Training data generation apparatus, training data generation method, and program

Abstract

To provide a device, a method and a program for easily acquiring a casting analysis result to be teacher data for constructing a surrogate model in a short time.SOLUTION: An acquisition unit configured to acquire a basic shape model having a product portion and a substantially columnar riser portion, in which a surface intersecting a reference plane perpendicular to a height direction of the riser portion is set as an upper surface of the riser portion; a setting unit configured to determine an upper limit value of a height from the upper surface and a step value equal to or smaller than the upper limit value; an arithmetic unit configured to obtain a dimension value obtained by multiplying an integer equal to or smaller than a quotient obtained by dividing the upper limit value by the step value and equal to or larger than 0 by the step value; a shape model expansion unit configured to create an expanded shape model by extending the upper surface of the riser portion by the dimension value; and an analysis model creation unit configured to divide the expanded shape model into elements; A training data creation device comprising: an analysis unit configured to execute a coagulation analysis under a predetermined analysis condition to acquire coagulation analysis data; and a training data creation unit configured to create, for an extended shape model, training data including the predetermined analysis condition, the coagulation analysis data, and shape data based on the extended shape model.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a teacher data creation device, a teacher data creation method, and a program for supporting the creation of teacher data, and more particularly to the creation of casting analysis data as teacher data for constructing a surrogate model. [Background technology]

[0002] Casting analysis using methods such as the finite element method and the finite difference method has long been used as a design tool for castings and casting plans, and the time required for analysis has been shortened as computer calculation speeds have improved. However, depending on the size and shape of the target casting, it can still take several hours to several tens of hours of calculation time to obtain an optimal casting plan, such as the size of the riser. This is because the creation of analytical models, including the element division process, is still done by the operator, and element division must be performed for each shape of the casting plan being analyzed.

[0003] In this regard, Patent Document 1 states, "In the design of a casting plan for weirs and feeders used in a casting process, a neural network is constructed in which data on the shapes of the product, weir, and feeder are input and solidification times are output, and for a plurality of products, the relationship between the product shape and the solidification time of the product is trained in the neural network for the product, the relationship between the weir shape and the solidification time of the weir is trained in the neural network for the weir, and the relationship between the feeder shape and the solidification time of the feeder is trained in the neural network for the feeder, and (1) the dimensions of the product to be manufactured are (2) assuming the dimensions of the weir and the feeder, and inputting these into a neural network related to the weir and a neural network related to the feeder, respectively, (3) calculating the solidification times of the product, the weir, and the feeder, respectively, (4) comparing these solidification times to determine whether the assumed dimensions of the weir and the feeder are appropriate, (5) if they are appropriate, fabricating the weir and the feeder based on those dimensions, and (6) if they are inappropriate, changing those dimensions and repeating steps (2) to (4).”

[0004] However, this technology requires three preparatory steps: constructing each of the mentioned neural networks in advance, investigating and obtaining the solidification times of the product, weir, and riser for various casting products, and training each neural network that has been constructed based on the relationship between these shapes and solidification times. Therefore, the accuracy of the analysis depends on the degree of learning of each neural network, that is, the number and accuracy of the data that is trained. In other words, while this technology can perform the analysis itself to find the optimal casting plan quickly and efficiently, it still requires a great deal of man-hours when including these prerequisite preparatory steps.

[0005] Furthermore, to improve the accuracy of numerical analysis using such physical models, it is necessary to accurately and comprehensively understand the physical phenomena of interest and reflect them in the analysis conditions. However, in casting analysis, which deals with the complex behavior of high-temperature molten metal, from its flow to solidification, it is difficult to confirm this behavior through experiments, and the analysis conditions become extremely complex and highly accurate, resulting in enormous calculation times. In other words, it has become extremely difficult to simultaneously improve accuracy and speed with conventional casting analysis.

