A digital twin visualization method based on static field reduced-order model

Abstract

The present invention belongs to the field of digital twin technology, and discloses a digital twin visualization method based on a static field reduction model, comprising the following steps: establishing a CAE simulation model; obtaining a number of training sample data sets; constructing a static field reduction model to obtain a corresponding static field reduction model; constructing a static field fault diagnosis model to obtain a corresponding static field fault diagnosis model; constructing a digital twin model of a digital entity; collecting a real-time monitoring data set; obtaining a real-time physical field data set and a real-time static field fault diagnosis result; and visually displaying the real-time physical field data set and the corresponding real-time static field fault diagnosis result on the digital twin model of the digital entity. The present invention solves the problems of large time cost investment, high hardware requirements, poor visualization effect and simple functions in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital twins, and in particular relates to a digital twin visualization method based on a static field reduced-order model. Background Art

[0002] Existing three-dimensional computer-aided engineering (CAE) simulation technology can calculate the distribution of physical quantities in space under given boundary conditions. For example, CAE simulation technology can be used to simulate the air velocity, pressure, and temperature distribution on the surface of aircraft and automobiles, FEA simulation can be used to simulate the stress and deformation distribution inside parts when they are subjected to force, and electromagnetic field finite element calculation can be used to simulate the electromagnetic field distribution of motors, circuit boards, antennas, etc.

[0003] The principle of this simulation is that by dividing the space into countless small blocks, it is easier to derive the basic physical formulas that each small block obeys. By combining the physical formulas obeyed by all the small blocks, a set of equations with dimensions of up to hundreds of millions can be formed. By solving this set of equations, the values ​​of all physical quantities in the physical formulas on each small block can be obtained. This makes it possible to calculate the distribution of physical quantities in space. At present, this technology has become a basic tool in the field of equipment design, playing a role in replacing product testing and greatly improving the speed, efficiency and complexity of product design.

[0004] In the field of product design, building a digital twin model based on CAE simulation technology allows product users to obtain the distribution of physical quantities inside the product in real time while using the product, thereby laying the foundation for achieving further goals such as status monitoring, life prediction, and fault warning, and has become an important research direction.

[0005] Defects of the existing technology:

[0006] In the field of product design, existing digital twin models often use traditional CAE simulation technology. Since each calculation requires solving a very large set of equations, CAE simulation takes a long time. Generally speaking, a conventional CAE simulation in a project takes several hours to several days, which is time-consuming and requires high hardware. For a completed version of a product design, usually only the distribution of physical quantities under a few typical working conditions can be simulated. Designers cannot fully understand the product performance of this version of the design under any working condition, and the visualization effect of the digital twin model is poor. In addition, the existing digital twin model has simple functions, only has data visualization functions, and lacks corresponding data analysis capabilities. Summary of the invention

[0007] In order to solve the problems of high time cost investment, high hardware requirements, poor visualization effect and simple functions in the prior art, the present invention aims to provide a digital twin visualization method based on a static field reduced-order model.

[0008] The technical solution adopted by the present invention is:

[0009] A digital twin visualization method based on a static field reduced-order model comprises the following steps:

[0010] Define the model input and output involved in building the digital twin of the physical entity, and build a CAE simulation model of the physical entity based on the CAD 3D model and actual physical properties of the physical entity;

[0011] According to the model input and model output, an input data set is established, and based on the CAE simulation model, CAE simulation is performed on the input data set to obtain a number of corresponding training sample data sets;

[0012] According to a number of training sample data sets, a static field reduced-order model is constructed to obtain a corresponding static field reduced-order model;

[0013] According to several training sample data sets, a static field fault diagnosis model is constructed to obtain a corresponding static field fault diagnosis model;

[0014] The static field reduced-order model and the static field fault diagnosis model are added to the corresponding CAD three-dimensional model, and the model input is used as the corresponding monitoring target to obtain the digital twin model of the digital entity;

[0015] According to the model input of the digital twin model, the corresponding real-time monitoring data set is collected, and the real-time monitoring data is input into the digital twin model;

[0016] Using a static field reduced-order model of a digital entity, a real-time physical field data set corresponding to a real-time monitoring data set is obtained, and the real-time physical field data set is input into a static field fault diagnosis model to obtain a real-time static field fault diagnosis result corresponding to the real-time physical field data set;

[0017] The real-time physical field data set and the corresponding real-time static field fault diagnosis results are visualized on the digital twin model of the digital entity.

