Prediction method, prediction apparatus, and program

Abstract

To efficiently perform prediction of a temperature distribution of a target object such as a substrate.SOLUTION: A prediction method is a method for predicting a temperature distribution of a target object on which one or more components are disposed by means of an AI model, acquires each physical property parameter of a component and the target object, and first mesh data in which the component and the target object are disposed on meshes of a prescribed number of meshes previously designated for the AI model, performs a mesh division process of dividing the meshes according to the other position when there is an instruction to move the component to another position on the target object, generates second mesh data in which the component and the target object are disposed on the meshes after the mesh division process, predicts and outputs the temperature distribution by inputting each physical property parameter and the second mesh data to the AI model.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to a prediction method, a prediction device, and a program. [Background technology]

[0002] Patent Document 1 discloses a hexahedral mesh generation support device that generates hexahedral meshes for analysis used in numerical analysis simulations such as the finite element method, which are performed during the product development and design stage. This hexahedral mesh generation support device inputs multiple hexahedral mesh data, which are information on hexahedral meshes to be connected, formed by nodes and edges connecting the nodes, and connects the hexahedral meshes by matching the mesh patterns at the joints if the mesh patterns at the joints do not match. The hexahedral mesh generation support device also generates continuously connected hexahedral mesh data by modifying the hexahedral mesh data based on the meshes at the joints with the matched mesh patterns. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-215662 Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, there has been a demand for faster processing speeds and more sophisticated functionality in electronic devices. However, control boards and other components built into the housings of electronic devices are often equipped with numerous heat-generating elements, such as electronic components that generate heat when power is supplied. This heat can potentially slow down processing speeds. Therefore, during the design phase of electronic device development, there is a need to accurately evaluate the heat generated on a board based on the placement of electronic components to be mounted on the board. The temperature distribution on a board with heat-generating elements can be predicted using simulation techniques such as Computational Fluid Dynamics (CFD), but when a board contains multiple heat-generating elements, the analysis can take an extremely long time, from tens of minutes to tens of hours.

[0005] As a replacement for CFD, AI surrogate models have emerged that use artificial intelligence (AI) to learn from the calculation results of simulations such as CFD and make predictions. However, the use of these AI surrogate models presents the following challenges, for example: Specifically, when using an AI surrogate model to predict the temperature distribution caused by a heat-generating element placed on a circuit board, the position of the heat-generating element (electronic component) that served as the basis for the learning results may be changed (i.e., moved) to another position on the circuit board during product development and design.

[0006] However, if the position of a heating element is changed (moved), the heating element on the board may not be located on the mesh line that is the target of prediction by the AI ​​surrogate model, which poses a problem that prediction by the AI ​​surrogate model is not possible. Patent Document 1 discloses a technique for adjusting and aligning the node positions of two different hexahedral meshes when joining them, but does not disclose a simulation technique that takes into account the inability of the AI ​​surrogate model to make predictions when the position of the heating element is changed as described above.

[0007] The present disclosure has been devised in view of the above-described conventional situation, and aims to efficiently predict the temperature distribution of an object such as a substrate. [Means for solving the problem]

[0008] The present disclosure provides a method for predicting the temperature distribution of an object on which one or more parts are arranged using an AI model, the method acquiring physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh with a predetermined number of meshes pre-specified in the AI ​​model, and when an instruction is given to move the part to another position on the object, performing a division process to shape the mesh or the part in accordance with the other position, generating second mesh data in which the part and the object are arranged in the mesh after the division process, and inputting each of the physical property parameters and the second mesh data into the AI ​​model, thereby predicting and outputting the temperature distribution.

[0009] The present disclosure also provides a prediction device that predicts the temperature distribution of an object on which one or more parts are arranged using an AI model, the prediction device including a processor and a memory, wherein the processor, in cooperation with the memory, acquires physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh with a predetermined number of meshes pre-specified in the AI ​​model, and, when an instruction is given to move the part to another position on the object, performs a division process to shape the mesh or the part in accordance with the other position, generates second mesh data in which the part and the object are arranged in the mesh after the division process, and inputs the physical property parameters and the second mesh data into the AI ​​model, thereby outputting a prediction result of the temperature distribution.

[0010] The present disclosure also provides a program for causing a prediction device, which is a computer, to execute the following steps: a process for predicting the temperature distribution of an object on which one or more parts are arranged using an AI model; a process for acquiring physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh with a predetermined number of meshes pre-specified in the AI ​​model; a process for, when an instruction to move the part to another position on the object is received, performing a division process to shape the mesh or the part in accordance with the other position, and generating second mesh data in which the part and the object are arranged in the mesh after the division process; and a process for inputting the physical property parameters and the second mesh data into the AI ​​model, and outputting a prediction result of the temperature distribution.

[0011] These comprehensive or specific aspects may be realized as a system, an apparatus, a method, an integrated circuit, a computer program, or a recording medium, or may be realized as any combination of a system, an apparatus, a method, an integrated circuit, a computer program, and a recording medium. [Effects of the Invention]

[0012] According to the present disclosure, it is possible to efficiently predict the temperature distribution of an object such as a substrate. [Brief explanation of the drawings]

[0013] [Figure 1] Conceptual diagram of temperature distribution prediction using AI [Figure 2] FIG. 1 is a block diagram illustrating an example of a hardware configuration of a prediction device according to an embodiment of the present invention. [Figure 3] A flowchart showing an example of the overall operation procedure of a prediction tool executed by a processor and an AI processor of a prediction device in chronological order. [Figure 4] A flowchart showing the detailed operation procedure of the prediction tool in chronological order when there is a movement of the part to be predicted. [Figure 5] An example of the setting display screen for the prediction tool [Figure 6]An example of a screen displaying the predicted temperature distribution of an object using a prediction tool. [Figure 7A] A schematic diagram showing an example of the processing image of mesh division method 1 for a part to be predicted. [Figure 7B] A schematic diagram showing another example of processing image of the mesh division method 1 for the part to be predicted. [Figure 8] A flowchart showing a detailed procedure example of mesh division method 1 in chronological order in step St13 of FIG. [Figure 9] A flowchart showing a detailed procedure example of mesh division method 1 in chronological order in step St13 of FIG. [Figure 10] A schematic diagram showing an example of the processing image of mesh division method 2 for the part to be predicted. [Figure 11] A flowchart showing a detailed procedure example of mesh division method 2 in chronological order in step St14 of Figure 4. [Figure 12] A diagram showing a schematic image of the processing image of the third example of operation for mesh division of the part to be predicted. [Figure 13] A diagram showing a schematic image of the processing image of the third example of operation for mesh division of the part to be predicted. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, with reference to the drawings as appropriate, embodiments specifically disclosing the prediction method, prediction device, and program according to the present disclosure will be described in detail. However, more detailed description than necessary may be omitted. For example, detailed description of well-known matters and redundant description of substantially identical configurations may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art. Note that the accompanying drawings and the following description are provided to enable those skilled in the art to fully understand the present disclosure, and are not intended to limit the subject matter recited in the claims.

[0015] First, an overview of predicting temperature distribution using AI will be explained with reference to Figure 1. Figure 1 is a conceptual diagram of predicting temperature distribution using AI.

[0016] Conventionally, the temperature distribution of an object such as a substrate containing a heat-generating element is predicted using simulation techniques such as Computational Fluid Dynamics (CFD) using physical property parameters such as the heat generation amount, thermal conductivity, and emissivity of the heat-generating element. However, predicting the temperature distribution using CFD has the problem that it takes an extremely long time, for example, several hours to several tens of hours, to perform the calculations.

