Vector field analysis system and analysis method

Abstract

A system for analyzing a vector field according to an embodiment disclosed herein comprises: a data reception unit that receives a plurality of pieces of vector field data input by a user; a preprocessing unit that calculates pieces of gradient data of the plurality of pieces of vector field data and generates vector field similarity data on the basis of the similarity of the calculated pieces of gradient data; a result analysis unit that includes a pre-trained artificial intelligence model, wherein the vector field similarity data is input to the artificial intelligence model, and the artificial intelligence model infers an analysis result dependent on the change factors; and an output unit that displays the change factors inferred by the result analysis unit and the analysis result that is dependent on the change factors.

Description

Vector field analysis system and analysis method

[0001] The disclosed embodiment relates to a system and analysis method for analyzing a change factor of a vector field, and more specifically, to a vector field analysis system and analysis method for predicting and analyzing a change in a vector field according to each change factor through artificial intelligence.

[0002] A vector field represents a function or mapping that assigns a vector to each point in space, representing a collection of vectors with a direction and magnitude at each point. These vector fields are applied to a wide range of fields, including electromagnetic fields, gravitational fields, and fluid dynamics. They can be particularly useful in designing ships, aircraft, and vehicles based on flow velocity or wind volume, as well as in the hardware design of wind power plants.

[0003] The analysis study on the change factors that change the conventional vector field was conducted on a trial and error basis, as disclosed below.

[0004] [Previous technical literature]

[0005] A Multidisciplinary Study on the Influence of Various Elements Related to Automotive Rear Spoilers on Automotive Aerodynamic Performance, Kim Jeong-min, Korea Exhibition Industry Convergence Research Institute, 2020

[0006] Previous studies analyzing vector field variation factors based on trial and error (T&E) have required significant costs for evaluator training and experimental environment construction. Furthermore, because the analysis of vector field variation factors is conducted manually, the results can vary across evaluators, making it difficult to ensure objectivity. Furthermore, these techniques have shown problems with low accuracy or predictive power when the number of vector field variation factors increases or the influence of interactions increases.

[0007] Meanwhile, Physics-Informed Neural Networks (PINNs) are a conventional technology that trains AI models using partial differential equations representing vector fields as loss functions. PINNs focus on predicting physical information that users may not have access to, and have been limited to techniques for tracking and analyzing changes in vector fields due to changing factors.

[0008] The disclosed embodiment relates to a vector field analysis system and analysis method that performs preprocessing capable of representing relationships between vector fields and then inputs preprocessed vector field data into artificial intelligence that classifies image data.

[0009] A vector field analysis system according to one disclosed embodiment includes: a data receiving unit that receives a plurality of vector field data input from a user; a preprocessing unit that calculates gradient data of the plurality of vector field data and generates vector field similarity data based on the similarity of the calculated gradient data; a result analysis unit that includes a pre-learned artificial intelligence model, inputs the vector field similarity data into the artificial intelligence model, and infers an analysis result according to the change factor by the artificial intelligence model; and an output unit that displays a change factor inferred by the result analysis unit and an analysis result according to the change factor.

[0010] The above preprocessing unit can normalize the gradient data based on cosine similarity, sine similarity, or correlation.

[0011] The above preprocessing unit can produce the gradient data based on the average gradient of each pixel interval of the vector field data or the gradient between pixels included in a preset range.

[0012] The above preprocessing unit can produce the gradient data based on numerical differentiation between data points of the vector field data.

[0013] The above preprocessing unit can perform convolution on the partial derivative value according to the numerical differentiation based on a preset filter, and produce the gradient data based on the convolution.

[0014] The above preprocessing unit can convert the vector field data into 2D grid vector field data.

[0015] The above result analysis unit can learn the artificial intelligence model based on the vector field similarity data and label data generated by the above preprocessing unit.

[0016] The above result analysis unit can infer the analysis results including whether or not there is a change in the vector field according to the change factor.

[0017] The above result analysis unit can infer the analysis result as a probability value indicating the degree of change in the vector field according to the above change factor.

[0018] The above result analysis unit can visualize the analysis results of the artificial intelligence model based on Grad-CAM (gradient-weighted class activation mapping).

