By solving the ordinary differential equations and fitting the measured data of the permanent magnet synchronous motor, the state variables at the termination time are solved. Then, the loss function calculated in the forward calculation is solved, and the adjoint state method is used to update the model parameters of the ordinary differential neural network to obtain the adjoint state with respect to the hidden state. Similarly, the adjoint state with respect to the parameters of the ordinary differential neural network and time is obtained. The global parameter gradient is obtained by iteratively solving in the reverse direction, and the parameters of the ordinary differential neural network are updated to form a digital twin reduced-order model of the permanent magnet synchronous motor.

7. A system for constructing a health monitoring model for permanent magnet synchronous motors of armored vehicles based on a reduced-order digital twin model, characterized in that, It includes a finite element simulation module, a feature set acquisition module, a feature set processing module, and a fitting and solution module; The finite element simulation module establishes an electromagnetic-temperature coupled field finite element model based on the parameters of the permanent magnet synchronous motor, and uses the electromagnetic-temperature coupled field finite element model to simulate the operating parameters of the permanent magnet synchronous motor at a preset speed. The feature set acquisition module constructs an original feature set based on the operating parameters of the permanent magnet synchronous motor at a preset speed; The feature set processing module uses principal component analysis to reduce the dimensionality of the original feature set and accumulate the explained variance, resulting in a matrix composed of principal component eigenvectors. The fitting and solution module uses an ordinary differential neural network to continuously solve the ordinary differential equations to fit the historical state of the reduced-dimensional time series data obtained from the simulation, and then solves the adjoint state of the network parameters in reverse to form a digital twin reduced-order model of the permanent magnet synchronous motor. The three-phase current, three-phase back electromotive force, electromagnetic torque and temperature data of permanent magnet synchronous motor are obtained and input into the ordinary differential neural network model obtained by reducing the order of the multi-physics field coupled finite element model of permanent magnet synchronous motor, that is, the digital twin reduced order model of permanent magnet synchronous motor, and the health status is monitored within a set time to determine whether there is a potential fault. An error threshold is set. When the error between the current or back electromotive force and the prediction of the ordinary differential neural network model exceeds the threshold, an early warning is issued for stator winding inter-turn short circuit faults, permanent magnet demagnetization faults, and overheating phenomena. The degree of short circuit fault, permanent magnet demagnetization rate, and overheating degree are displayed according to the error magnitude, and an adjustment strategy is further formed.

8. A computer device, characterized in that, It includes a processor and a memory, the memory being used to store a computer-executable program, the processor reading part or all of the computer-executable program from the memory and executing it, and when the processor executes part or all of the computer-executable program, it can implement the method for constructing a health monitoring model of an armored vehicle permanent magnet synchronous motor based on a reduced-order digital twin model or the method for monitoring the health of a permanent magnet synchronous motor based on a reduced-order digital twin model as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the method for constructing a health monitoring model for a permanent magnet synchronous motor of an armored vehicle based on a reduced-order digital twin model, as described in any one of claims 1-6.

Images