[0017] For example, the target object may be a substrate including multiple elements that are heat-generating bodies. The elements may be, for example, coils, capacitors, diodes, integrated circuits (ICs), or large-scale integration (LSIs). The number, size, and type of elements included in a substrate vary depending on the application, and analyzing the temperature distribution for each substrate using simulation technology would take an enormous amount of time. Note that the target object is not limited to a substrate, and may be any electronic device that includes a heat-generating body, such as an engine or a product that includes a heat-generating body (e.g., a smartphone, a personal computer, etc.).

[0018] Conventional analysis methods take a long time to predict temperature distribution, but by using AI technologies such as Convolutional Neural Networks (CNN), it may be possible to significantly reduce the time required to predict the temperature distribution of an object, from a few seconds to a few tens of seconds.

[0019] Information including physical property parameters such as the heat generation amount of a heating element included in the object, the in-plane thermal conductivity around the heating element, the out-of-plane thermal conductivity and emissivity, etc. When the object is viewed in three dimensions, for example, the input data is information in which physical property parameters are associated with each of the meshes divided into squares by vertical and horizontal mesh division lines in the x-, y-, and z-axis directions.

[0020] By inputting such input data into an AI model, the temperature distribution predicted by the AI ​​model is calculated. Unlike physical models, AI models are generated by machine learning using training data that includes physical property parameters and the predicted temperature distribution. For example, the AI ​​model is a surrogate model (hereinafter referred to as an "AI surrogate model") that uses input data to predict the temperature distribution of an object in which a heating element is located. The physical model referred to here refers to a simulation technology that uses classical physics, such as CFD. By using an AI surrogate model to calculate the predicted temperature distribution of an object (hereinafter referred to as the "predicted temperature distribution"), analysis time can be significantly reduced. Below, an AI surrogate model will be used as an example of an AI model for calculating the temperature distribution of an object.

[0021] However, the use of AI surrogate models is thought to pose the following challenges. Specifically, when using an AI surrogate model to predict the temperature distribution caused by a heating element placed on a circuit board, the position of the heating element (electronic component) that served as the basis for the learning results may be changed (i.e., moved) to another position on the circuit board during product development and design. This can occur when, in order to determine where to place a heating element on a circuit board, developers and others refer to the temperature distribution that would occur if the heating element were placed in the current location to determine whether or not there are any problems with that location. However, when the position of the heating element is changed (moved), the heating element on the circuit board may not be located on the mesh line (mesh division line) that is the target of prediction by the AI ​​surrogate model, which poses the challenge of making predictions using the AI ​​surrogate model.

[0022] More specifically, an AI surrogate model has a predetermined number of meshes (e.g., X-axis: 256, Y-axis: 256, Z-axis: 128, etc.), and unless both the heating element and the target object, such as the circuit board, are positioned within the meshes, the reliability of the temperature distribution obtained by the AI ​​surrogate model cannot be ensured. In other words, to ensure the reliability of the temperature distribution obtained by the AI ​​surrogate model, developers and others need to move components so that they are positioned within the meshes of the specified number of meshes of the AI ​​surrogate model. However, this is considered unrealistic in product development and design. Therefore, there is a need for a technology that can predict a reliable temperature distribution even when components need to be moved, without requiring developers and others to take careful care in positioning the moved components within the meshes.

[0023] 1. Configuration of the prediction device Next, with reference to FIG. 2, an example configuration of the prediction device 10 according to the present embodiment will be described. FIG. 2 is a block diagram showing an example hardware configuration of the prediction device 10 according to the present embodiment. In FIG. 2, the prediction device 10 and the external database 20 are illustrated as separate entities, but a configuration equivalent to the external database 20 may be built into the prediction device 10. Although not clearly shown in FIG. 2, the prediction device 10 and the external database 20 may be directly connected to each other so as to enable data communication, or may be connected via a network so as to enable data communication. For ease of explanation, an example in which the prediction device 10 and the external database 20 are directly connected to each other so as to enable data communication will be described below.

[0024] The prediction device 10 predicts the temperature distribution of an object, such as a circuit board, on which components including one or more heat-generating elements are arranged, using an AI surrogate model. The prediction device 10 is a computer device, such as a desktop personal computer or a stationary server computer. Note that the prediction device 10 is not limited to these computer devices and may be a portable computer device, such as a smartphone or a tablet terminal. The prediction device 10 includes a communication I / F 11, a memory 12, an input device 13, a display device 14, a processor 15, and an AI processor 16.

[0025] The communication I / F 11 is an interface circuit that performs wireless or wired communication between the prediction device 10 and the external database 20. Here, I / F represents an interface. The communication between the prediction device 10 and the external database 20 may be via a network. The communication method used by the communication I / F 11 is, for example, a Wide Area Network (WAN), a Local Area Network (LAN), a Long Term Evolution (LTE), a mobile communication such as 4G or 5G, a power line communication, a short-range wireless communication (e.g., Bluetooth (registered trademark) communication), or a communication for a mobile phone.

[0026] The memory 12 is configured using, for example, Random Access Memory (RAM) and Read Only Memory (ROM), and temporarily stores programs necessary for the operation of the prediction device 10 and data acquired or generated during operation. The RAM is, for example, a work memory used during the operation of the prediction device 10. The ROM stores and holds, for example, programs for controlling the prediction device 10 in advance.

[0027] The input device 13 is a device that receives input from a user of the prediction device 10. The input device 13 may be, for example, a mouse, a keyboard, a touch panel, or a combination of these.

[0028] The display device 14 displays the prediction result (see FIG. 6) executed by the prediction device 10. The prediction result is, for example, a prediction result of the temperature distribution of an object predicted by the AI ​​processor 16 using an AI surrogate model. The display device 14 is, for example, a display such as a Liquid Crystal Display (LCD) or an organic EL display. The input device 13 and the display device 14 may be configured integrally, in which case the input device 13 is a touch panel display or the like.

[0029] The processor 15 is configured by at least one of, for example, a central processing unit (CPU), a digital signal processor (DSP), or a field programmable gate array (FPGA). The processor 15 functions as a controller that manages the overall operation of the prediction device 10. The processor 15 performs control processing for overseeing the operation of each part of the prediction device 10, data input / output processing between each part of the prediction device 10, data arithmetic processing, and data storage processing. The processor 15 operates according to a program stored in the memory 12. The processor 15 uses the memory 12 during operation, and temporarily stores data generated or acquired by the processor 15 in the memory 12. In the prediction device 10, the processor 15 and the AI ​​processor 16 each work in cooperation with the memory 12 to function as a prediction tool (in other words, application software) that predicts the temperature distribution of an object.

[0030] The AI ​​processor 16 constitutes part of the prediction tool (see above) and is configured, for example, by a Graphical Processing Unit (GPU). However, the AI ​​processor 16 is not limited to being configured by a GPU. The AI ​​processor 16 can use any of multiple AI surrogate models to perform the corresponding processing. The AI ​​processor 16 executes processing using one of the multiple AI surrogate models selected by the processor 15. Based on instructions from the processor 15, the AI ​​processor 16 reads and acquires input data (see FIG. 1) from the memory 12, and then executes the selected AI surrogate model to predict the temperature distribution of an object, such as a substrate on which one or more heating elements are arranged. The AI ​​processor 16 sends the predicted temperature distribution of the object to the processor 15.

[0031] The external database 20 is configured by a storage device such as a Database Management System (DBMS) or a hard disk, and stores and accumulates data on the prediction results such as the temperature distribution of the object calculated by the prediction device 10.

[0032] 2. Overall operation procedure of the prediction tool Next, an example of the overall operation procedure of the prediction tool executed by the prediction device 10 according to this embodiment and an example of a detailed operation procedure when a part is moved on the target object will be described with reference to Figures 3, 4, 5, and 6. Figure 3 is a flowchart showing an example of the overall operation procedure of the prediction tool executed by the processor 15 and AI processor 16 of the prediction device 10 in chronological order. Figure 4 is a flowchart showing an example of the detailed operation procedure of the prediction tool in chronological order when a part to be predicted is moved. Figure 5 is a diagram showing an example of a setting display screen for the prediction tool. Figure 6 is a diagram showing an example of a screen displaying the results of a prediction of the temperature distribution of the target object by the prediction tool.

