Multi-physical domain agent model modeling and solving method, system and medium

Abstract

The invention belongs to the technical field of airborne electromechanical system digital twin modeling and simulation, and particularly relates to a multi-physical-domain agent model modeling and solving method and system and a medium, and the method comprises the steps: 1) carrying out the modeling of a general agent model based on a structured number table; firstly, a system mechanism model is used as a unique data source to solve and generate a model solution space database; secondly, the range of input parameters is determined, the influence of the input parameters on output parameter changes is analyzed, and all possible input combinations are formed; thirdly, all possible input combinations are submitted to the system mechanism model for calculation, response results are output, and input and output relations are clustered and sorted into a group of structured number tables; fourthly, converting the structured data table into an array in the code, and finally converting the algebraically expressed system mechanism model into a code array; 2) solving the general agent model established in the step 1); the method comprises the following steps: firstly, reading externally input to-be-solved input parameter values; secondly, determining a parameter value closest to a to-be-solved input parameter in a representation solution space array in a general agent model code; and thirdly, querying an output value of the closest parameter value, and outputting the output value as a solving result.

Description

Technical Field

[0001] This invention belongs to the field of digital twin modeling and simulation technology of airborne electromechanical systems, specifically involving a multi-physics domain proxy modeling and solving method, system and medium. Background Technology

[0002] A digital model of an electromechanical system is a computer-recognizable, processable, analytical, and applicable digital representation of the multi-physical domain mechanisms, phenomena, and processes of a system. The functions of an aircraft's electromechanical system are mainly divided into energy support (electricity, hydraulics, and pneumatics), environmental support (equipment, personnel environment), and other functions (cargo airdrop, braking, etc.), encompassing subsystems such as auxiliary power, environmental control, fuel, power supply, hydraulics, and airdrop systems, involving thousands of devices across multiple physical domains including mechanical, electrical, hydraulic, and pneumatic systems. With the development of aircraft digital design, verification, and operation and maintenance technologies, the digital model of the electromechanical system, as a crucial functional system component of an aircraft, needs to characterize its operational features and processes in virtual integration verification, semi-physical verification, flight simulators, and digital aircraft support systems. It has numerous application scenarios and requirements throughout the entire aircraft lifecycle digital engineering process.

[0003] To improve the power-to-weight ratio and reliability of the system, existing advanced electromechanical systems feature a dual integration of control and energy architectures. The integrated control architecture combines the previously separate data acquisition and control computers of multiple subsystems into a single computer, managing and controlling each subsystem of the electromechanical system via a high-speed bus. The system composition includes mechanical and electrical equipment such as motors, pumps, and valves, control equipment such as brake controllers and motor controllers, bus devices such as interface units and processing computers, etc., making the control and mechanical and electrical interface relationships between the subsystems more complex. The integrated energy architecture connects the previously separate channels for generating, configuring, and consuming electrical, hydraulic, and pneumatic energy into a comprehensive energy network. The system composition includes energy conversion components such as high-voltage energy storage batteries, electro-hydraulic conversion devices, and turbine generators.

[0004] The coupling characteristics of integrated electromechanical systems in the control and energy domains necessitate the construction of simulation models containing multiple subsystems to fully express the control, fluid, mechanical, and electrical characteristics. However, current system simulations such as AMEsim, SABER, SimLink, and ANSYS only have the ability to simulate one or two physical domains. To achieve multi-physical domain simulation of large-scale and complex integrated electromechanical systems, it is necessary to adopt the FMU co-simulation approach.