[0019] A vector field analysis method according to another disclosed embodiment includes: receiving a plurality of vector field data input from a user; calculating gradient data of the plurality of vector field data; generating a vector field similarity image based on a similarity of the calculated gradient data; inputting the vector field similarity image into a pre-trained artificial intelligence model, and then allowing the artificial intelligence model to infer a change factor and an analysis result according to the change factor; and displaying the change factor inferred by the artificial intelligence model and the analysis result according to the change factor.

[0020] Generating the above vector field similarity image may include normalizing the gradient data based on cosine similarity, sine similarity, or correlation.

[0021] Generating the vector field similarity image may include calculating the gradient data based on an average gradient of each pixel interval of the vector field data or a gradient between pixels included in a preset range.

[0022] Generating the vector field similarity image may include: calculating the gradient data based on numerical differentiation between data points of the vector field data; performing convolution on partial derivative values ​​according to the numerical differentiation based on a preset filter; and calculating the gradient data based on the convolution.

[0023] It may further include learning the artificial intelligence model based on the vector field similarity image and label data.

[0024] Inferring the above analysis results may include visualizing the analysis results including whether or not the vector field changes according to the change factor, the analysis results as a probability value indicating the degree of change in the vector field according to the change factor, or the analysis results of the artificial intelligence model based on Grad-CAM (gradient-weighted class activation mapping).

[0025] A preprocessing method for vector field analysis according to another disclosed embodiment includes: converting a plurality of vector field data input from a user into a 2D grid; calculating gradient data of the plurality of vector field data; normalizing the gradient data based on the similarity of the calculated gradient data; and generating vector field similarity data based on the normalized gradient data.

[0026] Generating the above vector field similarity data may include calculating the gradient data based on an average gradient of each pixel interval of the vector field data or a gradient between pixels included in a preset range.

[0027] The vector field analysis system and analysis method according to the disclosed embodiment can provide the possibility of utilizing artificial intelligence for vector field analysis by performing preprocessing capable of representing relationships between vector fields and then inputting the image data into artificial intelligence for classifying the image data.

[0028] In addition, the disclosed vector field analysis system and analysis method can provide expandability in research on the analysis of the relationship between vector fields and change factors in various research fields, such as analysis of cardiovascular disease occurrence factors and optimization of power generation facilities, which were previously analyzed only in terms of the relationship between physical values ​​derived from vector fields and change factors.

[0029] FIG. 1 is a diagram illustrating data collected by a vector field analysis system according to one disclosed embodiment.

[0030] Figure 2 is a control block diagram of the disclosed vector field analysis system.

[0031] Figure 3 is a drawing for explaining the disclosed vector field analysis method.

[0032] Figure 4 is a flowchart illustrating a preprocessing process for generating vector field similarity data according to one embodiment.

[0033] Figure 5 is a drawing for specifically explaining the flowchart of Figure 4.

[0034] FIG. 6 is a diagram for explaining a preprocessing process for generating vector field difference data according to another embodiment.

[0035] Figure 7 is a flowchart illustrating an embodiment in which the disclosed vector field analysis system outputs analysis results.

[0036] Figure 8 is a diagram illustrating an example of the analysis results.

[0037] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. However, the technical concept of the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided to ensure that the disclosure is thorough and complete and to sufficiently convey the spirit of the present invention to those skilled in the art.

[0038] In this specification, when a component is referred to as being on another component, it means that it can be formed directly on the other component, or a third component may be interposed between them. Furthermore, in the drawings, shapes and sizes are exaggerated for the purpose of effectively explaining the technical contents.

[0039] Additionally, although terms such as first, second, and third have been used to describe various components in various embodiments of this specification, these components should not be limited by these terms. These terms are only used to distinguish one component from another. Thus, what is referred to as a first component in one embodiment may be referred to as a second component in another embodiment. Each embodiment described and illustrated herein also includes its complementary embodiments. Additionally, the term "and / or" has been used herein to mean including at least one of the components listed before and after.

[0040] In the specification, singular expressions include plural expressions unless the context clearly dictates otherwise. In addition, terms such as "comprise" or "have" are intended to specify the presence of a feature, number, step, component, or combination thereof described in the specification, and should not be construed as excluding the possibility of the presence or addition of one or more other features, numbers, steps, components, or combinations thereof. In addition, in the present specification, "connection" is used to mean both indirectly connecting a plurality of components and directly connecting them.

[0041] In addition, when describing the present invention below, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

[0042] FIG. 1 is a diagram illustrating data collected by a vector field analysis system according to one disclosed embodiment.