[0033] In FIG. 3, the processor 15 instructs the launch of a prediction tool, which is application software, based on a user operation using, for example, the input device 13 (step St1). This allows the prediction tool, which is application software pre-installed in an executable state in the prediction device 10, to be launched. In the descriptions of FIGS. 3 to 6, each process is described as being executed by the prediction tool. However, since the prediction tool can be executed by at least one of the processor 15 and the AI ​​processor 16, the "prediction tool" may be replaced with "at least one of the processor 15 and the AI ​​processor 16." In order to predict the temperature distribution of the object, the prediction tool reads and acquires one or more heating elements and physical property parameters (e.g., shape, condition) of the object and first mesh data from the memory 12 (step St2). Note that, as shown in FIG. 3, the series of processes from step St2 to step St6 corresponds to pattern 1, in which the movement of a component, such as a heating element, on an object, such as a substrate, is not instructed by a user operation. Pattern 2, in which the movement of a component, such as a heating element, on an object, such as a substrate, is instructed by a user operation, will be described later with reference to FIG. 4.

[0034] "Shape" refers to data on the shape and size (dimensions) of each heating element and object that can be adopted in the product development and design stage. The "shape" data may be stored in advance in memory 12 or may be obtained from an external database 20 by a prediction tool.

[0035] The "conditions" include data on the physical parameters (e.g., heat generation amount, emissivity, and thermal conductivity) of the heating element and the object. Here, as described with reference to FIG. 1, the data on each physical parameter is stored in association with each mesh of a predetermined number of meshes included in the first mesh data. The "conditions" data, like the "shape" data, may be stored in advance in the memory 12 or may be acquired by a prediction tool from the external database 20. In addition to the above-described physical parameters, the "conditions" may also include data on the importance (i.e., weighting coefficients) of components that affect the object by generating heat when predicting the temperature distribution of the object. The importance of the heating element is "3," the importance of metal is "2," and the importance of other components is "1." The "conditions" may include factors other than the above-described importance.

[0036] The "first mesh data" refers to a predetermined number of meshes (target number of meshes) specified in advance for each AI surrogate model. For example, to predict the two-dimensional temperature distribution of an object with a heating element disposed thereon, the input data for the AI ​​surrogate model must match the arrangement of the heating element and the object to one of 64 meshes in the x direction (horizontal direction) and 64 meshes in the y direction (vertical direction). In this case, the first mesh data associates the identification information of the AI ​​surrogate model to be used, the total number of meshes (i.e., 64 x 64 = 4096), and the mesh width, which is the spacing between adjacent meshes. Note that the adjacent mesh widths in all meshes included in the first mesh data do not necessarily have to be the same, but may be the same.

[0037] The prediction tool selects one AI surrogate model MDL1 (see FIG. 5) from among multiple AI surrogate models based on a user operation using the input device 13 (step St3). The prediction tool selects a physical quantity to be predicted (in other words, a physical quantity to be displayed as a prediction result) using the AI ​​surrogate model MDL1 (see FIG. 5) selected in step St3 based on a user operation using the input device 13 (step St4). Here, for example, it is assumed that "temperature distribution" is selected as the physical quantity. The prediction tool predicts the physical quantity (e.g., temperature distribution) selected in step St4 by inputting the data on the physical property parameters (e.g., shape and conditions) of the heating element and the target object and the first mesh data read in step St2 into the AI ​​surrogate model MDL1 (see FIG. 5) selected in step St3 (step St5).

[0038] The prediction tool outputs and displays the physical quantity (for example, the temperature distribution of the object such as a substrate on which a heating element is arranged) obtained by the prediction in step St5 on the display device 14 (step St6, see, for example, FIG. 6).

[0039] 4, it is assumed that after step St4 in FIG. 3, the prediction tool receives an instruction to move a component that is a heat generating element on an object such as a circuit board based on a user operation using the input device 13 (step St11). As a result, the component that is a heat generating element on the object such as a circuit board is moved to a different position (an example of another position) from the position before the movement, and the coordinate values ​​of the component are changed accordingly. Therefore, as pattern 2, the prediction tool performs a division process to reshape the mesh or component according to the position of the moved component so that the moved component exists (is located) on a mesh with the number of meshes specified by the AI ​​surrogate model.

[0040] In this embodiment, two types of division processing are provided for "accuracy-oriented" patterns (specifically, mesh division method 1 shown in Figs. 8 and 9, and mesh division method 2 shown in Fig. 11), and one type for "real-time-oriented" patterns (specifically, the part shaping method shown in Figs. 12 and 13). The prediction tool can execute both the mesh division method processing for the "accuracy-oriented" pattern described above and the part shaping method processing for the "real-time-oriented" pattern described above.

[0041] The prediction tool accepts a selection of either "emphasis on accuracy" or "emphasis on real-time" based on a user operation using the input device 13 (step St12). This selection can be realized, for example, by a user operation of one of the selection icons BT2 displayed in the third designation area PRT3 of the setting screen WD1 shown in FIG. 5. In the illustrated example of FIG. 5, "emphasis on real-time" has been selected, and the prediction tool executes processing of the component shaping method for the "emphasis on real-time" pattern based on this selection. A detailed description of FIG. 5 will be given later.

[0042] The prediction tool accepts the selection of "accuracy-oriented" mesh division method 1 (for example, mesh division method 1, which will be described later with reference to FIGS. 8 and 9) through a user operation, executes processing for this "accuracy-oriented" mesh division method 1, and generates second mesh data by editing (correcting) the first mesh data in accordance with the movement of parts (step St13). Details of the processing of step St13 will be described later with reference to FIGS. 7A, 7B, 8, and 9. Note that in the illustrated example of FIG. 5, only one "accuracy-oriented" mesh division method can be selected (i.e., only one of mesh division method 1 and mesh division method 2 can be selected), but selection icons for mesh division method 1 and mesh division method 2 may be displayed and arranged in the third designation region PRT3. Note that mesh division method 1 can be selected when a more accurate mesh division method is achieved by adding more meshes than the number of meshes specified in the AI ​​surrogate model MDL1 (see FIG. 5) selected in step St4. The prediction tool inputs the second mesh data generated in step St13 and the data on the physical property parameters (e.g., shape, conditions) of the heating element and the target object read in step St2 into the AI ​​surrogate model MDL1 (see Figure 5) selected in step St3, thereby predicting the physical quantity (e.g., temperature distribution) selected in step St4 (step St17).

[0043] Meanwhile, the prediction tool accepts a selection of "accuracy-oriented" (for example, mesh division method 2, which will be described later with reference to FIG. 11) through a user operation, executes processing for this "accuracy-oriented" mesh division method 2, and generates second mesh data by editing (correcting) the first mesh data in accordance with the movement of the parts (step St14). Details of the processing of step St14 will be described later with reference to FIGS. 10 and 11. The prediction tool predicts the physical quantity (for example, temperature distribution) selected in step St4 by inputting the second mesh data generated in step St14 and the data of the physical property parameters (for example, shape and conditions) of the heating element and the target object read in step St2 into the AI ​​surrogate model MDL1 (see FIG. 5) selected in step St3 (step St17).