[0006] Meanwhile, in recent years, with the dramatic improvement of AI technology making it easier to carry out machine learning such as neural networks, a method of calculating and predicting physical phenomena using a proxy model (hereinafter also referred to as a surrogate model) has become common. For example, Patent Document 2 describes a computer-implemented method for manufacturing or controlling a technical system, which includes: (a) inputting samples of parameter sets suitable for manufacturing or controlling the technical system together with a feasibility identifier assigned to each of the parameter sets (S1), wherein the feasibility identifier records each parameter set as technically feasible if it satisfies a predetermined system criterion for the manufacturing or control of the technical system, or as technically infeasible if it does not satisfy the predetermined system criterion, or as erroneous if an evaluation of the parameter set using a computer simulation results in an error; (b) recording the error (S2) generating a computerized surrogate model for the technical system by a regression method based on the parameter sets of each of the samples recorded as technically feasible, rather than infeasible or infeasible, (c) determining an optimized parameter set based on the surrogate model by a computer optimization method (S3), and (d) outputting the optimized parameter set (PS_opt) for manufacturing or controlling the technical system (S4)." The proposed method discloses a technology for obtaining an optimized parameter set by using a surrogate model trained with a parameter set obtained from an analysis of a physical model. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2000-326051 [Patent Document 2] Special Publication No. 2023-515640 Summary of the Invention [Problem to be solved by the invention]

[0008] Therefore, if a surrogate model is constructed by having AI learn from the results of casting analysis using a conventional physical model as training data, the casting analysis process can be omitted. Furthermore, since there is no need to perform complex element division and set analysis conditions from scratch, the time required to achieve the desired shape optimization, such as a feeder design, can be dramatically reduced. However, in order to construct a surrogate model that achieves sufficient prediction accuracy, a considerable number of analysis results for multiple different shapes must be prepared as training data. The present invention aims to provide an apparatus, method, and program that quickly and easily obtains casting analysis results that serve as training data for constructing a surrogate model. [Means for solving the problem]

[0009] The first aspect of the present invention is an acquisition unit that acquires a basic shape model having a product part and at least one substantially columnar riser that is integrally connected to the product part, a reference plane that is perpendicular to the height direction of the riser, and a plane where the riser and the reference plane intersect is the top surface of the riser; a setting unit that determines an upper limit value of the height of the riser, with an upward direction from the top surface being positive, and interval values ​​that are equal to or less than the upper limit value; a calculation unit that calculates a quotient by dividing the upper limit value by the interval value, and multiplies the interval value by an integer that is equal to or less than the quotient and is equal to or greater than 0; a shape model extension unit that creates an extended shape model having a deformed riser portion extended by the dimension value in the direction of the extension; an analytical model creation unit that divides the extended shape model into elements to create an analytical model; an analysis unit that performs solidification analysis on the analytical model under predetermined analytical conditions to obtain solidification analysis data for all of the elements that make up the analytical model; and a training data creation unit that creates training data for the extended shape model, the predetermined analytical conditions, the solidification analysis data, and shape data based on the extended shape model. and an output unit that outputs the teacher data.

[0010] A second aspect of the present invention is the solidification analysis data for all of the elements constituting the analytical model is obtained by performing a solidification analysis on the analytical model under predetermined analytical conditions. The method comprises the steps of: preparing a basic shape model having a product part and at least one substantially columnar feeder part integrally connected to the product part, a reference plane perpendicular to the height direction of the feeder part, and a plane where the feeder part and the reference plane intersect is defined as the top surface of the feeder part; determining an upper limit value for height, with an upward direction from the top surface being positive, and step values ​​that are equal to or less than the upper limit value; determining a quotient by dividing the upper limit value by the step value, and multiplying the step value by an integer that is equal to or less than the quotient

[0011] In the training data creation method according to the second embodiment of the present invention, the created extended shape models preferably include the basic shape model. That is, it is preferable that at least one of the extended shape models is equivalent to the basic shape model. In other words, it is preferable that for all the feeders, at least one extended shape model is included, having a deformed feeder section whose dimension value is 0 (zero), i.e., 0 (zero), which is the product of the integer 0 (zero) and the step value.

[0012] A third aspect of the present invention is a step of acquiring a basic shape model having a product part and at least one substantially columnar feeder part integrally connected to the product part, a reference plane perpendicular to the height direction of the feeder part, the plane where the feeder part and the reference plane intersect being the top surface of the feeder part; a step of acquiring an upper limit value of the height, where an upward direction from the top surface is positive, and an increment value that is less than or equal to the upper limit value; a step of obtaining a quotient by dividing the upper limit value by the increment value and multiplying the increment value by an integer less than or equal to the quotient; a step of creating an extended shape model having a deformed feeder part where the top surface of the feeder part is extended upward by the increment value; a step of creating an analytical model by dividing the extended shape model into elements; a step of performing solidification analysis on the analytical model under predetermined analytical conditions to obtain solidification analysis data for all of the elements constituting the analytical model; and a step of creating teacher data for the extended shape model, the predetermined analytical conditions, the solidification analysis data, and the shape data based on the extended shape model.

