System for switching artificial intelligence model of physical model analysis simulation, and operating method of system

Abstract

A system for switching an artificial intelligence model of a physical model analysis simulation, and an operating method of the system are provided. The operating method of the system for switching an artificial intelligence model of a physical model analysis simulation comprises the steps of: generating first physical model input data on the basis of a design condition; obtaining first analysis data generated as a result of a physical model simulation performed using the first physical model input data; primarily training an artificial intelligence model, which receives the first physical model input data so as to output predictive analysis data; primarily evaluating the performance of the artificial intelligence model; setting an input data recommendation condition by using the evaluation result of the artificial intelligence model; and generating second physical model input data on the basis of the input data recommendation condition.

Description

Artificial intelligence model conversion system for physical model analysis simulation and its operation method

[0001] The present invention relates to an artificial intelligence model conversion system for physical model analysis simulation and an operating method thereof, and more particularly, to a system for performing efficient development by applying and optimizing recommended conditions in the generation of input data for training an artificial intelligence model in order to convert a physical model analysis simulation into an inference using an artificial intelligence model, and an operating method thereof.

[0002] In areas such as product design, verification, and testing, efforts are ongoing to apply artificial intelligence (AI) model inference to the analysis of computer-based physical models related to computer-aided engineering (CAE), physical equation models, and mathematical empirical formulas. As is widely known, AI model inference requires a trained AI model, and adequate input and output data are essential for high-quality training.

[0003] If this is applied to the analysis of computer-based physical models, design conditions, process conditions, etc. are required as input data, and measurement data and inspection data representing the material properties of the results obtained through the design conditions and process conditions are required as the target of inference of the artificial intelligence model. Such measurement data and inspection data are obtained through simulations using existing computer-based physical models instead of production using design conditions, and obtaining a large amount of training data requires simulation resources and time proportional to the amount of training data required.

[0004] The technical problem to be solved by the present invention relates to a system for optimizing the development process by generating input / output data to which recommended conditions for training an artificial intelligence model are applied by an artificial intelligence model conversion system of a physical model analysis simulation, and an operating method of the system.

[0005] The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0006] The method of operating an artificial intelligence model conversion system for physical model analysis simulation according to an embodiment of the present invention for solving the above-described technical problem includes: a step of generating first physical model input data based on design conditions; a step of acquiring first analysis data generated as a result of a physical model simulation performed using the first physical model input data; a step of first training an artificial intelligence model that receives the first physical model input data and outputs predicted analysis data; a step of first evaluating the performance of the artificial intelligence model; a step of setting input data recommendation conditions using the evaluation result of the artificial intelligence model; and a step of generating second physical model input data based on the input data recommendation conditions.

[0007] In some embodiments of the present invention, the method may further include: obtaining second analysis data generated as a result of a physical model simulation performed using the second physical model input data; performing a second training by adding the second physical model input data as input data to the artificial intelligence model; and performing a second evaluation of the performance of the artificial intelligence model generated as a result of the second training.

[0008] In some embodiments of the present invention, prior to the step of first evaluating the performance of the artificial intelligence model, a step of receiving a model performance target value from a user is further included, and if the result of the second evaluation does not satisfy the model performance target value, the step of generating second physical model input data using the input data recommendation condition and the step of second training of the artificial intelligence model may be repeated.

[0009] In some embodiments of the present invention, the step of first evaluating the performance of the artificial intelligence model includes a step of first evaluating based on a difference between the predicted interpretation data output from the artificial intelligence model generated by the first training and the first interpretation data, wherein the difference may be calculated as at least one of indicators for evaluating the performance of the artificial intelligence model, such as MSE (Mean Squared Error), RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), and DICE.

[0010] In some embodiments of the present invention, the input data recommendation conditions may include index conditions and filtering conditions.

[0011] In some embodiments of the present invention, the step of setting input data recommendation conditions using the evaluation results of the artificial intelligence model may include the steps of: setting arbitrary sampling conditions for the first physical model input data; obtaining evaluation results when data sampled according to the arbitrary sampling conditions is input to the first trained artificial intelligence model; and setting the index conditions and filtering conditions.