[0044] Meanwhile, the prediction tool accepts the selection of "real-time oriented" (for example, the part shaping method shown in FIGS. 12 and 13) by user operation, and executes processing of this "real-time oriented" part shaping method (step St15). Details of the processing of step St15 will be described later with reference to FIGS. 12 and 13. Furthermore, the prediction tool corrects the data of the physical property parameters (for example, heat generation amount, emissivity, thermal conductivity) of the corresponding part read in step St2 in association with part shaping, using the processing result of step St15 (step St16). The prediction tool predicts the physical quantity (for example, temperature distribution) selected in step St4 by inputting the physical property parameters of the part corrected in step St16, the data of each physical property parameter (for example, shape, conditions) of the target object read in step St2, and the first mesh data into the AI ​​surrogate model MDL1 (see FIG. 5) selected in step St3 (step St17).

[0045] The prediction tool outputs and displays the physical quantity (for example, the temperature distribution of the object such as a substrate on which a heating element is arranged) obtained by the prediction in step St17 on the display device 14 (step St18, see, for example, FIG. 6).

[0046] In FIG. 5, the setting screen WD1 has a first designated region PRT1, a second designated region PRT2, and a third designated region PRT3.

[0047] The first specification area PRT1 hierarchically displays check boxes for specifying whether or not to include a target in the prediction targets predicted by the AI ​​surrogate model MDL1. The check boxes are displayed in association with the target (e.g., a circuit board) and the names of various components arranged on the target.

[0048] The second designation area PRT2 displays data related to the attributes of the object itself, such as a component or circuit board, that has been designated as a prediction target by a user operation using a checkbox in the first designation area PRT1. As shown in Fig. 5, the attributes include the component name (or the object name), material, heat generation amount, coordinate values ​​on the object (specifically, x-coordinate value, y-coordinate value, z-coordinate value), and size (dimensions, specifically, length in the x-direction, length in the y-direction, and length in the z-direction). For example, a user operation on the second designation area PRT2 instructs the prediction device 10 to move the component that is to be moved.

[0049] The third specification region PRT3 displays an appearance image IPT1 in which one or more components PRs designated as prediction targets predicted by the AI ​​surrogate model MDL1 are three-dimensionally arranged on an object (e.g., a board SUB1), selection icons BT2, and a color bar CB1. The selection icons BT2 include an icon for specifying the processing of the above-mentioned "accuracy-oriented" mesh division method 1 or mesh division method 2 ("Accuracy" in the example of FIG. 5) and an icon for specifying the processing of the above-mentioned "real-time-oriented" component shaping method ("Real Time" in the example of FIG. 5). The color bar CB1 is a legend indicating the heat generation state when the prediction results, such as the temperature distribution of one or more components PRs and the board SUB1 indicated in the appearance image IPT1, are displayed as a heat map (see FIG. 6). The prediction tool predicts specified physical quantities (e.g., temperature distribution, flow, pressure, etc.) by inputting into the AI ​​surrogate model MDL1 physical property parameters (e.g., heat generation rate, emissivity, thermal conductivity for each mesh) and first or second mesh data (e.g., number of meshes, mesh width) specified by user operation in each of the first to third specified areas PRT1 to PRT3 on the setting screen WD1 shown in Fig. 5. The prediction tool outputs and displays the prediction results on the prediction result display screen WD2 (see Fig. 6).

[0050] 6, the prediction result display screen WD2 has at least a first specified area PRT1 and a prediction result area PRT4. The first specified area PRT1 is the same as that shown in FIG.

[0051] The prediction result region PRT4 displays the prediction result RST1 of the physical quantities (e.g., temperature distribution) of the board SUB1 on which the components PRs are placed, as predicted by the AI ​​surrogate model MDL1. The color bar CB1 is the same as that shown in FIG. 5, so its explanation is omitted. The prediction result display screen WD2 shown in FIG. 6 allows a user such as a developer to grasp at a bird's-eye view the heat generation state of the entire board SUB1, as the temperature distribution around heat-generating elements (e.g., IC chips) or metals placed among the components PRs is high.

[0052] 3. Detailed operation procedure for precision-oriented mesh division method 1 Next, detailed operational procedures of mesh division method 1 in step St13 in Fig. 4 will be described with reference to Figs. 7A, 7B, 8, and 9. Fig. 7A is a diagram schematically showing an example of a processing image of mesh division method 1 for a part to be predicted. Fig. 7B is a diagram schematically showing another example of a processing image of mesh division method 1 for a part to be predicted. Figs. 8 and 9 are flowcharts showing detailed procedure examples of mesh division method 1 in step St13 in Fig. 4 in chronological order.

[0053] 7A shows an example of the results of mesh division of part Pa1 according to mesh division method 1. The example of FIG. 7A shows a case where the shape of part Pa1 is parallel to both the x-axis and the y-axis. For ease of understanding, part Pa1 is described as being represented by a two-dimensional model (i.e., two orthogonal axes, the x-axis and the y-axis), but the description of FIG. 7A can also be applied to a case where part Pa1 is represented by a three-dimensional model (i.e., three orthogonal axes, the x-axis, the y-axis, and the z-axis).

[0054] A first mesh division line Mx1a in the x-axis direction and a first mesh division line My1a in the y-axis direction (an example of a mesh division line) are arranged as a processing result of mesh division method 1 so as to tangent to and cover the outer shape of part Pa1. These first mesh division lines Mx1a and My1a are mesh division lines arranged according to the axial planes. An axial plane is a surface perpendicular to each coordinate axis (specifically, the x-axis, y-axis, and z-axis) (specifically, the axial plane which is the yz plane, the zx plane, and the xy plane). In other words, the first mesh division lines Mx1a and My1a are mesh division lines for dividing part Pa1 on each axial plane based on a bounding box which covers the outer shape of part Pa1.

[0055] 7A, a second mesh division line My1b (an example of a mesh division line) for dividing a first mesh division line (for example, a first mesh division line My1a adjacent in the y-axis direction) into multiple lines is arranged as a processing result of the mesh division method 1. Similarly, a second mesh division line Mx1b (an example of a mesh division line) for dividing a first mesh division line (for example, a first mesh division line My1a adjacent in the x-axis direction, although not shown in FIG. 7A) into multiple lines is arranged as a processing result of the mesh division method 1. These second mesh division lines Mx1b and My1b are mesh division lines arranged according to their respective limit widths. The limit width is the spacing (i.e., the mesh width) between adjacent mesh division lines arranged as a processing result of the mesh division by the mesh division method 1. In other words, when performing mesh division processing, the predictive tool determines the arrangement of the mesh division lines while restricting the mesh width so that it does not become smaller than the limit width.

[0056] Fig. 7B shows an example of the results of mesh division of part Pa2 according to mesh division method 1. The example of Fig. 7B shows a case where the shape of part Pa1 is not necessarily parallel to both the x-axis and the y-axis. For ease of understanding, part Pa2 is described as being represented by a two-dimensional model (i.e., two orthogonal axes, the x-axis and the y-axis), but the description of Fig. 7B can also be applied to a case where part Pa1 is represented by a three-dimensional model (i.e., three orthogonal axes, the x-axis, the y-axis, and the z-axis).

[0057] In the example of FIG. 7B, the outer shape of part Pa2 is parallel to the x-axis direction but non-parallel to the y-axis direction. However, as in the example of FIG. 7A, the prediction tool places a first mesh division line Mx2a in the x-axis direction and a first mesh division line My2a in the y-axis direction (an example of a mesh division line) as a processing result of mesh division method 1 so as to contact and cover part Pa2. Furthermore, the prediction tool places a second mesh division line My2b (an example of a mesh division line) for dividing the first mesh division line (for example, the first mesh division line My2a adjacent in the y-axis direction) into multiple parts as a processing result of mesh division method 1. Similarly, the prediction tool places a second mesh division line Mx2b (an example of a mesh division line) for dividing the first mesh division line (for example, the first mesh division line Mx2a adjacent in the x-axis direction, not shown in FIG. 7B) into multiple parts as a processing result of mesh division method 1. However, at this time, the prediction tool shapes and places part Pa2, like part Pa2m, so that its outer shape fits within the mesh between the first mesh division line and the second mesh division line. In other words, the first mesh division lines Mx2a and My2a become mesh division lines for dividing part Pa2m on each axial plane based on a bounding box that covers the outer shape of part Pa2m.