[0013] In each of the first to third embodiments of the present invention, The solidification analysis data preferably includes the time required for the elements constituting the hot spot in the product part to form the hot spot, the solidification completion time of each of the elements constituting the product part, the temperature of each of the elements constituting the product part at the maximum value of the solidification completion time, and the porosity value of each of the elements constituting the product part at the maximum value of the solidification completion time. [Effects of the Invention]

[0014] The present invention provides an apparatus, method, and program that can quickly and easily obtain the required number of casting analysis results as training data for constructing a surrogate model, and by using these, even those without specialized knowledge of casting can easily create the training data. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is a schematic diagram illustrating an example of a basic shape model used in the present invention. [Figure 2] 1 is a schematic diagram showing an example of the configuration of a system including an apparatus according to the present invention. [Figure 3] 1 is a schematic diagram showing an example of a hardware configuration of an apparatus according to the present invention. [Figure 4] FIG. 1 is a schematic diagram showing an example of the functional configuration of an apparatus according to the present invention. [Figure 5] 2 is a schematic diagram illustrating a cross section A in FIG. 1, for explaining an example of a method for creating an extended geometric model according to the present invention. FIG. [Figure 6] 1 is a flow chart showing the steps of an example of a method according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] First, the basic shape model used in the present invention will be described. FIG. 1 is a schematic diagram illustrating the basic shape model. The basic shape model 3 shown in FIG. 1 is an example of an exhaust manifold, which is a cast part, but is not limited to this example. The basic shape model 3 has a configuration in which a substantially columnar feeder 31 is integrally connected to a product part 30, which will be the cast part, to prevent shrinkage cavities from occurring during the casting process. At least one feeder 31 is required, and in the example shown in FIG. 1, five feeders 311-315 are integrally connected to the product part 30 at locations of the product part 30 where shrinkage cavities are likely to occur. Note that when the individual feeders 311-315 are not distinguished and are generally referred to, hereinafter, as feeder 31.

[0017] 1 is in a form in which its shape is visualized, the basic shape model 3 will be referred to hereinafter as the basic shape model 3, including a data format in which the shape is not visualized. The non-visualized data format of the extended shape model 4 described below will also be referred to as the extended shape model 4. When there is no need to distinguish between the basic shape model 3 and the extended shape model 4, the basic shape model 3 and the extended shape model 4 will be simply referred to as the shape model hereinafter. The shape model in the present invention may be divisible into elements, i.e., may have a data format from which a mesh model can be created, and may be a surface model such as STL having face information, or a solid model further having volume information.

[0018] In Fig. 1, the height direction z indicated by the chain line with an arrow indicates the height direction of the feeder head 31. The height direction z is equal to the vertical direction in actual casting, and the direction of the arrow corresponds to the vertically upward direction in actual casting. The feeder heads 311 to 315 are provided with a reference plane S perpendicular to the height direction z. B 1~S B 5 are defined for each. Below, the individual reference planes S B 1~S B When speaking generally without distinguishing between 5, the reference plane S B Reference plane S B The height value of the reference plane S is defined as 0 (zero) at the corresponding riser 31. B The position of the reference plane S can be arbitrarily determined in the corresponding riser portion 31. B 1~S B The heights of all 5s are defined as 0 (zero), but they may not be located on the same plane as each other.

[0019] In the basic shape model 3, the direction of the arrow in the height direction z is set to be positive, and the top surfaces of the riser sections 311 to 315 are set to the respective reference planes S B 1~S B 5. In other words, the upper surface of the riser 31 is aligned with the reference plane S B It is defined as the plane where and intersect.

[0020] In particular, the positions of the upper surfaces of all the risers 31, that is, all the reference planes S BIn the case of the basic shape model 3 having a shape in which the position of the feeder head is as close as possible to a surface (not shown) corresponding to the mold parting surface (parting surface) in the actual casting of the product portion 30, it is a shape model having a shape in which the feeder head effect is substantially absent. When a basic shape model 3 having such a shape is prepared and added as one shape to the extended shape model 4 described later, it is preferable because training data for a shape in which the feeder head effect is substantially absent can be obtained.

[0021] Next, referring to Figure 2, we will explain an example of a teacher data creation system 200 (hereinafter also simply referred to as a creation system) that includes a teacher data creation device 1 (hereinafter also simply referred to as a creation device) related to the first embodiment of the present invention.