[0012] According to an embodiment of the present invention for solving the above-described technical problem, a system for converting an artificial intelligence model of a physical model analysis simulation comprises: a processor; and a memory connected to the processor, wherein the memory stores instructions that cause the processor, when executed, to perform operations including: a step of generating first physical model input data based on a design condition; a step of acquiring first analysis data generated by a physical model simulation performed using the first physical model input data; a step of first training an artificial intelligence model that receives the first physical model input data and outputs predicted analysis data; a step of first evaluating the performance of the artificial intelligence model; a step of setting an input data recommendation condition using an evaluation result of the artificial intelligence model; and a step of generating second physical model input data based on the input data recommendation condition.

[0013] In some embodiments of the present invention, the instructions may further cause the processor to perform operations including: obtaining second analysis data generated as a result of a physical model simulation performed using the second physical model input data; performing a second training by adding the second physical model input data as input data to the artificial intelligence model; and performing a second evaluation of the performance of the artificial intelligence model generated as a result of the second training.

[0014] In some embodiments of the present invention, the command further causes the processor to perform, prior to the first step of evaluating the performance of the artificial intelligence model, a step of receiving a model performance target value from a user, and, if the result of the second evaluation does not satisfy the model performance target value, the step of generating second physical model input data using the input data recommendation condition and the second training step of the artificial intelligence model can be repeated.

[0015] Specific details of other embodiments are included in the detailed description and drawings.

[0016] The artificial intelligence model conversion system and operating method for physical model analysis simulation according to an embodiment of the present invention repeatedly generates input data according to recommended conditions set based on the evaluation results of the artificial intelligence model, and thus a significant reduction in the total time required to generate the artificial intelligence model can be expected. Meanwhile, even when generating the same number of physical model input data as before, it is possible to achieve excellent prediction performance of the analysis data.

[0017] The effects of the invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

[0018] FIG. 1 is a diagram for explaining an artificial intelligence model conversion system for a physical model analysis simulation according to an embodiment of the present invention.

[0019] FIG. 2 is a drawing for explaining the configuration of an artificial intelligence model conversion system for physical model analysis simulation according to an embodiment of the present invention.

[0020] FIG. 3 is a flowchart for explaining the operation method of an artificial intelligence model conversion system for physical model analysis simulation according to an embodiment of the present invention.

[0021] FIG. 4 is a drawing for explaining analysis data obtained by an operation method of an artificial intelligence model conversion system for physical model analysis simulation according to an embodiment of the present invention.

[0022] FIG. 5 is a diagram for explaining an artificial intelligence model for inferring prediction interpretation data using physical model input data in a conversion system (100) according to an embodiment of the present invention.

[0023] FIG. 6 is a flowchart illustrating a method for setting input data recommendation conditions by an artificial intelligence model conversion system of a physical model analysis simulation according to an embodiment of the present invention.

[0024] FIGS. 7a and 7b are graphs and diagrams for explaining the effect of an artificial intelligence model conversion system for a physical model analysis simulation according to an embodiment of the present invention.

[0025] The advantages and features of the present invention, and the methods for achieving them, will become clear with reference to the embodiments described in detail below together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, and the present embodiments are provided only to make the disclosure of the present invention complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

[0026] When a component is referred to as being "connected to" or "coupled to" another component, it includes both cases where it is directly connected or coupled to the other component or cases where there are other components intervening. Conversely, when a component is referred to as being "directly connected to" or "directly coupled to" another component, it indicates that there are no other components intervening. "And / or" includes each and any combination of one or more of the mentioned items.

[0027] The terminology used herein is for the purpose of describing embodiments only and is not intended to limit the present invention. In this specification, the singular also includes the plural unless specifically stated otherwise. As used herein, the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components, steps, operations, and / or elements.

[0028] Although the terms first, second, etc. are used to describe various components, these components are not limited by these terms. These terms are merely used to distinguish one component from another. Accordingly, it goes without saying that the first component mentioned below may also be a second component within the technical spirit of the present invention.

[0029] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used with the meaning commonly understood by those of ordinary skill in the art to which the present invention pertains. In addition, terms defined in commonly used dictionaries shall not be interpreted ideally or excessively unless explicitly and specifically defined.

[0030] FIG. 1 is a diagram for explaining an artificial intelligence model conversion system for a physical model analysis simulation according to an embodiment of the present invention.