[0058] The processing in Figures 8 and 9 is executed as a subroutine of step St13 in Figure 4. In the explanation of Figures 8 and 9, one or more components and boards are represented as a three-dimensional model (i.e., three orthogonal axes: x-axis, y-axis, and z-axis), but they may also be explained as being represented as a two-dimensional model (i.e., two orthogonal axes: x-axis and y-axis).

[0059] 8, the prediction tool sets importance data for one or more components PRs and the board SUB1 specified in the first specified region PRT1 by referring to the memory 12 (step St21). This importance data is used as a weighting coefficient when making predictions using the AI ​​surrogate model MDL1. For example, the importance is "3" for a heat-generating element, "2" for a metal with high thermal conductivity that conducts heat to the board SUB1, and "1" for other components. This importance data is linked to each component in advance and stored in the memory 12.

[0060] The prediction tool sets the specified number of meshes (target number of meshes), limit width value, and number of times to execute mesh division (e.g., 20 times) specified in the AI ​​surrogate model for each of the x-axis, y-axis, and z-axis directions (step St22). These values ​​to be set may be stored in memory 12 in advance, or may be read and acquired from external database 20 by the prediction tool.

[0061] The prediction tool performs a process of drawing (arranging) mesh division lines along the axial planes (see above) of the x-axis, y-axis, and z-axis for at least the heating elements and metals arranged on the object (e.g., substrate SUB1) (step St23). Based on the arrangement of the mesh division lines in step St23, the prediction tool performs a mesh division process for at least the heating elements and metals along the mesh division lines (step St24). The mesh division process here refers to a process of determining which part of each component falls within which mesh (in other words, where it is located) by three-dimensionally arranging and dividing the components, such as the heating elements and metals, by mesh division lines consisting of the first mesh division line and the second mesh division line (see FIGS. 7A and 7B).

[0062] The prediction tool compares the number of meshes for each of the x-axis, y-axis, and z-axis extracted as the processing result of step St24 with the target number of meshes set in step St22 (step St25). If the prediction tool determines that the number of meshes for each of the x-axis, y-axis, and z-axis extracted as the processing result of step St24 matches the target number of meshes set in step St22 (step St25, YES), it ends the processing of mesh division method 1 shown in FIG.

[0063] On the other hand, if the prediction tool determines that the number of meshes for each of the x-axis, y-axis, and z-axis extracted as the processing result of step St24 does not match the target number of meshes set in step St22 (step St25, NO), it determines whether the number of times the mesh division process of the current step St24 has been executed exceeds the number of times the mesh division process of the current step St24 has been executed (step St26).If it determines that the number of times the mesh division process of the current step St24 has been executed exceeds the number of times the mesh division process of the current step St22 has been executed (step St26, YES), the processing of the prediction tool proceeds to step St31 (see FIG. 9).

[0064] On the other hand, if the prediction tool determines that the number of executions of the mesh division process in the current step St24 does not exceed the number of executions of mesh division set in step St22 (NO in step St26), it adjusts the current limit width for the corresponding axis direction according to the following formula (step St27). Here, the corresponding axis direction is the z-axis direction, for example, if the mesh counts and target mesh counts for the x-axis and y-axis match, but the mesh count and target mesh count for the z-axis do not match. The adjustment formula in step St27 is, for example, (new) limit width = (actual (current) mesh count / target mesh count) × (current) limit width. Note that this formula is not limited to this example formula. However, as shown in FIG. 8, if the calculation result of step St27 shows that the (new) limit width is less than 1.0 and equal to or greater than 0.9, the prediction tool rounds it down to 0.90. If the value is greater than 1 and less than 1.1, the prediction tool rounds it up to 1.10, and then returns to step St23.

[0065] In FIG. 9, if the prediction tool determines that the current number of meshes on each of the x-axis, y-axis, and z-axis is greater than the target number of meshes set in step St22 (step St31, NO), it adjusts the limit width for each axis (e.g., the z-axis) on which the number of meshes is greater than the target number of meshes (step St32). The formula used in the adjustment in step St32 is, for example, the above-mentioned (new) limit width = (actual (current) number of meshes / target number of meshes) × (current) limit width. This allows the limit width to be increased from the current value on the corresponding axis so that the current number of meshes approaches the target number of meshes. The prediction tool executes the mesh division process (see step St24) using the limit width adjusted in step St32 (step St33). Thereafter, the series of processes from step St32 to step St34 is repeated until the current number of meshes on each of the x-axis, y-axis, and z-axis becomes equal to or less than the target number of meshes set in step St22 (step St34, NO). As a result, mesh division lines for several target meshes specified by the AI ​​surrogate model are applied to an object (for example, a board SUB1) on which one or more components to be predicted are placed.

[0066] On the other hand, if the prediction tool determines that the current number of meshes on each of the x-axis, y-axis, and z-axis is less than or equal to the target number of meshes set in step St22 (YES in step St31 or step St34), it calculates the difference in the number of meshes as the differential mesh number (A) (i.e., "target number of meshes" - "current number of meshes") for each axis (step St35). From the processing result of step St24, the prediction tool obtains the coordinate values ​​of all meshes on the x-axis, y-axis, and z-axis within a bounding box that covers all components placed on the target object (e.g., board SUB) (step St36).

[0067] The prediction tool acquires the coordinates of two adjacent mesh division lines on each of the x-axis, y-axis, and z-axis, and the value of the interval (mesh width) between them (step St37). The prediction tool extracts the meshes equivalent to the number of differential meshes (A) calculated in step St35 from the largest mesh width acquired in step St37, and adds and arranges new mesh division lines between (e.g., in the middle) the two mesh division lines that make up each mesh width (step St38). In other words, the series of processes from step St35 to step St38 represents the process of adding new mesh division lines to reach the target number of meshes specified by the AI ​​surrogate model. As a result, mesh division lines equivalent to the number of target meshes specified by the AI ​​surrogate model are applied to the object (e.g., board SUB1) on which one or more components to be predicted are arranged.

[0068] 4. Detailed operation procedure for precision-oriented mesh division method 2 Next, the detailed operational procedure of mesh division method 2 in step St14 in Fig. 4 will be described with reference to Fig. 10 and Fig. 11. Fig. 10 is a diagram schematically illustrating an example of a processing image of mesh division method 2 for a part to be predicted. Fig. 11 is a flowchart showing a detailed example of the procedure of mesh division method 2 in step St14 in Fig. 4 in chronological order. For ease of understanding, the explanation is given using a two-dimensional model of part Pa3 and board Pa4 (i.e., two orthogonal axes, the x-axis and the y-axis), but the explanation in Fig. 10 can also be applied to a three-dimensional model of part Pa3 and board Pa4 (i.e., three orthogonal axes, the x-axis, the y-axis, and the z-axis).

[0069] Figure 10 shows an example of the mesh division results for part Pa3 placed on board Pa4, according to mesh division method 2. The disadvantage of an AI surrogate model created using CNN or similar is that its input shape (the number of meshes) is fixed. Therefore, when a part on the target object (e.g., a board) is moved by a user operation, the moved part may not fit within the fixed mesh (see the upper part of Figure 10 (before remeshing)).

[0070] Therefore, as a prerequisite for mesh division method 2, the AI ​​surrogate model is created larger than the target object to be predicted. Specifically, the AI ​​surrogate model is prepared by training it with a space (e.g., additional space AD1 in FIG. 10) around the size (dimensions) of the target object to be predicted, with, for example, three to ten mesh division lines added. This not only enables high-speed mesh division processing, but is also expected to improve the prediction accuracy of the AI ​​surrogate model at the edge of the target object to be predicted. As shown in FIG. 10, the number of additional mesh division lines is set to Xa on the minimum side and Xb on the maximum side in the x-axis direction of substrate Pa4, and Ya on the minimum side and Yb on the maximum side in the y-axis direction of substrate Pa4. In the following description, Xa, Xb, Ya, and Yb are integers equal to or greater than 0.