[0022] The teacher data creation system 200 includes a teacher data creation device 1. The creation device 1 may be connected to a terminal 220 different from the creation device 1 via a communication line 210. The creation device 1 may also be connected to a server 230 or the like via the communication line 210. The terminal 220 may be a device for creating basic shape models 3 (see FIG. 1; the same applies below). The basic shape models 3 created by the terminal 220 may be transmitted to the creation device 1 via the communication line 210, or may be transmitted to the server 230 once, and a predetermined number of basic shape models 3 may be stored in the server 230. The creation device 1 acquires the basic shape models 3 from the terminal 220 or the server 230 via the communication line 210. Note that while the creation system 200 in FIG. 2 is configured such that the creation device 1 acquires the basic shape models 3 via the communication line 210, the teacher data creation device 1 itself may have the function of creating the basic shape models 3. In this case, the teacher data creation device 1 can directly acquire the basic shape models 3 without using the communication line 210.

[0023] [1] First embodiment Next, an example of the creation device 1 according to the first embodiment of the present invention will be described. Fig. 3 is a schematic diagram showing an example of the hardware configuration of the creation device 1. Fig. 4 is a schematic diagram showing an example of the functional configuration of the creation device 1.

[0024] See Figure 3. The creation device 1 is composed of a main body 10, which is its main part, and an input device 12 and a display device 13 that are directly connected to the main body 10. The evaluation device 1 is specifically a computer, and in addition to electronic devices such as a workstation (WS) or a personal computer (PC), electronic devices such as a smartphone, a tablet terminal, a wearable terminal, or an IoT (Internet of Things) device, or a single-board computer such as Raspberry Pi (registered trademark) may also be used, and these terminals may have built-in microphones, cameras, etc.

[0025] The main body 10 of the creation device 1 includes a CPU (Central Processing Unit) 101, a RAM (Random Access Memory) 102, a ROM (Read Only Memory) 103, an STR (Storage) 104, which is a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and an I / F (Interface) 105. It may also include a GPU (Graphics Processing Unit, a processor dedicated to image processing) 106. Each of these components 101 to 106 is connected to one another by an internal bus 107. The main body 10 may further include a housing 100 that houses these components.

[0026] The CPU 101 controls the overall operation of the creation device 1 via an internal bus 107. The RAM 102 is a work area used when the CPU 101 is operating. The ROM 103 stores the operation code of the CPU 101. When a GPU 106 is provided, data converted and processed by the CPU 101 is processed in parallel to perform calculations at high speed. The STR 104 is an auxiliary storage area for recording and reading information controlled or processed by the CPU 101. The I / F 105 is an interface for sending and receiving various information to and from external devices such as the input device 12 and display device 13 connected to the main body 10.

[0027] Of these I / Fs 105, the I / F 108 transmits and receives various information to and from a terminal 220, a server 230, etc. via a communication line 210 in the creation system 200 shown in FIG.

[0028] The I / F 109 transmits and receives information to and from the input device 12. A plurality of input devices 12 may be provided, and may include, for example, a command or numerical value input means such as a keyboard, an image acquisition means such as an image scanner, a pointing device such as a mouse, etc. An operator or the like using the creation device 1 inputs commands and various information for controlling the creation device 1 via these input devices 12.

[0029] Furthermore, the I / F 110 transmits and receives various types of information to and from the display device 13. The display device 13 outputs various types of information stored in the storage device STR 104, the processing status of the creation device 1, and the like. The display device 13 may be, for example, a display. The display may also function as the input device 12, such as a touch panel type. Furthermore, a plurality of display devices 13 may be provided.

[0030] Various information including the basic shape model 3 to be input to the creation device 1 may be input directly by an operator or the like via the input device 12, or may be input via the I / F 108 as a signal transmitted from a terminal 220, a server 230, or the like via a communication line 210 shown in Fig. 1. This various information may also include information for operating or controlling the creation device 1.

[0031] Next, the function of the creation device 1 will be described with reference to FIG. (1) Acquisition section 151 The acquisition unit 151 acquires the basic shape model 3. It is preferable that the acquisition unit 151 has a function to acquire not only the basic shape model 3 but also all information (hereinafter referred to as adjustment information) required by the operator to operate the creation device 1 under desired conditions. The adjustment information includes, for example, mesh data obtained by dividing the basic shape model 3 into elements, upper limit values ​​and interval values ​​set in the setting unit 152 described later, information required to determine the upper limit values ​​and interval values, and information for selecting the extended shape model 4 to be created by the shape model extension unit 154 described later. If the configuration is such that such adjustment information can also be acquired, it becomes possible to obtain a result reflecting the desired adjustment by the adjustment unit 159 described later.