[0018] Further, several training sample data sets are collected from different CAE simulation cycles of the same physical entity, and each training sample data set is composed of a theoretical working range of the physical entity and an input quantity sampling point data set generated by a corresponding DOE method, and the training sample data set includes several training sample data;

[0019] The training sample data consists of the historical monitoring data and the corresponding historical physical field data of each data point;

[0020] Historical monitoring data corresponds to model input, and historical physical field data corresponds to model output.

[0021] Furthermore, according to the model input and the model output, an input data set is established, and based on the CAE simulation model, CAE simulation is performed on the input data set to obtain a number of corresponding training sample data sets, including the following steps:

[0022] According to the model input, collect some input data and establish the corresponding input data set;

[0023] According to the model output, based on the CAE simulation model, CAE simulation is performed on the input data set to obtain several corresponding training sample data sets.

[0024] Furthermore, a static field reduced-order model is constructed according to a number of training sample data sets to obtain a corresponding static field reduced-order model, including the following steps:

[0025] Preprocessing the training sample data set to obtain a first training sample data set after preprocessing;

[0026] Converting the historical physical field data set of the first training sample data set after preprocessing into a historical training field matrix;

[0027] Perform singular value decomposition on the historical training field matrix to obtain the corresponding historical eigenvector;

[0028] According to the contribution of the historical eigenvectors to the final physical field, the historical main eigenvectors are screened and the corresponding historical main eigenvector coefficients are generated;

[0029] The preprocessed first training sample data set is used as the first sub-model input, and the corresponding historical main eigenvector coefficient is used as the first sub-model output, and a scalar reduced-order sub-model is constructed to obtain a corresponding scalar reduced-order sub-model;

[0030] Generate a historical computational physics field data set according to the historical main eigenvectors and the corresponding historical main eigenvector coefficients;

[0031] Taking the historical main eigenvectors and the historical main eigenvector coefficients as the input of the second sub-model, taking the corresponding historical calculation physical field data set as the output of the second sub-model, constructing the output field synthesis calculation sub-model, and obtaining the corresponding output field synthesis calculation sub-model;

[0032] The scalar reduction-order sub-model and the output field synthesis calculation sub-model are integrated to obtain the corresponding static field reduction-order model.

[0033] Furthermore, the formula for singular value decomposition of the historical training field matrix is:

[0034] A=U×Σ×V T

[0035] Where A is the historical training field matrix; U is the historical left eigenvector; Σ is the eigenvalue matrix; V T is the transpose of the historical right eigenvector.

[0036] Furthermore, according to the contribution of the historical eigenvectors to the final physical field, the historical main eigenvectors are screened, and the corresponding historical main eigenvector coefficients are generated, including the following steps:

[0037] According to the representation of the contribution of the eigenvalues ​​to the final physical field in the eigenvalue matrix, the error estimation of the historical main eigenvectors is defined using the error evaluation method;

[0038] According to the error estimation, the error evaluation index of the historical main eigenvector is constructed;

[0039] According to the error evaluation index, the historical main characteristic order of the historical main characteristic vector is obtained;

[0040] According to the historical main feature order, the historical left feature vector of the historical feature vector is screened to obtain the corresponding historical main feature vector;

[0041] According to the historical main eigenvectors, the eigenvalue matrix of the historical eigenvectors and the transpose of the historical right eigenvectors are processed to generate the corresponding historical main eigenvector coefficients.

[0042] Furthermore, the formula for generating the historical calculation physical field data set is:

[0043] F=a1U1+a2U2+...+a s U s +...+a r U r

[0044] Where F is the historical calculation physical field data set; a s 、a r are the main eigenvector coefficients of the sth and rth histories respectively; U s , U r are the sth and rth historical main eigenvectors respectively; s is the main eigenvector indicator; r is the historical main characteristic order.

[0045] Furthermore, according to a number of training sample data sets, a static field fault diagnosis model is constructed to obtain a corresponding static field fault diagnosis model, including the following steps:

[0046] Extracting historical physical field data sets corresponding to a number of training sample data sets, and preprocessing and integrating all the historical physical field data sets to obtain a preprocessed second training sample data set containing fault diagnosis labels;

[0047] The second training sample data set after preprocessing is divided into a model training sample set and a model testing sample set in a ratio of 7:3;

[0048] Use some historical physical field point cloud data without fault diagnosis labels to pre-train the DBN network and build an initial static field fault diagnosis model;

[0049] The initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm, and the IFWA optimization algorithm is used to optimize according to the optimization target to obtain the optimal initial network parameters of the DBN network;

[0050] According to the optimal initial network parameters of the DBN network, the network structure of the initial static field fault diagnosis model is optimized, and the model training sample set is input to perform optimization training to obtain the optimized static field fault diagnosis model construction;

[0051] Input the model test sample set, build the optimized static field fault diagnosis model, perform model testing, and obtain the second model accuracy;

[0052] If the second model accuracy is greater than the second model accuracy threshold, the optimal static field fault diagnosis model construction is output, otherwise, the optimization training continues.