[0071] Specifically, before starting the prediction process using the AI ​​surrogate model, as shown in the lower part of FIG. 10 (after remeshing), the prediction tool moves some of the mesh division lines from the additional space AD1 provided around the object to be predicted, and positions (remeshes) the outer shape (outer contour) of part Pa3 so that it exists (is located) on the mesh division lines. In the example shown in FIG. 10, one of the Xb mesh division lines, one of the Ya mesh division lines, and one of the Yb mesh division lines have been moved. This allows the prediction tool to complete the mesh division process (remeshes) for part Pa3 so that the mesh division lines touch the outer contour of part Pa3 (in other words, so that the entire part Pa3 fits within one of the meshes defined by the mesh division lines).

[0072] The processing in Fig. 11 is executed as a subroutine of step St14 in Fig. 4. For ease of understanding, Fig. 11 will be explained assuming that one or more components and boards are represented by a two-dimensional model (i.e., two orthogonal axes, the x-axis and the y-axis), but it may also be explained assuming that they are represented by a three-dimensional model (i.e., three orthogonal axes, the x-axis, the y-axis, and the z-axis).

[0073] In FIG. 11, the prediction tool acquires the number of meshes to be added (Xa and Xb lines on the minimum and maximum sides in the x-axis direction, and Ya and Yb lines on the minimum and maximum sides in the y-axis direction) from the AI ​​surrogate model (step St41). The prediction tool acquires the x-coordinate value (B) and the y-coordinate value (C) of the current mesh division line by referring to the memory 12 (step St42). The prediction tool detects that a mesh division button (not shown) has been pressed by a user (step St43). Note that the processing of step St43 may be omitted.

[0074] The prediction tool refers to the memory 12 to obtain the maximum and minimum x-coordinate values ​​(D) of the part after movement in step St11 of Fig. 4 (see, for example, part Pa3 shown in Fig. 10) (step St44). The prediction tool compares the x-coordinate value (B) of the current mesh division line with the maximum and minimum x-coordinate values ​​(D) of the part after movement (step St45). If the prediction tool determines that the x-coordinate value (B) of the current mesh division line differs from the maximum and minimum x-coordinate values ​​(D) of the part after movement, it checks the current number of meshes to be added (Xa on the minimum side and Xb on the maximum side in the x-axis direction) (step St46).

[0075] If the prediction tool determines that "Xa+Xb≧1" (i.e., at least one of Xa and Ya is 1), it moves one mesh division line from the minimum or maximum side in the x-axis direction and arranges the mesh division line so that it is located between the minimum and maximum sides in the x-axis direction of the part (step St47). By moving the mesh division line, the prediction tool updates the current number of meshes to be added (Xa on the minimum side and Xb on the maximum side in the x-axis direction) (step St47). The prediction tool also updates the x-coordinate value (B) of the current mesh division line (step St48). After this update, the processing of the prediction tool returns to step St45.

[0076] On the other hand, if the prediction tool determines that "Xa+Xb=0" (that is, both Xa and Xb are zero), it executes the processing of mesh division method 1 described with reference to FIGS. 8 and 9 (step St49).

[0077] Furthermore, if the prediction tool determines that the x-coordinate value (B) of the current mesh division line is equal to the maximum and minimum x-coordinate values ​​(D) of the moved component, it references the memory 12 to acquire the maximum and minimum y-coordinate values ​​(E) of the component after movement in step St11 of FIG. 4 (see, for example, component Pa3 shown in FIG. 10) (step St50). The prediction tool compares the y-coordinate value (C) of the current mesh division line with the maximum and minimum y-coordinate values ​​(E) of the moved component (step St51). If the prediction tool determines that the y-coordinate value (C) of the current mesh division line is different from the maximum and minimum y-coordinate values ​​(E) of the moved component, it checks the current number of meshes to be added (Ya on the minimum side and Yb on the maximum side in the y-axis direction) (step St52).

[0078] When the prediction tool determines that "Ya + Yb ≥ 1" (i.e., at least one of Xa and Ya is 1), it moves one mesh division line from the minimum or maximum side in the y-axis direction and arranges the mesh division line so that it is located between the minimum and maximum sides in the y-axis direction of the component (step St53). The prediction tool updates the current number of additional meshes (Ya on the minimum side and Yb on the maximum side in the y-axis direction) by moving the mesh division line (step St53). The prediction tool also updates the y-coordinate value (C) of the current mesh division line (step St54). After this update, the processing of the prediction tool returns to step St51. In other words, the prediction tool repeats the series of processes from step St51 to step St54 until it determines that the y-coordinate value (C) of the current mesh division line is equal to the maximum and minimum y-coordinate values ​​(E) of the component after the movement.

[0079] Furthermore, if the prediction tool determines that the y-coordinate value (C) of the current mesh division line is equal to the maximum and minimum y-coordinate values ​​(E) of the part after movement, it performs mesh division processing based on the position of the mesh division line at that time (i.e., the x-coordinate value and the y-coordinate value) (step St55).

[0080] On the other hand, if the prediction tool determines that "Ya+Yb=0" (that is, both Ya and Yb are zero), it executes the processing of mesh division method 1 described with reference to FIGS. 8 and 9 (step St49).

[0081] 5. Detailed operation procedure of real-time part shaping method Next, with reference to FIGS. 12 and 13, the detailed operational procedure of the part shaping method in step St15 of FIG. 4 will be described. FIG. 12 is a diagram schematically illustrating an example of a processing image of a third operational example of mesh division of a part to be predicted. FIG. 13 is a diagram schematically illustrating an example of a processing image of the third operational example of mesh division of a part to be predicted. For ease of understanding, the parts Pa3a, Pa3b, and board Pa4 are described as being represented by a two-dimensional model (i.e., two orthogonal axes, the x-axis and the y-axis). However, the description of FIG. 12 can also be applied to a three-dimensional model of the parts Pa3a, Pa3b, and board Pa4 (i.e., three orthogonal axes, the x-axis, the y-axis, and the z-axis). Note that FIG. 13 describes the parts Pa3a, Pa3b, and board Pa4 as being represented by a three-dimensional model.

[0082] In mesh division method 1 (see step St13 in FIG. 4 ) and mesh division method 2 (see step St14 in FIG. 4 ), in order to emphasize improving the accuracy of predictions by the AI ​​surrogate model, mesh division lines are placed on the heating element (e.g., part Pa3 shown in FIG. 10 ), and the shape and physical parameters of the heating element itself are not changed. However, when prioritizing real-time performance, which prioritizes prediction processing time over improving the accuracy of predictions by the AI ​​surrogate model, the prediction tool does not change the current positions of the mesh division lines (x-coordinate values ​​and y-coordinate values), but instead reshapes (in other words, corrects) the shape and physical parameters of the heating element itself to match the positions of the mesh division lines. Shape correction here refers to changing the size (dimensions) of the relevant part (e.g., the heating element).

[0083] As shown in the upper part of Figure 12, assume that the entire component Pa3a placed on the board Pa4 does not fit within the mesh defined by the mesh division line Mx3 in the x-axis direction and the mesh division line My3 in the y-axis direction. In other words, the component Pa3a is placed so that parts of it protrude from the mesh in both the x-axis and y-axis directions. Specifically, the component Pa3a is located within the two central meshes adjacent to each other in the x-axis direction, but is not positioned so as to fit within the meshes surrounding those two meshes, and parts of the component Pa3a protrude from the surrounding meshes.