[0005] There are three problems with system integration simulation based on FMU co-simulation: First, the principle of FMU is to achieve data exchange between different models by standardizing and encapsulating external interfaces, but the internal complexity of the model remains unchanged, resulting in high integration difficulty and a large amount of debugging work for FMU-based co-simulation. Second, the solution of FMU depends on the internally integrated general solver, which has low efficiency for solving complex models. At present, some simulation software such as AMEsim also provides neural network-based surrogate model reduction tools to improve the computational efficiency of FMU modules, but due to the lack of a unified reduction standard among tools, the FMU models after reduction by different tools still have the problem of high debugging difficulty. Summary of the Invention

[0006] The purpose of this invention is to address the problems of digital prototypes for electromechanical systems, which express the multi-physical domain characteristics of electromechanical systems through mathematical functions. These problems stem from limitations imposed by different simulation software modeling processes, solver differences, and proxy model generation mechanisms. Furthermore, the invention addresses the varying requirements of different application scenarios for electromechanical system models, such as design evaluation, testing, and digital operation and maintenance, regarding the hardware and software operating environment, operating speed, accuracy, and granularity of output parameters. These limitations lead to the need to develop multiple models for electromechanical systems, resulting in a large modeling workload, difficulties in managing the technical status of models, and development cycles that cannot meet equipment requirements.

[0007] The purpose of this invention is to propose a multi-physics domain proxy modeling and solving method, system, and medium. It generates a standard proxy model that can run and be quickly integrated in different software and hardware environments by using a high-precision system mechanism model as the sole data source. This enables multi-scenario, multi-granularity proxy modeling and agile delivery and integration based on a high-precision mechanism model as the authoritative data source, reduces the workload of proxy modeling and integration for electromechanical systems in specific scenarios, and expands the application scenarios of electromechanical system models.

[0008] The technical solution of the present invention:

[0009] A method for modeling and solving multi-physics domain proxy models, comprising:

[0010] 1) Modeling a general surrogate model based on structured tables;

[0011] First, a solution space database is generated by solving the system mechanism model as the sole data source. Second, the range of input parameters is determined, and the influence of input parameters on changes in output parameters is analyzed to form all possible input combinations. Third, all possible input combinations are submitted to the system mechanism model for calculation, the response results are output, and the relationship between input and output is clustered and organized into a set of structured tables. Fourth, the structured tables are converted into arrays in the code, and finally the algebraically expressed system mechanism model is transformed into a code array.

[0012] 2) Solve the general proxy model established in 1);

[0013] First, read the input parameter values ​​to be solved from the external input; second, determine the parameter value in the solution space array of the general surrogate model code that is closest to the input parameter to be solved; third, query the output value of the closest parameter value and output it as the solution result.

[0014] Furthermore, the system mechanism model is based on the system's physical mechanism and geometric dimensions, and expresses the system's physical characteristics through physical formulas. The parameters of the physical formulas are adjusted using experimental data to ensure that the difference between the input-output relationship and the actual physical product test data is no greater than a limit value, thus ensuring that the output value of the system mechanism model can remain consistent with the physical product response output under high confidence.

[0015] Furthermore, the physical formulas include Maxwell's equations and Navier-Stokes equations.

[0016] Furthermore, the range of input parameters is determined based on the system engineering scenario.

[0017] Furthermore, the influence of input parameters on changes in output parameters is analyzed. Parameters with significant influence are discretized at a fine granular level, while parameters with minor influence are discretized at a coarse granular level. The discretization results are then combined to form all possible input combinations.

[0018] Furthermore, neural networks are used to cluster the input and output relationships, organizing them into a set of structured tables.

[0019] Furthermore, standard data storage protocols are used to convert the data table into an array in the code.

[0020] Furthermore, the sorting program is run to determine the parameter value in the solution space array of the surrogate model code that is closest to the input parameter to be solved; the output value of the closest parameter value is queried according to the predefined data protocol and output as the solution result.

[0021] A multi-physics domain proxy modeling and solving system includes:

[0022] General Proxy Modeling Module:

[0023] This module uses the system mechanism model as the sole data source to generate a model solution space database; it determines the range of input parameters, analyzes the impact of input parameters on changes in output parameters, and forms all possible input combinations; it submits all possible input combinations to the system mechanism model for calculation, outputs response results, clusters the input and output relationships, and organizes them into a set of structured tables; it converts the structured tables into arrays in the code, and finally transforms the algebraically expressed system mechanism model into a code array;

[0024] General surrogate model solving module:

[0025] This module is used to read the input parameter values ​​to be solved from external input; determine the parameter value in the solution space array of the general surrogate model code that is closest to the input parameter to be solved; and query the output value of the closest parameter value as the solution result.