[0053] Furthermore, the initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm, and according to the optimization target, the IFWA optimization algorithm is used to perform optimization to obtain the optimal initial network parameters of the DBN network, including the following steps:

[0054] The initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm;

[0055] Set the IFWA population parameters, maximum number of iterations and fitness function of the IFWA optimization algorithm, and use the optimization target as the position of the IFWA individuals in the IFWA population;

[0056] According to the IFWA population parameters, the Circle chaotic mapping sequence is used to initialize the IFWA population to obtain the initialized IFWA population;

[0057] According to the fitness function, calculate the fitness value of the IFWA individuals in the initialized IFWA population;

[0058] Get the explosion radius and number of sparks of each initial firework in the initial firework set;

[0059] According to the explosion radius and the number of sparks of each initial IFWA individual in the initialized IFWA population, fireworks are exploded to obtain an updated IFWA population;

[0060] Use the Gaussian mutation algorithm to perform Gaussian mutation on the initialized IFWA population to generate a Gaussian mutated IFWA population;

[0061] Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the initialized IFWA population to generate a dynamic reverse IFWA population;

[0062] Calculate the fitness values ​​of all IFWA individuals in the updated IFWA population, the Gaussian-mutated IFWA population, and the dynamically reversed IFWA population, and take the IFWA individual with the minimum fitness value as the optimal individual;

[0063] If the number of iterations reaches the threshold or the fitness value of the optimal individual meets the requirements, the optimal solution corresponding to the current optimal individual is output to obtain the optimal initial network parameters of the DBN network.

[0064] The beneficial effects of the present invention are:

[0065] The present invention provides a digital twin visualization method based on a static field reduced-order model. The static field reduced-order model is constructed through a large amount of CAE simulation data, and the traditional CAE simulation technology is improved. The real-time calculation of the physical problem can be realized according to the real-time physical field data set corresponding to the real-time monitoring data set at the cost of appropriate sacrifice of accuracy, and the original calculation time of hours to days is shortened to milliseconds to seconds, which reduces the time cost investment and hardware requirements. The real-time physical field data set can intuitively display the physical field point cloud distribution of the physical entity, and simulate the physical quantity distribution of the physical entity under any working condition based on the digital twin model. The designer can fully understand the product performance of this version design under any working condition, and improve the visualization effect of the digital twin model. The static field fault diagnosis model is added to perform real-time static field fault diagnosis on the real-time physical field data set, which can timely discover the potential fault risk of the physical entity based on the static field, has corresponding data analysis capabilities, and enhances the functionality of the digital twin model.

[0066] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flowchart of the digital twin visualization method based on the static field reduced-order model in the present invention. DETAILED DESCRIPTION

[0068] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0069] Embodiment 1:

[0070] like Figure 1 As shown, this embodiment provides a digital twin visualization method based on a static field reduced-order model, comprising the following steps:

[0071] S1: Define the model input and output involved in building the digital twin of the physical entity, and build a CAE simulation model of the physical entity based on the CAD 3D model and actual physical properties of the physical entity;

[0072] The method for constructing a CAD three-dimensional model and a CAE simulation model of a physical entity includes the following steps:

[0073] A-1: Use CAD software to perform 3D modeling on the physical entity and obtain the corresponding CAD 3D model;

[0074] A-2: Import the built CAD 3D model into CAE software and perform meshing on the 3D CAD software; meshing is both a pre-processing step for CAE and a method for generating geometric point cloud files for digital twins;

[0075] A-3: Set boundary conditions for the CAE simulation model according to the type of digital twin input;

[0076] A-4: Perform trial calculations on the simulation model with the boundary conditions set. And set the output data of the calculation results according to the output physical field type of the digital twin, thus completing the setting of the entire CAE simulation process;

[0077] S2: According to the model input and model output, an input data set is established, and based on the CAE simulation model, CAE simulation is performed on the input data set to obtain a number of corresponding training sample data sets, including the following steps:

[0078] S2-1: According to the model input, collect some input data and establish the corresponding input data set;