[0084] Therefore, the prediction tool does not change the current positions of the mesh division line Mx3 in the x-axis direction and the mesh division line My3 in the y-axis direction, and instead reshapes the shape of component Pa3a so that the outer periphery of component Pa3a touches the mesh division lines Mx3 and My3. For example, the prediction tool compares the area (or volume) of each mesh with that of a component, such as a heating element, for each mesh, starting with the smallest x-coordinate value, y-coordinate value, and z-coordinate value. If the prediction tool determines that "mesh area (or volume) × 0.5 ≧ component area (or volume)" holds true in this comparison, it determines that a circuit board, not a component, is placed in that mesh. If the prediction tool determines that "mesh area (or volume) × 0.5 < component area (or volume)" holds true, it determines that a component is placed in that mesh.

[0085] For example, the lower part of Figure 12 shows part Pa3b after shaping. As a result, the outer contour of part Pa3b is contained within a total of six meshes defined by mesh division lines Mx3 and My3. The prediction tool also corrects the physical property parameters of part Pa3b after shaping. For example, assume that the heat generation amount Q of part Pa3a is known in advance, and that of the total of six meshes containing part Pa3b, x1 and x2 are the mesh sizes (mesh widths) in the x-axis direction, and y1, y2, and y3 are the mesh sizes (mesh widths) in the y-axis direction.

[0086] The prediction tool corrects (in other words, allocates) the heat generation amount Q of the part Pa3a according to the following formula when the part Pa3a is shaped into the part Pa3b. The correction of the heat generation amount Q is, for example, (Q / (x1+x2)×(y1+y2+y3))×x1×y1. Note that the formula used for the correction is not limited to this formula.

[0087] As a premise for the explanation with reference to Figure 13, it is assumed that in the temperature environment of the atmosphere Ta, before and after changing the size of the component Pa3a such as a heating element, the temperature (T1, T2, T3, Tc) and the amount of heat transferred (Q1, Q2, Q3, ΣQ) do not change, but the emissivity (α1, α2, α3) and thermal conductivity (λ1, λ2, λ3) do change.

[0088] Temperatures T1, T2, T3, and Tc indicate the temperatures of the bottom, side, top, and center of component Pa3a, respectively. The heat transfer quantities Q1, Q2, and Q3, which correspond to the amount of heat generated, indicate the heat transfer quantity from the bottom of component Pa3a to the board Pa4, the heat transfer quantity from the side to the atmosphere, and the heat transfer quantity from the top to the atmosphere, respectively. The heat transfer quantity ΣQ corresponds to (Q1 + Q2 + Q3). Emissivities α1, α2, and α3 indicate the emissivities of the bottom, side, and top of component Pa3a, respectively. Emissivities α1', α2', and α3' indicate the emissivities of the bottom, side, and top of component Pa3b, respectively. Thermal conductivities λ1, λ2, and λ3 indicate the thermal conductivity in one direction along the surface, the other direction along the surface, and the height direction of component Pa3a, respectively. Thermal conductivities λ1', λ2', and λ3' indicate the thermal conductivities in one direction along the surface of part Pa3b, another direction perpendicular to the one direction along the surface, and the height direction, respectively. Surface areas S1, S2, and S3 indicate the areas of the bottom, side, and top of part Pa3a, respectively. Surface areas S1', S2', and S3' indicate the areas of the bottom, side, and top of part Pa3b, respectively. Length L1 indicates the length from the center to the side of part Pa3a. Length L1' indicates the length from the center to the side of part Pa3b.

[0089] For example, let us consider a calculation example for correcting the emissivity of the side surface of part Pa3a. Specifically, as described above, it is assumed that the amount of heat transferred Q2 remains constant even when the size of part Pa3a is changed from Pa3a to Pa3b. Therefore, Q2 = S2 × α2(T2 - Ta) = S2' × α2'(T2 - Ta). In other words, α2 × S2 = α2' × S2'. Therefore, the prediction tool calculates the emissivity α2' as the corrected emissivity of the side surface of part Pa3b as "α2' = (S2 / S2') × α2." Similarly, the prediction tool calculates the emissivities α1' and α3' as the corrected emissivity of the bottom and top surfaces of part Pa3b as "α1' = (S1 / S1') × α1" and "α3' = (S3 / S3') × α3," respectively.

[0090] For example, we will explain a calculation example for correcting the thermal conductivity of part Pa3a in one direction (λ1 direction) along its surface. Specifically, as described above, we assume that the amount of heat transferred, Q2, remains constant even when the size of part Pa3a is changed from Pa3a to Pa3b. Therefore, Q2 = λ1 × S2 × (Tc - T2) / L1 = λ1' × S2' × (Tc - T2) / L1' holds. In other words, λ1 × S2 / L1 = λ1' × α2' / L1'. Therefore, the prediction tool calculates the thermal conductivity λ1' as the corrected thermal conductivity in one direction (λ1 direction) along the surface of part Pa3b as follows: λ1' = λ1 × (S2 × L1') / (S2' × L1). Similarly, the prediction tool calculates the thermal conductivities λ2' and λ3' as the corrected thermal conductivities in the other direction (λ1 direction) perpendicular to the one direction along the surface of part Pa3b and in the height direction.

[0091] Summary of the Disclosure The above description of the embodiments discloses technical concepts corresponding to the following items.

[0092] (Item 1) The prediction method disclosed herein is a method for predicting the temperature distribution of an object on which one or more parts are arranged using an AI model, which acquires physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh with a predetermined number of meshes pre-specified in the AI ​​model, and when an instruction is given to move the part to another position on the object, performs a division process to shape the mesh or the part in accordance with the other position, generates second mesh data in which the part and the object are arranged in the mesh after the division process, and inputs each of the physical property parameters and the second mesh data into the AI ​​model, thereby predicting and outputting the temperature distribution. As a result, the prediction method makes it possible to position components so that they fit into the number of meshes specified by the AI ​​surrogate model according to their position after movement, thereby enabling the AI ​​model to be used to efficiently predict the temperature distribution of objects such as circuit boards.

[0093] (Item 2) In the prediction method described in Item 1, the division process includes a process of dividing the part into meshes by adding mesh division lines along axial planes of the x, y, and z axes of the object using a finite limit width, and a process of generating the second mesh data including the post-division mesh number when the post-division mesh number indicating the mesh number of the x, y, and z axes after the mesh division matches the specified mesh number. As a result, the prediction method can arrange the number of meshes specified by the AI ​​surrogate model to match the outer periphery (e.g., bounding box) of the moved part (e.g., heating element), thereby obtaining highly accurate prediction results.

[0094] (Item 3) In the prediction method according to item 1 or 2, the division process further includes a process of adjusting the limit width based on the post-division mesh number and the specified mesh number, and a process of repeating the mesh division using the adjusted limit width until the post-division mesh number matches the specified mesh number. As a result, according to the prediction method, even if the target number of meshes is not reached as a result of performing mesh division processing using a default value for the limit width (e.g., the initial value), the limit width itself can be repeatedly adjusted to match the target number of meshes, making it possible to arrange the number of meshes specified by the AI ​​surrogate model.

[0095] (Item 4) In the prediction method according to item 2 or 3, the parts include at least a first part that affects the temperature distribution, and the mesh division is performed with the first part as the target. As a result, this prediction method can target heat-generating elements that are likely to affect temperature distribution due to heat generation among the components placed on an object such as a circuit board, and can arrange meshes with the number of meshes specified by the AI ​​surrogate model, thereby obtaining highly accurate prediction results.