[0043] Referring to Fig. 1, the vector field analysis system (1) can collect various vector field data from the outside. Specifically, the data that the vector field analysis system (1) can collect may include various data such as vector field data for wind (3), vector field data for blood flow (4), or other various wave fields (5).

[0044] Vector field data is utilized in various fields, such as wind power generation, automobile and aircraft design, climate change analysis, and blood flow analysis. The disclosed vector field analysis system (1) analyzes change factors that change the vector field and the resulting changes in vector field data, i.e., factors that affect the vector field. For example, in wind power generation, the design of a windmill blade can be a change factor, and the wind speed or wind volume resulting from the change in design can be vector field data.

[0045] The vector field analysis system (1) can collect a plurality of vector field data through a communication network (2) or receive input from a user. Specifically, the vector field analysis system (1) can be implemented as a computer or a portable terminal that can connect to the communication network (2). Here, the computer includes, for example, a notebook, a desktop, a laptop, a tablet PC, a slate PC, etc. having a web browser installed, and the portable terminal can include, for example, a wireless communication device that ensures portability and mobility, such as a PCS (Personal Communication System), a GSM (Global System for Mobile communications), a PDC (Personal Digital Cellular), a PHS (Personal Handyphone System), a PDA (Personal Digital Assistant), an IMT (International Mobile Telecommunication)-2000, a CDMA (Code Division Multiple Access)-2000, a W-CDMA (W-Code Division Multiple Access), a WiBro (Wireless Broadband Internet) terminal, a smart phone, etc., and a wearable device.

[0046] The communication network (2) includes various short-range, wired, and wireless networks. The communication network (2) can communicate by means of a communication module included in the vector field analysis system (1).

[0047] Specifically, the short-range communication module of the vector field analysis system (1) may include various short-range communication modules that transmit and receive signals using a wireless communication network at a short range, such as a Bluetooth module, an infrared communication module, an RFID (Radio Frequency Identification) communication module, a WLAN (Wireless Local Access Network) communication module, an NFC communication module, and a Zigbee communication module.

[0048] The wired communication module of the vector field analysis system (1) may include various wired communication modules such as a Controller Area Network (CAN) communication module, a Local Area Network (LAN) module, a Wide Area Network (WAN) module, or a Value Added Network (VAN) module, as well as various cable communication modules such as a Universal Serial Bus (USB), a High Definition Multimedia Interface (HDMI), a Digital Visual Interface (DVI), RS-232 (recommended standard 232), power line communication, or plain old telephone service (POTS).

[0049] The wireless communication module of the vector field analysis system (1) may include a wireless communication module that supports various wireless communication methods such as GSM (global System for Mobile Communication), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access), UMTS (universal mobile telecommunications system), TDMA (Time Division Multiple Access), and LTE (Long Term Evolution), in addition to a WiFi module and a WiBro (Wireless broadband) module.

[0050] Figure 2 is a control block diagram of the disclosed vector field analysis system.

[0051] Referring to FIG. 2, the vector field analysis system (1) includes a data receiving unit (11) that receives a plurality of vector field data, a control unit (10) that performs preprocessing of the received vector field data, compares the similarity of the preprocessed vector field data, and generates a degree of change in the vector field data (hereinafter referred to as an analysis result) according to a change factor, a storage unit (12) that stores the received vector field data, an artificial intelligence model used in the analysis operation of the control unit (10), and various vector field data, and an output unit (13) that outputs the analysis result.

[0052] Specifically, the data receiving unit (11) may include not only a communication module that receives various vector field data from the communication network (2) described above in FIG. 1, but also a user input command and a hardware configuration such as a USB (Universal Serial Bus).

[0053] The data receiving unit (11) may include hardware devices such as various buttons, switches, pedals, keyboards, mice, trackballs, various levers, handles, or sticks to receive user input, and may include at least one of a short-range communication module, a wired communication module, and a wireless communication module capable of communicating with a communication network (2).