[0031] Referring to FIG. 1, an artificial intelligence model conversion system (100, hereinafter referred to as “conversion system”) for a herd model analysis simulation according to an embodiment of the present invention can be controlled, for example, through a client (10) connected via a network (500).

[0032] The conversion system (100) may include at least one computing device for generating an artificial intelligence model using physical model input data such as design conditions and process conditions, and analysis data generated as a result of physical model simulation.

[0033] Such a computing device may include, for example, a memory storing at least one instruction for performing an artificial intelligence model conversion method of a physical model interpretation simulation, one or more processors configured to execute the instruction stored in the memory, and a storage device storing execution results by the processor and an artificial intelligence model, etc.

[0034] The above computing device may include a server computer suitable for processing large amounts of data, but the present invention is not limited thereto, and the conversion system (100) may include a PC (Personal Computer) for creating a lightweight learning model through machine learning and executing inference using the same.

[0035] The conversion system (100) can perform operations such as generating physical model input data, performing physical model simulation, and training and evaluating an artificial intelligence model for conversion of a physical model analysis simulation to an artificial intelligence model, and a detailed description related thereto will be described in more detail using FIG. 3, etc.

[0036] In some other embodiments of the present invention, a physical model simulation using physical model input data may be performed through a simulation system separate from the conversion system (100). That is, a simulation system connected to the conversion system (100) and a client (10) via a network (500) may receive physical model input data generated by the conversion system (100), and the simulation system may generate analysis data generated by the physical model simulation. The conversion system (100) may generate an artificial intelligence model based on the analysis data generated by the simulation system.

[0037] In order to control the conversion system (100), a user can access the conversion system (100) using a client (10) and execute an artificial intelligence model conversion method of a physical model analysis simulation according to an embodiment of the present invention through the conversion system (100). The client (10) can include a computing device such as a PC, a notebook computer, a laptop computer, a smartphone, or a tablet PC, for example.

[0038] FIG. 2 is a diagram for explaining the configuration of an artificial intelligence model conversion system for physical model analysis simulation according to an embodiment of the present invention.

[0039] Referring to FIG. 2, the artificial intelligence model conversion system (100) of the physical model analysis simulation according to the embodiment of the present invention may include a processor (110), a memory (120), an interface (130), and a storage device (140).

[0040] The processor (110) can control the operation of the switching system (100) of the present invention and the components included in the switching system (100) to correspond to the input information. The processor (110) can collect, process, or classify information necessary for the operation of the switching system (100) by executing instructions stored in the memory (120).

[0041] Meanwhile, the processor (110) can perform calculations necessary for performing physical model simulations using physical model input data and training an artificial intelligence model to output predicted interpretation data through inference. Furthermore, calculations for performing processes such as evaluating the artificial intelligence model and generating physical model input data based on recommended conditions set based on the evaluation results can be performed by the processor (110).

[0042] In some embodiments of the present invention, the processor (110) may include one or more GPUs (Graphics Processing Units), TPUs (Tensor Processing Units), NPUs (Neural Processing Units), etc., to accelerate and perform matrix operations, vector operations, and floating-point operations required for training an artificial intelligence model and inference using the artificial intelligence model. Since these units are composed of a plurality of processing elements arranged in parallel, they can process operations required for machine learning faster than a central processing unit (CPU) of a general structure.

[0043] The memory (120) can temporarily or non-temporarily store data processed by the processor (110). Accordingly, the memory (120) can store data such as applications and programs executed in the switching system (100), while also storing an artificial intelligence model generated by the processor (110). Accordingly, the processor (110) can load the artificial intelligence model stored in the memory (120) and use it to infer predictive analysis data.

[0044] The interface (130) may include a communication interface for communicating with external devices and servers through various communication methods, and an input / output interface for interacting with a user. The processor (110) may receive design conditions and input data recommendation conditions for generating physical model input data through the interface (130), and in some embodiments, may also receive analysis data generated as a result of a physical model simulation from a connected simulation system through the interface (130).

[0045] The storage device (140) can store programs and data required for the operation of the switching system (100). For example, the storage device (140) can store program data related to the operating method of the switching system and provide the program to the memory (120) when the program is executed by the processor (110).