[0026] A computer storage medium storing a computer program for executing the multi-physics proxy model modeling and solving method.

[0027] The beneficial effects of this invention are:

[0028] To address the multi-scenario, multi-granularity integration and interaction requirements of complex multi-physics domain models in integrated electromechanical systems, this invention proposes a multi-physics domain proxy modeling and solving method, system, and medium. This enables cross-platform integration of current multi-physics domain characteristic models of electromechanical systems across multiple scenarios, improves the reusability efficiency of electromechanical system models, and supports the application of electromechanical digital prototype products. Specifically, it offers the following advantages:

[0029] 1) Integration of surrogate model generation and solution: It eliminates the need for the dynamic link libraries required by current FMU model integration methods. Only a single piece of code without any function calls and a set of floating-point numbers are needed to express and solve for fluid, mechanical, and electrical properties, significantly improving the model's adaptability to computations across different operating systems.

[0030] 2) Sensitivity analysis and neural network algorithms are used to perform two dimensionality reductions on the solution space of the mechanism model. While ensuring the accuracy of the model, the complex proxy model of the multi-physics system can be simplified to be directly expressed by arrays.

[0031] 3) For the generated array representation proxy model, code is written according to a certain data protocol, and the solution process can be independent of database dependency.

[0032] 4) For the surrogate model solver, the built-in model data protocol allows direct reading of input variables and output values ​​from the code array without database operations.

[0033] 5) For the surrogate model solver, a built-in sorting algorithm is used to find the input value of the model input variable that is closest to the variable to be solved by comparing the sorting of the variable to be solved and the input variable of the surrogate model. The solution is completed without the use of any other function library, and the implementation is adapted to different operating system software environments to the greatest extent. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the modeling and solution process of the environmental control evaporator proxy model.

[0035] Figure 2 This is a schematic diagram of the power loss storage sequence of the evaporator pack. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0037] Example 1: Differences in function libraries and data management strategies across different operating systems and software are the main reasons affecting the operation of proxy models in different environments. This invention addresses the challenge of model reuse and efficient integration of multi-physics domain models for integrated electromechanical systems under different simulation hardware and software environments, proposing a method for modeling and solving multi-physics domain proxy models. The specific solution is as follows:

[0038] 1) Modeling a general surrogate model based on structured tables

[0039] The fundamental principle of surrogate modeling is to transform mathematical expressions containing complex algebraic functions into arrays where input and output relationships directly correspond. This changes the model computation process from an algebraic solution process to a data query process, reducing the model's reliance on dedicated solvers. The specific process is as follows:

[0040] First, a model solution space database is generated by solving a high-precision system mechanism model as the sole data source. The high-precision mechanism model is based on the system's physical mechanism and geometric dimensions, and expresses the system's physical characteristics using physical formulas such as Maxwell's equations and Navier-Stokes equations. The parameters of the physical formulas are adjusted using experimental data to ensure that the difference between the input-output relationship and the actual physical product test data is no greater than a limit value, thus ensuring that the model's output value remains consistent with the physical product's response output at a high confidence level.

[0041] Taking turbine efficiency calculation as an example, the isentropic efficiency is commonly used for turbine efficiency:

[0042]

[0043] in:

[0044] h 01 ,h 02 Total enthalpy of the actual process;

[0045] h 02s Total enthalpy at the exit of an isentropic process;

[0046] T 01 ,T 02 ,T 02s : The corresponding total temperatures;

[0047] Under the ideal gas assumption, h = c p T.

[0048] Substituting the enthalpy expression, we get:

[0049]

[0050] Calibration parameters:

[0051] parameter illustrate Typical value γ The specific heat ratio of the gas is approximately 1.33-1.4. 1.36 <![CDATA[c p ]]> Specific heat capacity (at normal pressure) 1150 J / (kg·K) <![CDATA[T 01 ,T 02 ]]> Total temperature at inlet / outlet 1400K / 1000K <![CDATA[P 01 ,P 02 ]]> Total pressure at inlet / outlet 1.6MPa / 0.4MPa

[0052] Secondly, a commonly used surrogate model generation method is adopted. Based on the system engineering scenario, the range of input parameters is determined, and the influence of input parameters on the changes in output parameters is analyzed. For parameters with significant influence, finer-grained discretization is performed, while for those with less influence, coarser-grained discretization is performed. The discretization results are then combined to form all possible input combinations.