[0079] S2-2: According to the model output, based on the CAE simulation model, CAE simulation is performed on the input data set to obtain several corresponding training sample data sets;

[0080] In this embodiment, the physical entity adopts a motor. When the motor is running, it is difficult to obtain the temperature information on its rotor because the gap between the rotor and the stator is too small and the rotor rotates at a high speed, so it is impossible to deploy a temperature sensor within the cost range. In this embodiment, a digital twin model of the motor is constructed. The model follows the physical laws related to the motor rotor temperature, defines the model input as the sensor information obtainable from the outside of the motor, such as the rotation speed, ambient temperature, current, and voltage, and defines the model output as the temperature field data of the rotor. The field data represents the distribution of temperature on the rotor and is a complete description of the temperature field.

[0081] Several training sample data sets are collected from different CAE simulation cycles of the same physical entity, and each training sample data set is composed of a theoretical working range of the physical entity and an input quantity sampling point data set generated by a corresponding design of experiment (DOE) method, and the training sample data set includes several training sample data;

[0082] The training sample data consists of the historical monitoring data and the corresponding historical physical field data of each data point;

[0083] The historical monitoring data corresponds to the model input, and the historical physical field data corresponds to the model output;

[0084] S3: Based on a number of training sample data sets, a static field reduced order model is constructed to obtain a corresponding static field reduced order model, including the following steps:

[0085] S3-1: preprocessing the training sample data set to obtain a first training sample data set after preprocessing;

[0086] Preprocessing includes removing erroneous data, screening duplicate data, and filling in missing data for each training sample data set in turn;

[0087] S3-2: converting the historical physical field data set of the first training sample data set after preprocessing into a historical training field matrix;

[0088] The number of points in the historical monitoring data set is m, and the number of points in the historical physical field data set is n. The dimension of the corresponding historical training field matrix is ​​m×n, and the number of rows in a rectangular matrix is ​​much larger than the number of columns.

[0089] S3-3: Perform singular value decomposition on the historical training ground matrix to obtain the corresponding historical eigenvector;

[0090] The formula for singular value decomposition of the historical training field matrix is:

[0091] A=U×Σ×V T

[0092] Where A is the historical training field matrix; U is the historical left eigenvector; Σ is the eigenvalue matrix; V T is the transpose of the historical right eigenvector; U is an m×n unitary matrix, V is an n×n unitary matrix, Σ is an m×n matrix, and only the diagonal (row number = column number) has values, and the other positions are zero. The values ​​on the diagonal are the eigenvalues. According to the properties of singular value decomposition, the eigenvalues ​​gradually decrease from the upper left corner to the lower right corner, close to but not equal to zero;

[0093] Transposing both sides of the equation, we get:

[0094] A T =VΣ T U T

[0095] In the formula, A T is the transpose of the historical training field matrix; Σ T is the transpose of the eigenvalue matrix; U T is the transpose of the historical left eigenvector; V is the historical right eigenvector;

[0096] Then multiply A by the transpose of A to get:

[0097] A T A=VΣ T U T UΣV T =VΣ 2 V T

[0098] Since U is a unitary matrix, Σ 2 is a diagonal matrix. The values ​​on the diagonal are the squares of the eigenvalues ​​after singular value decomposition. T A is a square matrix. Through the above transformation, the singular value decomposition problem is transformed into the problem of finding the eigenvalues ​​and eigenvectors of the square matrix, which can be solved using the general method of linear algebra;

[0099] Get A T The square matrix composed of the eigenvectors of A, that is, V of the singular value decomposition, obtains the first eigenvalue, and takes the square root of the first eigenvalue to obtain the eigenvalue of the singular value decomposition;

[0100] U=A(ΣV T ) -1

[0101] This embodiment uses a special method to solve U instead of the traditional matrix theory singular value decomposition. In the original method, since Σ is an m×n matrix, (ΣV T ) -1It is an n×m matrix. The solved U is an m×m matrix. Since the dimension of m is extremely large, it will lead to unbearable computation and storage. However, since only the diagonal values ​​of Σ are not zero, the values ​​of all vectors of U starting from the n+1th column will eventually be multiplied by zero, which is useless. Therefore, when solving U, Σ can be artificially constructed into a diagonal square matrix, and then the solved U is an m×n matrix.