[0054] The storage unit (12) can store not only the vector field data received by the data receiving unit (11), but also a plurality of artificial intelligence models operated by the control unit (10) and a large amount of data required for learning the artificial intelligence models. To this end, the storage unit (12) may be implemented as at least one of non-volatile memory devices such as cache, ROM (Read Only Memory), PROM (Programmable ROM), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), and flash memory, or volatile memory devices such as RAM (Random Access Memory), or storage media such as a hard disk drive (HDD, Hard Disk Drive), or CD-ROM, but is not limited thereto. The storage unit (12) is a memory implemented as a separate chip from the control unit (10) as illustrated in FIG. 2, but may also be implemented as a single chip with the control unit (10) if necessary.

[0055] The output unit (13) outputs the results analyzed by the control unit (10), that is, the presence or absence of change in vector field data according to the change factor, the degree of change, or the result visualizing the changing vector field. To this end, the output unit (13) may include various hardware devices such as a digital light processing (DLP) panel, a plasma display panel, a liquid crystal display (LCD) panel, an electroluminescence (EL) panel, an electrophoretic display (EPD) panel, an electrochromic display (ECD) panel, a light emitting diode (LED) panel, or an organic light emitting diode (OLED) panel.

[0056] Meanwhile, the output unit (13) may include a GUI (Graphical User Interface), i.e., a software device, such as a touch pad for receiving user input commands. The touch pad may be implemented as a touch screen panel (TSP) and may form a mutual layer structure with the data receiving unit (11).

[0057] The control unit (10) can be divided into a preprocessing unit (110) that performs preprocessing of a plurality of vector field data received by the data receiving unit (11), and a result analysis unit (120) that inputs vector field similarity image or vector field difference data performed by the preprocessing unit (110) into a learned artificial intelligence model and infers analysis results.

[0058] As mentioned above in the background of the invention, the conventional general Trial & Error-based vector field change factor analysis technique had several problems. In addition, the vector field analysis technique using artificial intelligence-based PINN (Physics-informed neural network) was limited to analyzing only the relationship between the physical values ​​derived from the vector field and the change factors. The disclosed control unit (10) can clearly and quickly derive the change factors and the change of the vector field according to the change factors by utilizing vector field image data for the analysis factors of the vector field and the change of the vector field according to the change factors. However, before inputting the vector field image data into the artificial intelligence model, the control unit (10) may perform a preprocessing process described below in order to ensure the clarity and speed described above.

[0059] Specifically, the preprocessing unit (110) performs preprocessing according to at least two embodiments.

[0060] A preprocessing unit (110) according to one embodiment calculates gradient data of input vector field data and generates vector field similarity data based on the similarity of the calculated gradient data. A preprocessing unit (110) according to another embodiment performs 2D gridding of vector field data and then calculates a difference value of the 2D gridded vector field data. The preprocessing unit (110) performs normalization of data including vector field difference values ​​between -1 and 1 (or between 0 and 1) and then generates vector field difference data.

[0061] In both of the aforementioned embodiments, vector field data can be normalized through the preprocessing unit (110). This normalization process has the effect of improving the learning performance of the artificial intelligence model used in the result analysis unit (120), which can ultimately lead to increased accuracy in analysis results based on change factors.

[0062] Meanwhile, the preprocessing unit (110) according to one embodiment may also perform 2D gridding of vector field data before generating gradient data. The specific process of generating vector field similarity data or difference data by the preprocessing unit (110) is described later through other drawings.

[0063] The result analysis unit (120) trains an artificial intelligence model to generate analysis results, and inputs vector field similarity data or difference data performed by the preprocessing unit (110) into the trained artificial intelligence model. The result analysis unit (120) generates result values ​​inferred by the artificial intelligence model, i.e., whether or not the vector field changes depending on the change factor, and the degree of change. In addition, the result analysis unit (120) can also visualize the analysis results.

[0064] Specifically, the artificial intelligence model used by the result analysis unit (120) may utilize various neural networks used in image analysis. For example, the result analysis unit (120) may utilize various image analysis models such as ResNet, CNN (Convolution Neural Network), GAN (Generative Adversarial Network), Faster R-CNN, SSD (Single Shot Multibox Detector), or YoLo (You Only Look Once).

[0065] The result analysis unit (120) performs supervised learning of an artificial intelligence model using label data containing change factors and vector field similarity data or difference data corresponding to the results. When learning is completed, the result analysis unit (120) inputs newly received vector field data (data on which a preprocessing process has been performed) into the artificial intelligence model, and can output the inferred result, i.e., the change factors of the newly input vector field data and the influence of the change factors, as probability values. The result analysis unit (120) transmits the inferred result to the output unit (13).