[0053] Taking turbine efficiency calculation as an example:

[0054] Input variables symbol scope illustrate Total inlet temperature <![CDATA[T 01 ]]> 1300–1600K Hot end design Total import pressure <![CDATA[P 01 ]]> 1.0–1.8 MPa Provided by the compressor Export pressure <![CDATA[P 02 ]]> 0.2–0.8 MPa According to nozzle back pressure rotational speed N 8000–18000rpm Turbine speed Cooling airflow ratio β 0–0.1 Cooling air mixture ratio

[0055] Local sensitivity analysis:

[0056] The most common method is to calculate the partial derivatives of the model's input variables:

[0057]

[0058] in:

[0059] x i Input variables (e.g., pressure ratio, rotational speed);

[0060] y: Output variable (turbo efficiency);

[0061] Numerically, partial derivatives are usually approximated using the finite difference method:

[0062]

[0063] Global sensitivity analysis (Sobol method):

[0064] When using the global method Sobol sensitivity analysis, the Sobol index that will be used is:

[0065] First-order Sobol index:

[0066]

[0067] Total effect Sobol index:

[0068]

[0069] in:

[0070] S i Input variable x i Individual impact on output;

[0071] S Ti Input variable x i Together with the total effect of interactions with other variables;

[0072] Var(Y): The total variance of the output; typically obtained through Monte Carlo sampling and fitting a surrogate model. Example of sensitivity analysis (Sobol method):

[0073] parameter Sensitivity Index <![CDATA[T 01 ]]> 0.35 <![CDATA[P 01 ]]> 0.25 <![CDATA[p 02 ]]> 0.2 N 0.15 β 0.05

[0074] Discretization example:

[0075] T 01 Fine discretization → [1300,1350,1400,1450,1500,1550,1600]

[0076] P 01 Fine discretization → [1.0, 1.2, 1.4, 1.6, 1.8]

[0077] P 02 Fine discretization → [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]

[0078] N: Coarse discretization → [10000, 14000, 18000]

[0079] β: Coarse discretization → [0.0, 0.05, 0.1]

[0080] Third, all possible input combinations are submitted to a high-precision mechanism model for calculation, and the output response results are used to cluster the input and output relationships using a neural network to further reduce and organize them into a set of structured tables.

[0081] Taking turbine efficiency calculation as an example:

[0082] Data structuring (sample format):

[0083] <![CDATA[T 01 (K)]]> <![CDATA[P 01 (MPa)]]> <![CDATA[P 02 (MPa)]]> N (rpm) β <![CDATA[η t ]]> 1400 1.4 0.4 14000 0.05 0.89 … … … … … …

[0084] K-Means clustering algorithm:

[0085] Input variable: x = [T 01 ,P 01 ,P 02 [,N,β]:

[0086] Input condition η: Turbine efficiency

[0087] Data preprocessing: Standardize input features to ensure all input components are on the same scale, avoiding orders-of-magnitude differences such as pressure ratio and temperature.

[0088]

[0089] Cluster center initialization:

[0090] Choose K initial cluster centers μ1, μ2, ..., μ K Used to measure the distance from a data point to a cluster.

[0091] Iterative updates:

[0092] Step 1. Assign data points to the nearest cluster centers:

[0093]

[0094] Step 2. Update the centroid of each cluster:

[0095]

[0096] Repeat steps 1 and 2 until the center no longer changes or the maximum number of iterations is reached.