[0102] S3-4: According to the contribution of the historical eigenvectors to the final physical field, the historical main eigenvectors are screened and the corresponding historical main eigenvector coefficients are generated, including the following steps:

[0103] S3-4-1: Based on the characterization of the contribution of the eigenvalues ​​to the final physical field in the eigenvalue matrix, the error estimation of the historical main eigenvectors is defined using the error evaluation method;

[0104] The contribution of each independently changing physical field to the final physical field depends on the size of the corresponding eigenvalue in Σ. The eigenvalues ​​in Σ gradually approach zero from the upper left to the lower right. The corresponding column vectors in the U matrix, each column represents an independent physical field whose contribution to the final physical field decreases from left to right.

[0105] The formula for error estimation is:

[0106]

[0107] In the formula, ε r is the error estimate of the first r-order eigenvector; σ is the eigenvector; r is the historical main characteristic order; n is the number of points in the historical physical field data set;

[0108] S3-4-2: Based on the error estimation, construct the error evaluation index of the historical main characteristic vector, the formula is:

[0109] I r =logε r

[0110] In the formula, ε r is the error estimate of the first r-order eigenvectors; l r is the error evaluation index of the first r-order eigenvectors;

[0111] S3-4-3: Obtain the historical main characteristic order of the historical main characteristic vector according to the error evaluation index;

[0112] Get the evaluation index when r = 1 and r = n-1, take the average value, and you can get the median value I of the evaluation index m ; Starting from r = 2, as the value of r increases, the evaluation index is calculated until the evaluation index I r <I m, then the r value at this time is the historical main characteristic order that the reduced-order model needs to take;

[0113] Find the main historical characteristic order that begins to decline rapidly, and take the linear combination of the characteristic vectors of all previous orders as the approximate expression of the final physical field, which greatly reduces the amount of calculation while ensuring tolerable errors;

[0114] S3-4-4: According to the historical main feature order, the historical left feature vector of the historical feature vector is screened to obtain the corresponding historical main feature vector;

[0115] S3-4-5: Based on the historical main characteristic vector (U s |s=1,2,...,r), process the eigenvalue matrix of the historical eigenvector and the transpose of the historical right eigenvector to generate the corresponding historical main eigenvector coefficient (ΣV T =a s |s=1,2,...,r);

[0116] S3-5: using the preprocessed first training sample data set as the first sub-model input, using the corresponding historical main eigenvector coefficient as the first sub-model output, and constructing a scalar reduced-order sub-model to obtain a corresponding scalar reduced-order sub-model;

[0117] S3-6: Generate a historical computational physics field data set based on the historical main eigenvectors and the corresponding historical main eigenvector coefficients;

[0118] The formula for generating the historical calculation physics field data set is:

[0119] F=a1U1+a2U2+...+a s U s +...+a r U r

[0120] Where F is the historical calculation physical field data set; a s 、a r are the main eigenvector coefficients of the sth and rth histories respectively; U s , U r are the sth and rth historical main eigenvectors respectively; s is the main eigenvector indicator; r is the historical main eigenorder;

[0121] S3-7: Using the historical main eigenvectors and the historical main eigenvector coefficients as the second sub-model input, using the corresponding historical calculation physical field data set as the second sub-model output, constructing the output field synthesis calculation sub-model, and obtaining the corresponding output field synthesis calculation sub-model;

[0122] S3-8: Integrate the scalar reduced-order sub-model and the output field synthesis calculation sub-model to obtain the corresponding static field reduced-order model;

[0123] S4: Based on several training sample data sets, the Deep Belief Network (DBN)-Improved Fireworks Algorithm (IFWA) algorithm is used to construct a static field fault diagnosis model to obtain the corresponding static field fault diagnosis model, including the following steps:

[0124] S4-1: extracting historical physical field data sets corresponding to a number of training sample data sets, and preprocessing and integrating all the historical physical field data sets to obtain a preprocessed second training sample data set containing a fault diagnosis label;

[0125] Fault diagnosis labels include local low temperature overheating, local medium temperature overheating, local high temperature overheating, global low temperature overheating, global medium temperature overheating and global high temperature overheating of the rotor and stator of the motor;

[0126] S4-2: Divide the preprocessed second training sample data set into a model training sample set and a model testing sample set in a ratio of 7:3;

[0127] S4-3: Use some historical physical field point cloud data without fault diagnosis labels to pre-train the DBN network and build an initial static field fault diagnosis model;

[0128] S4-4: Taking the initial network parameters of the DBN network as the optimization target of the IFWA optimization algorithm, and using the IFWA optimization algorithm to optimize according to the optimization target to obtain the optimal initial network parameters of the DBN network, including the following steps:

[0129] S4-4-1: The initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm;