[0046] FIG. 3 is a flowchart for explaining the operation method of an artificial intelligence model conversion system for physical model analysis simulation according to an embodiment of the present invention.

[0047] Referring to FIG. 3, the operating method of the artificial intelligence model conversion system of the physical model analysis simulation according to the embodiment of the present invention may include repeating the steps of generating first physical model input data based on design conditions (S110), obtaining first analysis data generated as a result of the physical model simulation using the first physical model input data (S120), first training an artificial intelligence model that receives wp1 physical model input data and outputs predicted analysis data (S130), first evaluating the performance of the artificial intelligence model (S140), step of setting input data recommendation conditions using the evaluation results of the artificial intelligence model when the first evaluation results do not satisfy the model performance target value (S150), step of generating second physical model input data based on the input data recommendation conditions (S160), step of adding the second physical input data to the artificial intelligence model as input data and performing second training (S170), and step of second evaluating the performance of the artificial intelligence model (S180).

[0048] First, a step (S110) of generating first physical model input data based on design conditions may be performed. The first physical model input data may include, for example, design data or process data that may be set in relation to a manufacturing process. The design data may include, for example, setting values ​​for operating process equipment in a specific unit process to perform a process. The process data may include, for example, measurement data measured by at least one measuring instrument installed in a process while performing a specific process on an object, and data obtained from inspection equipment for measuring physical, chemical, electrical, or optical properties obtained on an object as a result of performing the process.

[0049] Next, a step (S120) of acquiring first analysis data generated as a result of a physical model simulation using first physical model input data may be performed. This will be described with reference to FIG. 4.

[0050] FIG. 4 is a drawing for explaining analysis data obtained by an operation method of an artificial intelligence model conversion system for a physical model analysis simulation according to an embodiment of the present invention.

[0051] In some embodiments of the present invention, the conversion system (100) may perform a physical model simulation using the first physical model input data. The physical model simulation may include a simulation performed using an approximate numerical analysis method based on mesh division, such as at least one of the Finite Difference Method, the Finite Element Method, or the Finite Volume Method, for an object manufactured using the first physical model input data as design or process conditions.

[0052] For example, the first physical model input data may include data generated in various combinations regarding design or process conditions such as input current density, plating solution flow, copper foil area, shielding film distance, etc. Meanwhile, the first analysis data may include numerical analysis results generated as a result of a physical model simulation using the first physical model input data, for example, data such as current density of a panel manufactured using the first physical model input data.

[0053] In some embodiments of the present invention, the conversion system (100) may receive first analysis data generated as a result of a physical model simulation performed by a separate simulation system connected via a network (500). To this end, the conversion system (100) may provide the first physical model input data to the simulation system as numerical conditions for the physical model simulation.

[0054] Next, a first step (S130) of training an artificial intelligence model that receives first physical model input data and outputs predicted interpretation data may be performed. This will be described with reference to FIG. 5.

[0055] FIG. 5 is a diagram for explaining an artificial intelligence model for inferring prediction interpretation data using physical model input data in a conversion system (100) according to an embodiment of the present invention.

[0056] Referring to FIG. 5, the conversion system (100) can train an artificial intelligence model to predict the results of a physical model simulation from provided physical model input data. In this case, the input data is first physical model input data, and the output data is predicted analysis data.

[0057] An artificial intelligence model for predicting the results of a physics model simulation can be generated, for example, using a deep learning technique including multiple hidden layers, but the present invention is not limited thereto. Furthermore, if the artificial intelligence model includes a deep learning model, it can adopt the structure of an artificial neural network such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a physics-informed neural network that utilizes physical information embedded in partial differential equations.

[0058] In some embodiments of the present invention, the conversion system (100) may perform data preprocessing tasks such as data cleaning, feature compression, feature encoding, and feature-specific normalization prior to training an artificial intelligence model using the first physical model input data.

[0059] Meanwhile, in some other embodiments, multiple artificial intelligence models can be generated through the first training process by dividing the provided first physical model input data into multiple data and performing training of the artificial intelligence models.

[0060] Next, a first step (S140) of evaluating the performance of the artificial intelligence model can be performed.