[0032] (2) Setting section 152 Further, the upper limit value H set by the setting unit 152 will be described with reference to Fig. 5. Fig. 5 is a cross section A of the feeder head 312 shown in Fig. 1 and the product section 30 integrally connected thereto, and is a schematic diagram illustrating the upper limit value H according to the present invention and the interval value h described later. The setting unit 152 sets the reference plane S of the feeder head 31 of the basic shape model 3 acquired by the acquisition unit 151. B The upper limit value H of the height is determined with the upward direction (the direction of the arrow in the height direction z) being positive. The upper limit value H may be set to any value exceeding 0 (zero). In FIG. 5, the reference plane S of the riser portion 312 is B This indicates that an upper limit value H2 of the height is set, with the height of the feeder head 311 being 0 (zero). Although not shown, upper limit values ​​H1 and H3 to H5 can be similarly set for the feeder head 311 and the feeder heads 313 to 315 shown in Figure 1. These upper limit values ​​H1 to H5 may be different values ​​for each corresponding feeder head 311 to 315. When referring generally to the upper limit values ​​H1 to H5 without distinguishing between them, they will hereinafter be referred to as upper limit value H.

[0033] Next, the step value h set by the setting unit 152 will be described with reference to FIG. 5. The setting unit 152 determines the step value h to be a positive value that is equal to or less than the upper limit value H. The step value h may be any value as long as it is a positive value that is equal to or less than the upper limit value H. In FIG. 5, the step value h2 is a value smaller than the upper limit value H2 and is set to a value that is equal to or less than the reference plane S B2, i.e., a positive value greater than a height of 0 (zero). Although not shown, the increment values ​​h1 and h3 to h5 can be similarly determined for the feeder head portion 311 and feeder head portions 313 to 315 shown in Figure 1. These increment values ​​h1 to h5 may be different values ​​for each corresponding feeder head portion 311 to 315. When the increment values ​​h1 to h5 are to be generally described without distinguishing between them, they will hereinafter be referred to as increment value h.

[0034] (3) Arithmetic unit 153 The calculation unit 153 determines the quotient by dividing the upper limit value H by the step value h. Figure 5 shows an example in which the upper limit value H2 for the feeder head 312 is exactly three times the step value h2. In this example, the quotient by dividing the upper limit value H2 by the step value h2 is exactly 3. For the feeder head 311 and the feeder heads 313-315 shown in Figure 1, the quotient can also be determined by dividing the upper limit value H1 and the upper limit values ​​H3-H5 (none of which are shown) by the step value h1 and the step value h3-h5 (none of which are shown). Here, if the upper limit value H is not an integer multiple of the step value h, the upper limit value H will not be divisible by the step value h and a remainder will be generated, but in this case the quotient will be the integer part.

[0035] (4) Shape model extension part 154 The shape model extension unit 154 multiplies the integer equal to or less than the quotient obtained by the calculation unit 153 and equal to or greater than 0 (zero) by the step value h, and extends the top surface of the riser portion 31 of the basic shape model 3, i.e., the reference surface S B is extended in the direction of the arrow in the height direction z to create a deformed feeder 41, and an extended geometric model 4 having this deformed feeder 41 is created. For example, in the feeder 312 shown in FIG. 5, the reference plane S BThe shape model extension unit 154 can create at least one of four types of modified feeder sections 412: modified feeder section 4121 which is elongated by the value obtained by multiplying step value h2 by 1, modified feeder section 4122 which is elongated by the value obtained by multiplying step value h2 by 2, modified feeder section 4123 which is elongated by the value obtained by multiplying step value h2 by 3, and modified feeder section 4120 (equivalent to feeder section 312) which is elongated by 0 (zero), which is the value obtained by multiplying step value h by 0 (zero), i.e., no extension of feeder section 312. The extension obtained by multiplying step value h by 0 (zero), i.e., no extension, is the same as feeder section 312, but when processed by the shape model extension unit 154, it is treated as modified feeder section 4120, which is one of the modified feeder sections 412.