[0130] S4-4-2: Set the IFWA population parameters, maximum number of iterations and fitness function of the IFWA optimization algorithm, and use the optimization target as the position of the IFWA individuals in the IFWA population;

[0131] S4-4-3: According to the IFWA population parameters, the Circle chaotic mapping sequence is used to initialize the IFWA population to obtain the initialized IFWA population. The formula is:

[0132]

[0133] In the formula, q l' is the initial IFWA individual of Circle chaos mapping; q* l' is the randomly generated initial IFWA individual; l' is the indicator;

[0134] S4-4-4: According to the fitness function, calculate the fitness value of the IFWA individual in the initialized IFWA population. The formula is:

[0135]

[0136] In the formula, f(q l' ) is the initial IFWA individual q l' The fitness value; MSE is the prediction mean square error function; Y' l' , Y l' is the predicted value and the true value;

[0137] S4-4-5: Get the explosion radius and number of sparks of each initial firework in the initial firework set. The formula is:

[0138]

[0139] In the formula, S l' is the initial IFWA individual q l' The number of sparks; M' is a constant; f max is the maximum fitness value in the initialized IFWA population; f(q l' ) is the initial IFWA individual q l' The fitness value of ; τ is an infinitesimal constant;

[0140]

[0141] In the formula, R l' is the initial IFWA individual q l' Explosion radius; f is the explosion radius adjustment constant; min is the minimum fitness value in the initialized IFWA population;

[0142] S4-4-6: According to the explosion radius and number of sparks of each initial IFWA individual in the initialized IFWA population, fireworks are exploded to obtain an updated IFWA population. The formula is:

[0143] q' l' =q l' +S l' ×rand(-1,1)

[0144] In the formula, q' l' is the updated IFWA individual; rand(-1,1) is a random number between -1 and 1; q l' For the initial IFWA individuals;

[0145] S4-4-7: Use the Gaussian mutation algorithm to perform Gaussian mutation on the initialized IFWA population to generate a Gaussian mutated IFWA population. The formula is:

[0146] q" l' =q l' +S l' ×G(1,1)

[0147] In the formula, q" l' is an IFWA individual with Gaussian variation; G(1,1) is a random number with Gaussian distribution whose mean and variance are both 1;

[0148] S4-4-8: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the initialized IFWA population to generate a dynamic reverse IFWA population. The formula is:

[0149] q'" l' =γ(L max +L min )-q l'

[0150] In the formula, q'" l' is a dynamic reverse IFWA individual; X best is the positive candidate optimal solution; γ is the decreasing inertia coefficient, γ=0.9-0.5t / T; L max , L min are the maximum and minimum values ​​of the vector space respectively; t is the current iteration number; T is the maximum iteration number;

[0151] S4-4-9: Calculate the fitness values ​​of all IFWA individuals in the updated IFWA population, the Gaussian mutated IFWA population, and the dynamically reversed IFWA population, and take the IFWA individual with the minimum fitness value as the optimal individual;

[0152] S4-4-10: If the number of iterations reaches the threshold or the fitness value of the optimal individual meets the requirements, the optimal solution corresponding to the current optimal individual is output to obtain the optimal initial network parameters of the DBN network;

[0153] S4-5: According to the optimal initial network parameters of the DBN network, the network structure of the initial static field fault diagnosis model is optimized, and the model training sample set is input to perform optimization training to obtain the optimized static field fault diagnosis model construction;

[0154] S4-6: input the model test sample set, construct the optimized static field fault diagnosis model, perform model testing, and obtain the second model accuracy;

[0155] S4-7: If the second model accuracy is greater than the second model accuracy threshold, then outputting the optimal static field fault diagnosis model construction, otherwise, continuing the optimization training;

[0156] S5: adding the static field reduced-order model and the static field fault diagnosis model to the corresponding CAD three-dimensional model, and taking the model input as the corresponding monitoring target to obtain the digital twin model of the digital entity;

[0157] S6: According to the model input of the digital twin model, collect the corresponding real-time monitoring data set, and input the real-time monitoring data into the digital twin model; the real-time monitoring data set of this embodiment is data collected by sensors;

[0158] S7: using the static field reduced-order model of the digital entity, obtaining a real-time physical field data set corresponding to the real-time monitoring data set, and inputting the real-time physical field data set into the static field fault diagnosis model to obtain a real-time static field fault diagnosis result corresponding to the real-time physical field data set;

[0159] S8: Visualize the real-time physical field data set and the corresponding real-time static field fault diagnosis results on the digital twin model of the digital entity.