[0061] Evaluating the performance of an artificial intelligence model may include, for example, a first evaluation based on the difference between the predicted interpretation data output from the artificial intelligence model generated through the first training and the first interpretation data. At this time, the conversion system (100) may use at least one of the indices for evaluating the performance of an artificial intelligence model, such as the Mean Squared Error (MSE), the Root Mean Squared Error (RMSE), the Mean Absolute Error (MAE), the Mean Absolute Percentage Error (MAPE), and the DICE, to derive the difference. In some embodiments, the conversion system (100) may use two or more evaluation methods.

[0062] In some embodiments of the present invention, prior to the first evaluation step, the switching system (100) may further include a step of receiving a model performance target value of the artificial intelligence model from the user.

[0063] Next, a step (S150) of setting input data recommendation conditions may be performed using the evaluation results of the artificial intelligence model. The step of setting input data recommendation conditions is described using FIG. 6.

[0064] FIG. 6 is a flowchart illustrating a method for setting input data recommendation conditions by an artificial intelligence model conversion system for a physical model analysis simulation according to an embodiment of the present invention.

[0065] Referring to FIG. 6, the process of setting input data recommendation conditions by the conversion system (100) may include a step of generating an arbitrary sampling condition (S210), a step of evaluating the performance of an artificial intelligence model (S220), and a step of setting index and filtering conditions (S230).

[0066] First, in the step of generating arbitrary sampling conditions (S210), sampling conditions for the first physical model input data provided as input data for the artificial intelligence model can be set. The setting of the sampling conditions can be performed using an already known algorithm, and may include conditions such as random sampling such as simple random sampling or stratified random sampling, or undersampling or oversampling that reduces or increases the number of samples of a class, but the present invention is not limited thereto.

[0067] In the step (S220) of evaluating the performance of the subsequent artificial intelligence model, a process is performed to obtain an evaluation result when sampling conditions are applied to the artificial intelligence model trained so far, in this case, the artificial intelligence model for which the first training has been completed. That is, this is a step of obtaining an evaluation result based on the difference between the predicted interpretation data obtained as an output when the sampled physical model input data is provided as input to the artificial intelligence model, and the interpretation data corresponding to the sampled physical model input data.

[0068] Meanwhile, as explained above, in the case where multiple artificial intelligence models are created in the first training stage (S130) of the artificial intelligence model, evaluation results based on the difference in prediction interpretation data between the multiple artificial intelligence models can be obtained.

[0069] Finally, a step (S230) of setting index and filtering conditions may be performed. Here, the index may include evaluation conditions such as the distance from existing data, the statistical characteristics of the predicted interpretation data through the artificial intelligence model, and the nonlinearity of the predicted interpretation data. In other words, the index corresponds to conditions for evaluating the relationship between sampled data and existing data, i.e., the first physical model input data, based on different criteria.

[0070] The filtering conditions may include the number of input data recommendation conditions set using the set index, the number of second physical model input data to be generated using the recommendation conditions, and these values ​​may be input by the user.

[0071] Referring again to FIG. 3, a step (S160) of generating second physical model input data may be performed based on the set input data recommendation conditions. That is, the second physical model input data may be generated according to the previously set index and filtering conditions.

[0072] In some embodiments of the present invention, the number of second physical model input data generated based on input data recommendation conditions may be smaller than the number of first physical model input data initially generated. That is, by setting the number of second physical model input data repeatedly generated based on input data recommendation conditions to be smaller than the number of first physical model input data initially generated, the overall evaluation results of the artificial intelligence model can be gradually improved.

[0073] A step (S170) of obtaining second analysis data generated by a physical model simulation using second physical model input data may be performed. This can be performed in the same manner as the process of step S120 by obtaining characteristic values ​​of an object manufactured under design conditions and process conditions using an existing numerical analysis simulation as described above using FIG. 4.

[0074] Next, a second training step (S180) may be performed for the AI ​​model using the second physical model input data. For example, the second training step may include fine-tuning the parameters of the AI ​​model using the additionally provided second physical model input data while freezing the parameters, but the present invention is not limited thereto.

[0075] Finally, the second trained artificial intelligence model can be evaluated for performance (S140) and if the model performance target value is satisfied, the series of processes can be terminated. If the model performance target value is not satisfied, the process of setting input data recommendation conditions and generating second physical model input data can be performed again.