[0066] In addition to the preprocessing unit (110) and the result analysis unit (120), the control unit (10) may be implemented as a processor that executes an algorithm for controlling the operation of each component (data receiving unit, storage unit, output unit, etc.) of the vector field analysis system (1) or a program that reproduces the algorithm. A processor may refer to a data processing device built into hardware that has a physically structured circuit to perform a function expressed by a code or command included in a program. Examples of such data processing devices built into hardware may include, but are not limited to, processing devices such as a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), and a graphics processing unit (GPU). The control unit (10) may include one or more processors.

[0067] Meanwhile, the vector field analysis system (1) may include other configurations in addition to the configuration described above in FIG. 2, and each configuration may be integrated in hardware and implemented as one chip or memory.

[0068] Figure 3 is a drawing for explaining the disclosed vector field analysis method.

[0069] Referring to Fig. 3, the vector field analysis system (1) receives vector field data (200).

[0070] A vector field analysis system (1) according to one embodiment can perform 2D grid conversion of vector field data. The first vector field data (21) and the second vector field data (22) illustrated in FIG. 3 are both examples of vector field data converted into a 2D grid.

[0071] The vector field analysis system (1) performs preprocessing of 2D gridded vector field data (210).

[0072] The image (23) illustrated in Fig. 3 is vector field similarity data that is finally calculated by calculating the gradient of a two-dimensional vector field and calculating the similarity between the calculated gradient data. As will be described later in Fig. 4 and elsewhere, the vector field similarity data (23) is preprocessed so that the size of each pixel is between -1 and 1. That is, the normalized vector field similarity data (23) is applied to the learning and inference of the artificial intelligence model (121).

[0073] The vector field analysis system (1) inputs vector field similarity data (23) into a learned artificial intelligence model (121) and generates analysis results (220).

[0074] For example, an artificial intelligence model may include ResNet, which applies the concept of Residual Learning to solve the Degradation problem, which is the phenomenon in which performance decreases as the layers of a CNN become deeper.

[0075] The vector field analysis system (1) outputs the analysis results output by the artificial intelligence model (230).

[0076] Table (31) illustrated in Fig. 3 is a result value regarding whether a change occurred in the vector field being analyzed according to multiple change factors from 1 to n. That is, the analysis result of Fig. 3 may mean that the vector field did not change (0) according to the first change factor, and may mean that the vector field changed (1) according to the second change factor. In this way, the vector field analysis system (1) outputs whether or not a vector field changes according to each change factor, thereby enabling the user to clearly recognize the change factor that changes the vector field.

[0077] Figure 4 is a flowchart illustrating a preprocessing process for generating vector field similarity data according to one embodiment. Figure 5 is a diagram specifically illustrating the flowchart of Figure 4. To avoid redundant explanation, these two steps are described together below.

[0078] Referring to Fig. 4, the preprocessing unit (110) loads vector field data (211).

[0079] Step 211 means loading vector field data received by the data receiving unit (11) or stored in the storage unit (12) into the program.

[0080] The preprocessing unit (110) can perform 2D gridization of at least two or more vector field data (the first vector field data and the second vector field data of FIG. 5). However, the preprocessing unit (110) does not necessarily have to perform 2D gridization, and vector field similarity data (23) can be generated through the following operations without 2D gridization.

[0081] The preprocessing unit (110) produces gradient data (212).

[0082] The disclosed preprocessing unit (110) can produce gradient data by utilizing a numerical analysis method or a convolution operation-based method. Specifically, the preprocessing unit (110) can apply a numerical analysis method that applies numerical differentiation between pixels (or data points) of vector field data, or a convolution operation technique that applies convolution to partial derivative values ​​according to numerical differentiation.

[0083] Referring to FIG. 5, the first mathematical expression (24) may be an example in which the preprocessing unit (110) calculates gradient data through the slope between a pixel (data point) of a vector field image and its surrounding pixels. Here, the preprocessing unit (110) may be implemented to enable GPU parallel computing by applying a convolution operation technique that applies a filter (25). Meanwhile, the value of the filter may be arbitrarily set by the user.

[0084] When gradient data is produced, the preprocessing unit (110) produces vector field similarity based on the produced gradient data (213).