[0096] (Item 5) In the prediction method described in Item 1, the AI ​​model is trained based on the size of a mesh addition area in which a plurality of mesh division lines are added around the object, and the division process includes a process of acquiring the number of additional mesh lines indicating the number of mesh division lines in the mesh addition area and the coordinate values ​​of each mesh division line included in the first mesh data, and a process of generating the second mesh data including the coordinate values ​​of each mesh division line when the coordinate value of the part moved to the other position matches the coordinate value of any of the mesh division lines. As a result, this prediction method not only improves the accuracy of the prediction results using the AI ​​surrogate model for the end side of the part (e.g., heating element) after movement, but also allows for the placement of meshes with the number of meshes specified by the AI ​​surrogate model, thereby obtaining highly accurate prediction results.

[0097] (Item 6) In the prediction method described in Item 5, the division process further includes a process of moving any of the mesh division lines in the mesh addition region so that the mesh division lines match the coordinate values ​​of the part when the coordinate values ​​of the part moved to the other position do not match the coordinate values ​​of any of the mesh division lines. As a result, according to the prediction method, it is possible to move the mesh division lines from the spatial space via an AI surrogate model obtained in advance by learning using the spatial space around the object, and then arrange the mesh to match the moved parts.

[0098] (Item 7) In the prediction method described in Item 1, the division process includes a process of determining whether the part exists in a mesh defined by each of the mesh division lines based on a coordinate value of the part moved to the other position and a coordinate value of each mesh division line included in the first mesh data; a process of shaping the part based on the determination result so that the part exists in the mesh defined by each of the mesh division lines; and a process of correcting the physical property parameters of the part based on the coordinate values ​​of the part after shaping. As a result, according to this prediction method, it is possible to obtain prediction results such as the temperature distribution of an object in a short time and in real time by shaping the part to fit the mesh division lines without moving or changing the mesh division lines themselves.

[0099] (Item 8) In the layout generation method described in item 1, a designation operation screen for the component and the object is displayed on a display, and an instruction to move the component to the other position is input based on a user operation on the designation operation screen. As a result, according to the prediction method, it is possible to give an instruction to move a part through a simple and intuitive user operation.

[0100] (Item 9) A prediction device according to the present disclosure is a prediction device that predicts the temperature distribution of an object on which one or more parts are arranged using an AI model, and includes a processor and a memory. The processor, in cooperation with the memory, acquires physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh with a predetermined number of meshes pre-specified in the AI ​​model. When an instruction is given to move the part to another position on the object, the processor performs a division process to shape the mesh or the part in accordance with the other position, generates second mesh data in which the part and the object are arranged in the mesh after the division process, and inputs each of the physical property parameters and the second mesh data into the AI ​​model, thereby outputting a prediction result of the temperature distribution. This allows the prediction device to position the components so that they fit into the number of meshes specified by the AI ​​surrogate model according to their position after movement, thereby enabling the AI ​​model to be used to efficiently predict the temperature distribution of objects such as circuit boards.

[0101] (Item 10) The program disclosed herein causes a prediction device, which is a computer, to execute the following steps: acquire physical property parameters of each of the parts and the object and first mesh data in which the parts and the object are arranged in a mesh with a predetermined number of meshes pre-specified in the AI ​​model, when an instruction is given to move the part to another position on the object, perform a division process to shape the mesh or the part in accordance with the other position, and generate second mesh data in which the part and the object are arranged in the mesh after the division process; and output a prediction result of the temperature distribution by inputting each of the physical property parameters and the second mesh data into the AI ​​model. This allows the prediction device to position the components so that they fit into the number of meshes specified by the AI ​​surrogate model according to their position after movement, thereby enabling the AI ​​model to be used to efficiently predict the temperature distribution of objects such as circuit boards.

[0102] Although the embodiments have been described above with reference to the accompanying drawings, the present disclosure is not limited to such examples. It is clear that a person skilled in the art can conceive of various modifications, alterations, substitutions, additions, deletions, and equivalents within the scope of the claims, and it is understood that these also fall within the technical scope of the present disclosure. Furthermore, the components in the above-described embodiments may be combined in any manner without departing from the spirit of the invention. [Industrial Applicability]

[0103] The technology disclosed herein is useful as a prediction method, a prediction device, and a program for efficiently predicting the temperature distribution of an object such as a substrate. [Explanation of symbols]

[0104] 10 Prediction Device 11 Communication I / F 12 Memory 13 Input Devices 14 Display Devices 15 processors 16 AI processors 20 External Databases

Claims

1. A method for predicting a temperature distribution of an object having one or more components disposed thereon using an AI model, comprising: Acquire physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh of a predetermined number of meshes pre-specified in the AI ​​model; when there is an instruction to move the part to another position on the object, performing a division process to shape the mesh or the part in accordance with the other position, and generating second mesh data in which the part and the object are arranged in the mesh after the division process; The physical property parameters and the second mesh data are input into the AI ​​model to predict and output the temperature distribution. Forecasting methods.

2. The division process includes: a process of dividing the part into meshes by adding mesh division lines along the axial planes of the x, y, and z axes of the object using a finite limit width; and generating the second mesh data including the post-division mesh number when the post-division mesh number indicating the mesh number for each of the x, y, and z axes after the mesh division matches the specified mesh number. The prediction method of claim 1 .

3. The division process includes: a process of adjusting the limit width based on the number of meshes after division and the specified number of meshes; and repeating the mesh division using the adjusted limit width until the number of meshes after division matches the specified number of meshes. The prediction method according to claim 2 .

4. the components include at least a first component that affects the temperature distribution; The mesh division is performed on the first part. The prediction method according to claim 2 or 3.

5. The AI ​​model is trained based on the size of a mesh additional region in which a plurality of mesh division lines are added around the object, The division process includes: A process of acquiring a mesh addition number indicating the number of mesh division lines in the mesh addition region and coordinate values ​​of each mesh division line included in the first mesh data; and generating the second mesh data including the coordinate values ​​of each of the mesh division lines when the coordinate values ​​of the part moved to the other position match the coordinate values ​​of any of the mesh division lines. The prediction method of claim 1 .

6. The division process includes: and when the coordinate values ​​of the part moved to the other position do not match the coordinate values ​​of any of the mesh division lines, moving any of the mesh division lines in the mesh additional region so that the coordinate values ​​match the coordinate values ​​of the part. The prediction method according to claim 5 .

7. The division process includes: a process of determining whether or not the part exists in a mesh defined by each of the mesh dividing lines based on the coordinate values ​​of the part moved to the other position and the coordinate values ​​of each of the mesh dividing lines included in the first mesh data; a process of shaping the part based on the determination result so that the part exists within a mesh defined by each of the mesh division lines; and correcting the physical property parameters of the part based on coordinate values ​​of the part after shaping. The prediction method of claim 1 .

8. displaying a designation operation screen for the part and the object on a display; the instruction to move the component to the other position is input based on a user operation on the designation operation screen; The prediction method of claim 1 .

9. A prediction device that predicts a temperature distribution of an object having one or more components arranged thereon using an AI model, A processor and a memory, The processor, in cooperation with the memory, Acquire physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh of a predetermined number of meshes pre-specified in the AI ​​model; when there is an instruction to move the part to another position on the object, performing a division process to shape the mesh or the part in accordance with the other position, and generating second mesh data in which the part and the object are arranged in the mesh after the division process; The physical property parameters and the second mesh data are input into the AI ​​model, and a prediction result of the temperature distribution is output. Prediction device.

10. A prediction device, which is a computer, predicts the temperature distribution of an object having one or more components arranged thereon using an AI model. A process of acquiring physical property parameters of each of the parts and the object, and first mesh data in which the parts and the object are arranged in a mesh of a predetermined number of meshes pre-specified in the AI ​​model; a process of performing a division process to shape the mesh or the part in accordance with the new position when an instruction to move the part to another position on the object is given, and generating second mesh data in which the part and the object are arranged in the mesh after the division process; and a process of inputting the physical property parameters and the second mesh data into the AI ​​model to output a prediction result of the temperature distribution. program.

Images