[0076] FIGS. 7a and 7b are graphs and diagrams for explaining the effect of an artificial intelligence model conversion system for a physical model analysis simulation according to an embodiment of the present invention.

[0077] FIG. 7a is a box plot showing the performance evaluation results (Data1) of an artificial intelligence model generated by setting input data recommendation conditions according to an embodiment of the present invention and the performance evaluation results (Data2) of an artificial intelligence model generated by a conventional method, and FIG. 7b is a chart showing the numerical values ​​of the graph of FIG. 7a.

[0078] First, in the case of the comparative example (Data2), a physical model simulation was performed using a fixed number of physical model input data, such as 500, and training of the artificial intelligence model was performed. A total of 500 physical model simulations were performed, and the evaluation result of the artificial intelligence model was obtained as 92.16±1.35.

[0079] Meanwhile, in the case of the result (Data1) of the embodiment of the present invention, the evaluation result when the first 100 first physical model input data were generated, a physical model simulation was performed, and an artificial intelligence model was trained is described in the item of Iter 0 (100) and has a result value of 80.86±9.17. This corresponds to the result when the setting of the input data recommendation conditions and the generation of the second physical model input data were repeated 0 times.

[0080] In this regard, the input data recommendation conditions are set using the evaluation results, and the results of the generation of the second physical model input data, the second training of the artificial intelligence model, and the second evaluation are shown in Iter 1, 2, 쪋, 10, and Iter n represents the number of repetitions of the generation of the second physical model input data and the second training using the same. Meanwhile, in the examples of FIGS. 7a and 7b, it is explained that a fixed number (40) of second physical model input data is generated for each repetition, but as explained above, the number of second physical model input data can be set by the user and thus can vary as much as desired.

[0081] According to an embodiment of the present invention, when the generation of the second physical model input data and the second training using the same were repeated 6 times, the evaluation result was 93.01±3.19, which exceeds the evaluation result of the comparative example. A physical model simulation was performed using a total of 340 physical model input data, and the performance of the trained artificial intelligence model exceeds the performance of the existing artificial intelligence model. In addition, the number of physical model simulations, which accounts for a significant portion of the time required for the process, is performed only 340 times, which is less than 70% of the 500 times of the comparative example, so a significant reduction in the total time required can be expected.

[0082] In addition, when the number of repetitions is increased to generate the same number of physical model input data as 500 in the comparative example, the evaluation result is 94.34±2.84, which means that the prediction performance of the interpretation data is superior to that of the artificial intelligence model generated by the comparative example.

[0083] Meanwhile, assuming that model performance targets are input from users as described above, if the performance evaluation results of the generated AI model fall below the target model performance when using a fixed number of physical model input data, as in the comparative example, additional physical model input data will be required. This is simply input model data with arbitrary values, regardless of the recommended input data conditions, so it may be difficult to expect only an increase in AI model performance due to additional data.

[0084] On the other hand, in the case of the conversion system (100) according to the embodiment of the present invention, after receiving a model performance target value from a user, a second training is performed using the second physical model input data, and if the evaluation result of the artificial intelligence model trained through the second training does not satisfy the model performance target value, the generation of additional second physical model input data can be repeated with the input data recommendation condition set using the evaluation result. The generation of physical model input data and the second training process can be repeated until the model performance target value is satisfied.

[0085] The present invention can also be implemented as computer-readable code on a computer-readable recording medium. Computer-readable recording media include all types of recording devices that store data that can be read by a computer device. Examples of computer-readable recording media include hard disks, ROMs, RAMs, CD-ROMs, solid-state disks, magnetic tapes, floppy disks, and optical data storage devices.

[0086] Although the embodiments of the present invention have been described with reference to the attached drawings, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without altering the technical concept or essential features thereof. Therefore, the embodiments described above should be understood to be illustrative in all respects and not restrictive.

Claims

1. A step of generating first physical model input data based on design conditions; A step of obtaining first analysis data generated as a result of a physical model simulation performed using the first physical model input data; A first step of training an artificial intelligence model that receives the first physical model input data and outputs predicted interpretation data; A first step of evaluating the performance of the above artificial intelligence model; A step of setting input data recommendation conditions using the evaluation results of the above artificial intelligence model; and A step of generating second physical model input data based on the above input data recommendation conditions, How the artificial intelligence model conversion system for physical model interpretation simulation works.