[0085] Referring to FIG. 5, the preprocessing unit (110) can use sine similarity, cosine similarity, or correlation between gradients to calculate the similarity between the first vector field data (21) and the second vector field data (22). The second mathematical expression (26) is an example of a calculation method using cosine similarity.

[0086] When the vector field similarity is calculated, the preprocessing unit (110) generates vector field similarity data (23) based on this and transmits this to the result analysis unit (120) (214).

[0087] The result analysis unit inputs vector field similarity data (23) into the artificial intelligence model and then generates the analysis results described later in Fig. 7. A detailed description of this will be provided later in other drawings.

[0088] FIG. 6 is a diagram for explaining a preprocessing process for generating vector field difference data according to another embodiment.

[0089] Referring to Fig. 6, the preprocessing unit (110) performs 2D gridization of the first vector field data (21) and the second vector field data (22). However, the preprocessing unit (110) does not necessarily have to perform 2D gridization, and the vector field difference data (27) can be generated through the following operations without 2D gridization.

[0090] The preprocessing unit (110) calculates the difference value between the first vector field data (21) and the second vector field data (22) (215).

[0091] Specifically, the preprocessing unit (110) calculates the vector field difference value (U diff) can be used to calculate the third mathematical expression (28), and the difference value can represent the size of each vector of each vector field data.

[0092] The preprocessing unit (110) performs normalization of the calculated difference value (216).

[0093] Specifically, the preprocessing unit (110) can use the fourth mathematical formula (29) to calculate the normalization, and the normalization can also apply various mathematical formulas in addition to the fourth mathematical formula (29). That is, the preprocessing unit (110) is sufficient to normalize the calculated difference value to between 0 and 1.

[0094] The preprocessing unit (110) transmits the normalized vector field difference data to the result analysis unit (120) (214).

[0095] Figure 7 is a flowchart illustrating an embodiment in which the disclosed vector field analysis system outputs analysis results. Figure 8 is a diagram illustrating an example of the analysis results. To avoid redundant explanation, they are described together below.

[0096] Referring first to FIG. 7, the vector field analysis system (1) receives vector field data (200), preprocesses the vector field (310), and generates analysis results (320).

[0097] As described above, the vector field analysis system (1) trains an artificial intelligence model and then performs preprocessing on the received new vector field data to normalize it to between 0 and 1 (or between -1 and 1). In particular, the preprocessing for generating vector field similarity data according to an example improves the data generation speed by using gradient data, and because it is normalized to a value between -1 and 1, the artificial intelligence model can recognize the characteristics of the vector field more clearly. As another example, the preprocessing for generating vector field difference data also improves the speed by generating the difference value by a simple mathematical formula, and because it includes a normalization process between -1 and 1, it is suitable for artificial intelligence model learning.

[0098] When preprocessed vector field data is input into an artificial intelligence model, the artificial intelligence model can output a result value (331) that classifies the presence or absence of change by change factor, a result value (332) that predicts the degree of change by change factor, or a result value (333) that visualizes the vector field analysis result, as learned.

[0099] The result value (331) classifying the presence or absence of change by change factor may include whether or not each change factor affects the change of the vector field while indicating multiple change factors. Referring to Fig. 8, the output unit (13) may output in a table (132) whether or not the vector field has changed or not according to change factors from 1 to n in the inference mode (131).

[0100] The result value (332) predicting the degree of change for each change factor represents the probability that each change factor can change the vector field, and the output unit (13) can also output a table (133) that displays all the probabilities that the vector field can change according to change factors from 1 to n as predicted values. In the table (133), the probability that the vector field changes due to change factor 1 is 0.179, and the probability that the vector field changes due to change factor 2 is 0.817, so change factor 2 can be analyzed as the main factor that changes the vector field.

[0101] The result value (333) visualizing the vector field analysis result is an application of a technique for visualizing the judgment basis of an image-based artificial intelligence model such as Grad-CAM (Gradient-weighted Class Activation Mapping). For example, the image (134) output by the output unit (13) shows the degree to which the vector field of the wind volume changes according to the design of the edge part of the target to be designed, and is an example of an image visualized so that additional analysis of the presence or absence of a change in the change factor is possible through various colors and color intensities.

[0102] The vector field analysis system (1) can also output a user interface (137) that can upload a visualized analysis image (134) so ​​that the user can utilize it.