[0036] Similarly, for feeder 311 and feeder sections 313-315 in Figure 1, it is possible to create as many modified feeder sections 41 as there are integers equal to or less than the quotient calculated by calculation section 153 and equal to or greater than zero. For example, as with feeder section 312, if the quotient obtained by dividing the upper limit value H by the step value h is 3 for all of feeder sections 311 and 313-315, it is possible to create four forms of modified feeder section 41 for each. That is, although none of these are shown in the figures, it is possible to create four forms of modified feeder sections 4110-4113, as well as four forms of modified feeder sections 4130-4133, four forms of modified feeder sections 4140-4143 and four forms of modified feeder sections 4150-4153. In addition, when referring generally to the individual modified feeder sections 411 to 415, modified feeder sections 4110 to 4113 which are further variations of the individual modified feeder sections 411 to 415, modified feeder sections 4120 to 4123, and so on, they will hereinafter be referred to as modified feeder section 41.

[0037] The geometric model extension unit 154 can create an extended geometric model 4 having one form of modified feeder 41 for each of the feeders 311 to 315. That is, since the maximum number of combinations in which one form can be selected from each of the modified feeders 411 to 415 is 1024 (4 to the fifth power), a maximum of 1024 different extended geometric models 4 can be created.

[0038] The maximum number of different extended shape models 4 that can be created by the shape model extension unit 154 increases exponentially with the increase in the number of feeders 31 constituting the basic shape model 3 and the increase in the number of integers equal to or greater than 0 (zero) and equal to or less than the quotient of the upper limit value H divided by the step value h, i.e., the increase in the number of modified feeders 41. The more extended shape models 4 are created, the more training data is obtained for each extended shape model 4. However, it is preferable that the shape model extension unit 154 is configured to obtain only a necessary and sufficient number of training data for constructing the surrogate model of the present invention. That is, it is preferable that the shape model extension unit 154 is configured to arbitrarily select a multiple of the step value h for each feeder 31, i.e., to adjust the number of modified feeders 41 that can be created. Adjustment information for this purpose may be acquired from the acquisition unit 151. With this configuration, the adjustment unit 159 (described later) controls the operation of the shape model extension unit 154 to adjust the modified feeders 41 to be created, thereby creating different extended shape models 4 with the necessary and sufficient shapes and number of modified feeders 41.

[0039] In particular, the reference plane S B It is preferable to provide adjustment information to the creation device 1 so that the extended shape model 4 having a deformed feeder section 41, which is created by extending the basic shape model 3 by the product of the step value h multiplied by 0 (zero), i.e., the basic shape model 3 itself, is included in one of multiple different extended shape models 4 created in the shape model extension section 154, because this makes it possible to create at least one piece of training data based on the basic shape model 3 that does not substantially have the effect of a feeder section.

[0040] (5) Analysis model creation unit 155 The analytical model creation unit 155 creates an analytical model (not shown) by dividing the extended geometric model 4 created by the geometric model extension unit 154 into elements. The analytical model is so-called mesh data in which the extended geometric model 4 is divided into a plurality of minute elements each consisting of a polyhedron. Adjustment information such as the conditions for dividing into elements is acquired from the acquisition unit 151, an adjustment unit 159 (described later), or a storage unit 160.

[0041] (6) Analysis section 156 The analysis unit 156 performs solidification analysis on the analytical model created by the analytical model creation unit 155 under predetermined analytical conditions to obtain solidification analysis data for all elements constituting the analytical model. Here, solidification analysis may be a known solidification analysis using the finite element method or the like, or may also include molten metal flow analysis. The predetermined analytical conditions include physical properties of the metallic material assumed in the analytical model (e.g., solidification start temperature, solidification end temperature, solidification latent heat, solidification shrinkage rate, thermal conductivity, specific heat, flow limit solid fraction, etc.), the thermal conductivity of the assumed mold (not shown), the heat transfer coefficient at the interface between the mold and the analytical model, boundary conditions such as the inflow rate and temperature change of the molten metallic material, the initial temperature of the analytical model, and the assumed initial temperature of the mold. The analysis unit 156 may also include a physical model that takes into account gas-liquid two-phase flow, thermal stress, etc. These predetermined analytical conditions can be acquired as adjustment information from the acquisition unit 151, the adjustment unit 159 (described later), or the storage unit 160.

[0042] In addition, solidification analysis data is data obtained by solidification analysis, and includes, for example, the solidification completion time (the time until the solid fraction reaches 100% or the time until the flow limit solid fraction is reached) for all elements that make up the product portion 30, the temperature of all elements that make up the product portion 30 at the solidification completion time of the element that has the longest solidification completion time among the solidification completion times of all elements that make up the product portion 30, the porosity values ​​of all elements that make up the product portion 30, elements that can be formed as hot spots, and the time required for a hot spot to be formed (hereinafter also referred to as hot spot time), but is not limited to these data.