[0160] The present invention provides a digital twin visualization method based on a static field reduced-order model. The static field reduced-order model is constructed through a large amount of CAE simulation data, and the traditional CAE simulation technology is improved. The real-time calculation of the physical problem can be realized according to the real-time physical field data set corresponding to the real-time monitoring data set at the cost of appropriate sacrifice of accuracy, and the original calculation time of hours to days is shortened to milliseconds to seconds, which reduces the time cost investment and hardware requirements. The real-time physical field data set can intuitively display the physical field point cloud distribution of the physical entity, and simulate the physical quantity distribution of the physical entity under any working condition based on the digital twin model. The designer can fully understand the product performance of this version design under any working condition, and improve the visualization effect of the digital twin model. The static field fault diagnosis model is added to perform real-time static field fault diagnosis on the real-time physical field data set, which can timely discover the potential fault risk of the physical entity based on the static field, has corresponding data analysis capabilities, and enhances the functionality of the digital twin model.

[0161] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A digital twin visualization method based on a static field reduced-order model, characterized in that: The steps include: Define the model input and output involved in building the digital twin of the physical entity, and build a CAE simulation model of the physical entity based on the CAD 3D model and actual physical properties of the physical entity; According to the model input and model output, an input data set is established, and based on the CAE simulation model, CAE simulation is performed on the input data set to obtain a number of corresponding training sample data sets; According to a number of training sample data sets, a static field reduced-order model is constructed to obtain a corresponding static field reduced-order model; According to several training sample data sets, a static field fault diagnosis model is constructed to obtain a corresponding static field fault diagnosis model; The static field reduced-order model and the static field fault diagnosis model are added to the corresponding CAD three-dimensional model, and the model input is used as the corresponding monitoring target to obtain the digital twin model of the digital entity; According to the model input of the digital twin model, the corresponding real-time monitoring data set is collected, and the real-time monitoring data is input into the digital twin model; Using a static field reduced-order model of a digital entity, a real-time physical field data set corresponding to a real-time monitoring data set is obtained, and the real-time physical field data set is input into a static field fault diagnosis model to obtain a real-time static field fault diagnosis result corresponding to the real-time physical field data set; The real-time physical field data set and the corresponding real-time static field fault diagnosis results are visualized on the digital twin model of the digital entity.

2. The digital twin visualization method based on the static field reduced-order model according to claim 1, characterized in that: The plurality of training sample data sets are collected from different CAE simulation cycles of the same physical entity, and each training sample data set is composed of a theoretical working range of the physical entity and an input quantity sampling point data set generated by a corresponding DOE method, and the training sample data set includes a plurality of training sample data; The training sample data is composed of historical monitoring data and corresponding historical physical field data of each data point; The historical monitoring data corresponds to the model input, and the historical physical field data corresponds to the model output.

3. The digital twin visualization method based on the static field reduced-order model according to claim 2 is characterized in that: According to the model input and model output, an input data set is established, and based on the CAE simulation model, CAE simulation is performed on the input data set to obtain a number of corresponding training sample data sets, including the following steps: According to the model input, collect some input data and establish the corresponding input data set; According to the model output, based on the CAE simulation model, CAE simulation is performed on the input data set to obtain several corresponding training sample data sets.

4. The digital twin visualization method based on the static field reduced-order model according to claim 3 is characterized in that: According to several training sample data sets, a static field reduced order model is constructed to obtain a corresponding static field reduced order model, including the following steps: Preprocessing the training sample data set to obtain a first training sample data set after preprocessing; Converting the historical physical field data set of the first training sample data set after preprocessing into a historical training field matrix; Perform singular value decomposition on the historical training field matrix to obtain the corresponding historical eigenvector; According to the contribution of the historical eigenvectors to the final physical field, the historical main eigenvectors are screened and the corresponding historical main eigenvector coefficients are generated; The preprocessed first training sample data set is used as the first sub-model input, and the corresponding historical main eigenvector coefficient is used as the first sub-model output, and a scalar reduced-order sub-model is constructed to obtain a corresponding scalar reduced-order sub-model; Generate a historical computational physical field according to the historical main eigenvectors and the corresponding historical main eigenvector coefficients; Taking the historical main eigenvectors and the historical main eigenvector coefficients as the input of the second sub-model, taking the corresponding historical calculation physical field as the output of the second sub-model, constructing the output field synthesis calculation sub-model, and obtaining the corresponding output field synthesis calculation sub-model; The scalar reduction-order sub-model and the output field synthesis calculation sub-model are integrated to obtain the corresponding static field reduction-order model.