[0097] After clustering, the average efficiency is analyzed, and the mean turbine efficiency in each cluster is calculated:

[0098]

[0099] Example output:

[0100] Cluster <![CDATA[Central T 01 > <![CDATA[Central P 01 > <![CDATA[Central P 02 > Center N central β <![CDATA[Central efficiency η t > C1 1350 1.2 0.3 14000 0.0 0.91 C2 1600 1.6 0.6 10000 0.05 0.85 … … … … … … …

[0101] Fourth, standard data storage protocols are used to convert the number table into an array in the code, ultimately transforming the complex mechanism model expressed in algebra into an array of code.

[0102] 2) General surrogate model solver based on general retrieval algorithm

[0103] The principle of the general surrogate model solver is a set of program code that does not rely on function library calls. Since the solution space of the surrogate model has been transformed into an array of code, the solution process can essentially be transformed into a data sorting and querying problem. Therefore, simple querying and sorting code can directly complete the querying and output of model results, solving the problem of solving difficulties caused by differences in function libraries of different operating systems such as Windows, Linux, and Kylin, and simulation software. The specific solution is as follows:

[0104] First, read the values ​​of the input parameters to be solved from the external input.

[0105] Secondly, the sorting program is run to determine the parameter values ​​in the solution space array of the surrogate model code that are closest to the input parameters to be solved.

[0106] Third, according to a predefined data protocol, query the output value of the closest parameter value and output it as the solution result.

[0107] Example 2:

[0108] The evaporator is a key thermal management component and electrical load in the electromechanical system. Its principle involves an electric motor driving an evaporator pump to compress a high-temperature gas-liquid mixture. The compressed liquid is then cooled by a heat exchanger, resulting in a cooler liquid heat sink, thus transferring heat from a low temperature to a high temperature. The mechanism model uses CFD software to calculate the fluid component performance, AMEsim software to calculate the heat transfer performance, and SimLink software to calculate the controller logic, ultimately obtaining the evaporator's power consumption under different operating conditions. This is a typical multiphysics domain model. However, the CFD, AMEsim, and SimLink software required for this model's computation all run on Windows operating systems. In contrast, users of the actual model need to utilize the domestically developed Kylin operating system, rendering the existing model unsuitable for their needs.

[0109] To solve this model under the Kylin operating system, a proxy model modeling and solution process for the evaporator component model of the environmental control system is proposed. Figure 1 Therefore, we first used CFD software, AMEsim software, and SimLink software to perform joint modeling and simulation under the Windows operating system, and then used experimental data to verify and adjust the model parameters until the error between the model calculation results and the experimental data was reduced to within 5% of the allowable range.

[0110] Based on the calibrated model, the input parameters of rotation speed 0-500 rpm, pressure 0-300 kPa, inlet temperature 0-10 degrees Celsius, and outlet temperature 30 to -30 degrees Celsius were solved, obtaining 5000 sets of calculation results. Then, a neural network was used to cluster these 5000 sets of results, with a deviation limit of 1%, resulting in 1100 clusters. Table 1 shows the discrete clustering of the evaporator power loss. The results were then arranged according to... Figure 2 The data format shown is stored in the proxy model code.

[0111] Table 1 Power loss of evaporator after discrete clustering

[0112]

[0113] When using this loss table for model solving calculations, it is necessary to quickly look up the data based on the input parameters. Figure 2The model whose data format is closest to the input parameters is output to [the specified data type]. Write the quicksort code shown below to [the specified data type]. Figure 2 Sort the data in the data:

[0114]

[0115]

[0116] Based on the sorting results, the model calculation values ​​corresponding to the set of model input parameters that are closest to the parameters to be solved can be found, thus completing the final model solution.

[0117] In this embodiment, only a piece of code for sorting the sequence is executed in the Kylin operating system. This code does not call any other external software or functions and can run in both Windows and Kylin operating systems without modification, ensuring the consistency of the data source and the rapid migration and interaction of the model.

[0118] Example 3: A multi-physics domain proxy modeling and solving system is provided, including:

[0119] General Proxy Modeling Module:

[0120] This module uses the system mechanism model as the sole data source to generate a model solution space database; it determines the range of input parameters, analyzes the impact of input parameters on changes in output parameters, and forms all possible input combinations; it submits all possible input combinations to the system mechanism model for calculation, outputs response results, clusters the input and output relationships, and organizes them into a set of structured tables; it converts the structured tables into arrays in the code, and finally transforms the algebraically expressed system mechanism model into a code array;

[0121] General surrogate model solving module:

[0122] This module is used to read the input parameter values ​​to be solved from external input; determine the parameter value in the solution space array of the general surrogate model code that is closest to the input parameter to be solved; and query the output value of the closest parameter value as the solution result.