[0043] (7) Teacher Data Creation Unit 157 The teacher data creation unit 157 creates teacher data for each extended shape model 4, which includes the specified analysis conditions applied by the analysis unit 156, the coagulation analysis data acquired by the analysis unit 156, and shape data based on the extended shape model 4, i.e., the extended shape model 4 or its analysis model (not shown).

[0044] (8) Output unit 158 The output unit 158 ​​outputs the teacher data created by the teacher data creation unit 157 to the outside of the creation device 1. In addition to this, the output unit 158 ​​may have a function to output the basic shape model 3 or adjustment information acquired by the acquisition unit 151, the upper limit value H and step value h set by the setting unit 152, the quotient obtained by dividing the upper limit value H by the step value h calculated by the calculation unit 153, the extended shape model 4 created by the shape model extension unit 154, the analysis model created by the analysis model creation unit 155, and the like. The output unit 158 ​​may also have a function to output the processing status of the creation device 1 as appropriate.

[0045] (9) Adjustment section 159 The creation device 1 may also include an adjustment unit 159. The adjustment unit 159 has a function of adjusting the operations of the setting unit 152, the calculation unit 153, the shape model extension unit 154, the analysis model creation unit 155, the analysis unit 156, the teacher data creation unit 157, the output unit 158, and the storage unit 160, which will be described later, based on the adjustment information acquired by the acquisition unit 151.

[0046] (10) Storage section 160 The memory unit 160 has the function of storing and reading out various information processed by the acquisition unit 151, setting unit 152, calculation unit 153, shape model extension unit 154, analysis model creation unit 155, analysis unit 156, teacher data creation unit 157, output unit 158, and adjustment unit 159 in RAM 102 or STR 104 (see Figure 3).

[0047] [2] Second embodiment Next, a method for creating teacher data according to a second embodiment of the present invention will be described. Fig. 6 is a flowchart showing the steps of one example of a method according to the present invention. That is, the second embodiment of the present invention is a method for creating teacher data (s7) including the steps of preparing a basic shape model (s1), creating upper limits and intervals (s2), determining dimensional values ​​(s3), creating an extended shape model (s4), creating an analytical model (s5), and acquiring solidification analysis data (s6). Each of these steps will be described in detail below.

[0048] (1) Step (s1) of preparing a basic shape model First, a basic shape model 3 is prepared (s1, hereinafter sometimes referred to as s1 step. Each step described below may be described in the same manner). As shown in Fig. 1, the basic shape model 3 has a structure in which at least one riser portion 31 is integrally connected to a product portion 30. The basic shape model can be newly created using 3D CAD software or the like. Alternatively, an existing shape model can be used and the reference surface S B Alternatively, a basic shape model 3 may be prepared by modifying the shape of the riser portion 31 having the above-mentioned structure.

[0049] (2) Step (s2) of determining the upper limit value and the interval value In step (s2) of determining the upper limit and step values, an upper limit H of the height of the riser head 31 (see FIG. 5; the same applies below) is determined, and a positive step value h that is equal to or less than the upper limit H is also determined.

[0050] In step s2, the reference surface S of the riser portion 31 of the basic shape model 3 prepared in step s1 is B In the example of the feeder head 312 shown in FIG. 5, the reference plane S B An arbitrary upper limit value H2 is set with the height of the feeder head 311 and the feeder heads 313-315 shown in Figure 1 as 0 (zero). When an upper limit value H is set for the feeder head 311 and the feeder heads 313-315 shown in Figure 1, similarly arbitrary upper limits H1 and H3-H5 are set. These upper limits H may be different values ​​for each feeder head 31.

[0051] The step value h determined in the s2 step may be any positive value that is equal to or less than the upper limit value H. In the example shown in FIG. 5, the step value h2 is a value that is smaller than the upper limit value H2 and is located on the reference plane S B This indicates a positive value greater than 2 (height 0). Similarly, increment values ​​h1 and increment values ​​h3 to h5 (none shown) can be determined for feeder head 311 and feeder heads 313 to 315 shown in Figure 1. These increment values ​​h may be different values ​​for each feeder head 31.