[0103] The vector field analysis system (1) disclosed herein can provide the possibility of utilizing artificial intelligence in vector field analysis by performing preprocessing capable of representing the relationship between vector fields and then inputting the image data into an artificial intelligence that classifies the data. In addition, it can provide expandability in research on the analysis of the relationship between vector fields and change factors in various research fields such as analysis of cardiovascular disease occurrence factors and optimization of power generation facilities, which were previously analyzed only through the relationship between physical values ​​derived from vector fields and change factors.

Claims

1. A data receiving unit that receives multiple vector field data input from a user; A preprocessing unit that calculates gradient data of the plurality of vector field data and generates vector field similarity data based on the similarity of the calculated gradient data; A result analysis unit including a pre-learned artificial intelligence model, inputting the vector field similarity data into the artificial intelligence model, and inferring the analysis result according to the change factor by the artificial intelligence model; and A vector field analysis system including an output unit that displays change factors inferred by the above result analysis unit and analysis results according to the change factors.

2. In paragraph 1, The above preprocessing unit, A vector field analysis system that normalizes the gradient data based on cosine similarity, sine similarity or correlation.

3. In paragraph 1, The above preprocessing unit, A vector field analysis system that produces the gradient data based on the average gradient of each pixel interval of the vector field data or the gradient between pixels included in a preset range.

4. In paragraph 2, The above preprocessing unit, A vector field analysis system that produces the gradient data based on numerical differentiation between data points of the above vector field data.

5. In paragraph 4, The above preprocessing unit, A vector field analysis system that performs convolution on partial derivative values ​​according to the numerical differentiation based on a preset filter and produces the gradient data based on the convolution.

6. In paragraph 1, The above preprocessing unit, A vector field analysis system that converts the above vector field data into 2D grid vector field data.

7. In paragraph 1, The above results analysis section, A vector field analysis system that learns the artificial intelligence model based on vector field similarity data and label data generated by the above preprocessing unit.

8. In paragraph 1, The above results analysis section, A vector field analysis system that infers the analysis results including whether or not a vector field changes according to the above change factors.

9. In paragraph 1, The above results analysis section, A vector field analysis system that infers the analysis results as probability values ​​representing the degree of change in the vector field according to the above change factors.

10. In paragraph 1, The above results analysis section, A vector field analysis system that visualizes the analysis results of the artificial intelligence model based on Grad-CAM (gradient-weighted class activation mapping).

11. Receive multiple vector field data input from the user; Calculating gradient data of the plurality of vector field data and generating a vector field similarity image based on the similarity of the calculated gradient data; After inputting the above vector field similarity image into a pre-learned artificial intelligence model, the artificial intelligence model infers the change factor and the analysis result according to the change factor; A vector field analysis method comprising: displaying change factors inferred by the artificial intelligence model and analysis results according to the change factors.

12. In paragraph 11, Generating the above vector field similarity image is: A vector field analysis method comprising normalizing the gradient data based on cosine similarity, sine similarity or correlation.

13. In paragraph 11, Generating the above vector field similarity image is: A vector field analysis method comprising: calculating the gradient data based on the average gradient of each pixel interval of the vector field data or the gradient between pixels included in a preset range; 14. In paragraph 11, Generating the above vector field similarity image is: The gradient data is calculated based on the numerical differentiation between data points of the above vector field data; A vector field analysis method comprising: performing convolution on partial derivative values ​​according to the numerical differentiation based on a preset filter, and calculating the gradient data based on the convolution.

15. In paragraph 11, A vector field analysis method further comprising: learning the artificial intelligence model based on the vector field similarity image and label data.

16. In paragraph 11, Inferring from the above analysis results, A vector field analysis method comprising: visualizing the analysis results of the artificial intelligence model based on the analysis results or Grad-CAM (Gradient-weighted Class Activation Mapping), including the presence or absence of change in the vector field according to the change factor, as a probability value indicating the degree of change in the vector field according to the change factor.

17. Convert multiple vector field data input from the user into a 2D grid; Calculating gradient data of the above plurality of vector field data; Normalizing the gradient data based on the similarity of the generated gradient data; A preprocessing method for vector field analysis, comprising: generating vector field similarity data based on the above normalized gradient data.

18. In paragraph 17, Generating the above vector field similarity data is as follows: A vector field preprocessing method comprising: calculating the gradient data based on the average gradient of each pixel interval of the vector field data or the gradient between pixels included in a preset range.

Images