[0123] The final embodiment of the present invention provides a computer storage medium storing a computer program for executing the multi-physics proxy model modeling and solving method.

[0124] The above description is merely a specific embodiment of the present invention, providing a detailed description of the invention. Parts not covered herein are conventional techniques. However, the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for modeling and solving a multi-physics domain proxy model, characterized in that, include: 1) Modeling a general surrogate model based on structured tables; First, a model solution space database is generated by solving the system mechanism model as the sole data source. Secondly, determine the range of input parameters, analyze the impact of input parameters on changes in output parameters, and form all possible input combinations; Third, submit all possible input combinations to the system mechanism model for calculation, output the response results, cluster the input and output relationships, and organize them into a set of structured tables; Fourth, the structured number table is transformed into an array in the code, and finally the algebraic system mechanism model is transformed into a code array; 2) Solve the general proxy model established in 1); First, read the input parameter values ​​to be solved from the external input; second, determine the parameter value in the solution space array of the general surrogate model code that is closest to the input parameter to be solved; third, query the output value of the closest parameter value and output it as the solution result.

2. The multi-physics domain proxy modeling and solving method according to claim 1, characterized in that, The system mechanism model is based on the system's physical mechanism and geometric dimensions, and expresses the system's physical characteristics through physical formulas. The parameters of the physical formulas are adjusted using experimental data to ensure that the difference between the input-output relationship and the actual physical product test data is no greater than a limit value, thus ensuring that the output value of the system mechanism model remains consistent with the physical product response output under high confidence.

3. The multi-physics domain proxy modeling and solving method according to claim 2, characterized in that, The physical formulas include Maxwell's equations and Navier-Stokes equations.

4. The multi-physics domain proxy modeling and solving method according to claim 1, characterized in that, Determine the range of input parameters based on the system engineering scenario.

5. The multi-physics domain proxy modeling and solving method according to claim 4, characterized in that, The impact of input parameters on output parameter changes is analyzed. Parameters with significant impact are discretized at a fine granular level, while parameters with minor impact are discretized at a coarse granular level. The discretization results are then combined to form all possible input combinations.

6. The method for modeling and solving multi-physics domain proxy models according to claim 5, characterized in that, The relationship between input and output is clustered using neural networks and organized into a set of structured tables.

7. The multi-physics domain proxy modeling and solving method according to claim 4, characterized in that, The table is converted into an array in the code using a standard data storage protocol.

8. The method for modeling and solving multi-physics domain proxy models according to claim 1, characterized in that, Run the sorting program to determine the parameter value in the solution space array of the surrogate model code that is closest to the input parameter to be solved; query the output value of the closest parameter value according to the predefined data protocol, and output it as the solution result.

9. A multi-physics domain proxy modeling and solving system, applied to the method described in any one of claims 1-8, characterized in that, include: General Proxy Modeling Module: This module uses the system mechanism model as the sole data source to generate a model solution space database; it determines the range of input parameters, analyzes the impact of input parameters on changes in output parameters, and forms all possible input combinations; it submits all possible input combinations to the system mechanism model for calculation, outputs response results, clusters the input and output relationships, and organizes them into a set of structured tables; it converts the structured tables into arrays in the code, and finally transforms the algebraically expressed system mechanism model into a code array; General surrogate model solving module: This module is used to read the values ​​of the input parameters to be solved from external input; and to determine the parameter values ​​in the solution space array of the general surrogate model code that are closest to the input parameters to be solved. It then queries the output value of the closest parameter value and outputs it as the solution result.

10. A computer storage medium, characterized in that, The storage medium stores a computer program, which is used to execute the multi-physics domain proxy model modeling and solving method according to any one of claims 1-8.

Images