2. In paragraph 1, A step of obtaining second analysis data generated as a result of a physical model simulation performed using the second physical model input data; A second training step by adding the second physical model input data as input data to the artificial intelligence model; and Further comprising a second step of evaluating the performance of the artificial intelligence model generated as a result of the second training. How the artificial intelligence model conversion system for physical model interpretation simulation works.

3. In paragraph 2, Before the first step of evaluating the performance of the artificial intelligence model, a step of receiving a model performance target value from a user is further included. If the result of the second evaluation does not satisfy the model performance target value, the second physical model input data generation using the input data recommendation condition and the second training step of the artificial intelligence model are repeated. How the artificial intelligence model conversion system for physical model interpretation simulation works.

4. In paragraph 1, The first step in evaluating the performance of the above artificial intelligence model is: A first evaluation step is included based on the difference between the predicted interpretation data output from the artificial intelligence model generated by the first training and the first interpretation data, wherein the difference is calculated as at least one of the indices for evaluating the performance of the artificial intelligence model, such as MSE (Mean Squared Error), RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), and DICE. How the artificial intelligence model conversion system for physical model interpretation simulation works.

5. In paragraph 1, The above input data recommendation conditions are: Including index conditions and filtering conditions, How the artificial intelligence model conversion system for physical model interpretation simulation works.

6. In paragraph 5, The step of setting input data recommendation conditions using the evaluation results of the above artificial intelligence model is as follows: A step of setting arbitrary sampling conditions for the first physical model input data; A step of obtaining an evaluation result when inputting data sampled according to the above arbitrary sampling conditions into the first trained artificial intelligence model; and Comprising a step of setting the above index conditions and filtering conditions, How the artificial intelligence model conversion system for physical model interpretation simulation works.

7. Processor; and Memory connected to the processor; Including, The above memory, when executed, the processor, A step of generating first physical model input data based on design conditions; A step of obtaining first analysis data generated by a physical model simulation performed using the first physical model input data; A first step of training an artificial intelligence model that receives the first physical model input data and outputs predicted interpretation data; A first step of evaluating the performance of the above artificial intelligence model; A step of setting input data recommendation conditions using the evaluation results of the above artificial intelligence model; and storing a command that causes an operation to be performed, including a step of generating second physical model input data based on the above input data recommendation conditions; Artificial intelligence model conversion system for physical model interpretation simulation.

8. In paragraph 7, The above instruction causes the processor to, A step of obtaining second analysis data generated as a result of a physical model simulation performed using the second physical model input data; A second training step by adding the second physical model input data as input data to the artificial intelligence model; and To further perform an operation including a second evaluation step of the performance of the artificial intelligence model generated as a result of the second training, Artificial intelligence model conversion system for physical model interpretation simulation.

9. In paragraph 8, The above instruction causes the processor to, Before the first step of evaluating the performance of the artificial intelligence model, a step of receiving a model performance target value from the user is further performed. If the result of the second evaluation does not satisfy the model performance target value, the second physical model input data generation using the input data recommendation condition and the second training step of the artificial intelligence model are repeated. Artificial intelligence model conversion system for physical model interpretation simulation.

10. In paragraph 7, The first step in evaluating the performance of the above artificial intelligence model is: An artificial intelligence model conversion system for a physical model analysis simulation, comprising a first evaluation step based on a difference between predicted interpretation data output from an artificial intelligence model generated through the first training and the first interpretation data, wherein the difference is calculated as at least one of indicators for evaluating the performance of an artificial intelligence model, such as MSE (Mean Squared Error), RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), and DICE.

11. In paragraph 7, The above input data recommendation conditions are: Including index conditions and filtering conditions, Artificial intelligence model conversion system for physical model interpretation simulation.

12. In paragraph 11, The step of setting input data recommendation conditions using the evaluation results of the above artificial intelligence model is as follows: A step of setting arbitrary sampling conditions for the first physical model input data; A step of obtaining an evaluation result when inputting data sampled according to the above arbitrary sampling conditions into the first trained artificial intelligence model; and Comprising a step of setting the above index conditions and filtering conditions, Artificial intelligence model conversion system for physical model interpretation simulation.

Images