5. The digital twin visualization method based on the static field reduced-order model according to claim 4 is characterized in that: The formula for singular value decomposition of the historical training field matrix is: A=U×Σ×V T Where A is the historical training field matrix; U is the historical left eigenvector; Σ is the eigenvalue matrix; V T is the transpose of the historical right eigenvector.

6. The digital twin visualization method based on a static field reduced-order model according to claim 4 is characterized in that: According to the contribution of the historical eigenvectors to the final physical field, the historical main eigenvectors are screened and the corresponding historical main eigenvector coefficients are generated, including the following steps: According to the representation of the contribution of the eigenvalues ​​to the final physical field in the eigenvalue matrix, the error estimation of the historical main eigenvectors is defined using the error evaluation method; According to the error estimation, the error evaluation index of the historical main eigenvector is constructed; According to the error evaluation index, the historical main characteristic order of the historical main characteristic vector is obtained; According to the historical main feature order, the historical left feature vector of the historical feature vector is screened to obtain the corresponding historical main feature vector; According to the historical main eigenvectors, the eigenvalue matrix of the historical eigenvectors and the transpose of the historical right eigenvectors are processed to generate the corresponding historical main eigenvector coefficients.

7. The digital twin visualization method based on a static field reduced-order model according to claim 6, characterized in that: The formula for generating the historical calculation physics field is: F=a1U1+a2U2+...+a s U s +...+a r U r Where F is the historical calculation physical field data set; a s 、a r are the main eigenvector coefficients of the sth and rth histories respectively; U s , U r are the sth and rth historical main eigenvectors respectively; s is the main eigenvector indicator; r is the historical main characteristic order.

8. The digital twin visualization method based on a static field reduced-order model according to claim 4, characterized in that: According to several training sample data sets, a static field fault diagnosis model is constructed to obtain the corresponding static field fault diagnosis model, including the following steps: Extracting historical physical field data sets corresponding to a number of training sample data sets, and preprocessing and integrating all the historical physical field data sets to obtain a preprocessed second training sample data set containing fault diagnosis labels; The second training sample data set after preprocessing is divided into a model training sample set and a model testing sample set in a ratio of 7:3; Use some historical physical field point cloud data without fault diagnosis labels to pre-train the DBN network and build an initial static field fault diagnosis model; The initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm, and the IFWA optimization algorithm is used to optimize according to the optimization target to obtain the optimal initial network parameters of the DBN network; According to the optimal initial network parameters of the DBN network, the network structure of the initial static field fault diagnosis model is optimized, and the model training sample set is input to perform optimization training to obtain the optimized static field fault diagnosis model construction; Input the model test sample set, build the optimized static field fault diagnosis model, perform model testing, and obtain the second model accuracy; If the second model accuracy is greater than the second model accuracy threshold, the optimal static field fault diagnosis model construction is output, otherwise, the optimization training continues.

9. The digital twin visualization method based on a static field reduced-order model according to claim 8, characterized in that: The initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm, and the IFWA optimization algorithm is used to optimize according to the optimization target to obtain the optimal initial network parameters of the DBN network, including the following steps: The initial network parameters of the DBN network are used as the optimization target of the IFWA optimization algorithm; Set the IFWA population parameters, maximum number of iterations and fitness function of the IFWA optimization algorithm, and use the optimization target as the position of the IFWA individuals in the IFWA population; According to the IFWA population parameters, the Circle chaotic mapping sequence is used to initialize the IFWA population to obtain the initialized IFWA population; According to the fitness function, calculate the fitness value of the IFWA individuals in the initialized IFWA population; Get the explosion radius and number of sparks of each initial firework in the initial firework set; According to the explosion radius and the number of sparks of each initial IFWA individual in the initialized IFWA population, fireworks are exploded to obtain an updated IFWA population; Use the Gaussian mutation algorithm to perform Gaussian mutation on the initialized IFWA population to generate a Gaussian mutated IFWA population; Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the initialized IFWA population to generate a dynamic reverse IFWA population; Calculate the fitness values ​​of all IFWA individuals in the updated IFWA population, the Gaussian-mutated IFWA population, and the dynamically reversed IFWA population, and take the IFWA individual with the minimum fitness value as the optimal individual; If the number of iterations reaches the threshold or the fitness value of the optimal individual meets the requirements, the optimal solution corresponding to the current optimal individual is output to obtain the optimal initial network parameters of the DBN network.

Images