[0052] (3) Step (s3) to calculate the dimension value In step (s3) of determining the dimension value, the upper limit value H determined in step s2 is divided by the step value h also determined in step s2 to determine a quotient, and the dimension value is determined by multiplying an integer less than or equal to this quotient and greater than or equal to 0 (zero) by the step value h. The dimension value is a value required to create the modified feeder section 41 in step (s4) of creating an extended shape model, which will be described later.

[0053] A specific method for determining the dimension values ​​will be explained using the example shown in Figure 5. For example, if the upper limit value H2 is set to 9 cm and the step value h2 is set to 3 cm in step s2, then in step s3, the quotient obtained by dividing the upper limit value H (9 cm) by the step value h2 (3 cm) is 3, and so the integer values ​​that are less than or equal to 3 and greater than or equal to 0 (zero) are 0, 1, 2, and 3. The product of these four integers multiplied by the step value h2 (3 cm) is the dimension value; in this example, four dimension values ​​are determined: 0 cm, 3 cm, 6 cm, and 9 cm. In a similar manner, the dimension values ​​of the other feeder sections 31 can be determined. Note that, unlike the example shown in Figure 5, if the upper limit value H is not an integer multiple of the step value h, the upper limit value H will not be divisible by the step value h, resulting in a remainder; in this case, the quotient will be the integer part.

[0054] (4) Step (s4) of creating an extended shape model In step (s4) of creating the extended geometric model 4, as will be explained using the example shown in Fig. 1 and Fig. 5, an extended geometric model 4 having one modified feeder 41 for each of the feeders 311 to 315 of the basic geometric model 3 is created. The modified feeder 41 is created by referring to the above-mentioned dimensional values, and the reference plane S B This can be created by deforming the riser 31 so that the (upper surface) is extended upward (in the direction of the arrow in the height direction z) by the dimension value. This operation can be performed manually by an operator using modeling software, but if this function is implemented in the casting analysis software, it can be used because it can be created automatically more quickly.

[0055] As mentioned above, when each of the feeders 311-315 has four dimensional values, a maximum of 1,024 different extended geometric models 4 can be created by selecting one shape from each of the modified feeders 411-415. However, since it is sufficient to create a necessary and sufficient number of extended geometric models 4, it is preferable to preselect the dimensional values ​​for creating the desired modified feeder 41 prior to step s4. Alternatively, an integer less than or equal to the quotient and greater than or equal to zero may be selected in step s3 before calculating the dimensional values. This selection may be made by the operator by referring to the integer greater than or equal to zero obtained in step s3 or the dimensional values ​​of each feeder 31, or may be automatically selected based on predetermined selection criteria. Alternatively, a table of combinations of selected dimensional values ​​may be created, and the extended geometric model 4 may be created by creating the modified feeder 41 using a function of the casting analysis software that references this table.

[0056] As mentioned above, it is preferable that one of the extended shape models 4 includes the basic shape model 3 itself, so it is preferable that for all feeder sections 31, a combination in which the extension dimension value is 0 (zero) is always included.

[0057] 6 may be executed after all of the extended shape models 4 based on the selected dimension values ​​are created in step s4, but if step s5 and subsequent steps are to be executed automatically, it is also possible to repeat the following operations: create one extended shape model 4 in step s4, execute steps s5 and s6, create training data based on this extended shape model 4 in step s7, and then execute step s4 again to create a different extended shape model 4. In other words, training data based on all combinations of extended shape models 4 may be created by sequentially repeating steps s4 to s7 according to the combination table described above.

[0058] (5) Step (s5) of creating an analytical model Next, the expanded geometric model 4 created in step s2 is divided into elements to create an analytical model (not shown) (s5). Here, the division into elements is also called mesh generation, and involves dividing the modeled geometric object into fine elements using mesh generation software and defining the characteristics of each element so that it can be used for various analyses using the finite element method, etc. Such a mesh generation function may be implemented in known casting analysis software, and this may be used if available.

[0059] (6) Step (s6) of acquiring coagulation analysis data In the step (s6) of acquiring solidification analysis data, solidification analysis is performed under predetermined analytical conditions for the analytical model created in step s5 using, for example, known casting analysis software, thereby acquiring solidification analysis data for all elements constituting the analytical model. The predetermined analytical conditions may be set each time in step s6, but since the same analytical conditions are generally applied to all extended shape models 4, the predetermined analytical conditions may be set in the casting analysis software and automatically loaded. Furthermore, the solidification analysis referred to here is not limited to the solidification phenomenon of the cast metal, but may also include all analyses of the series of phenomena from casting to the completion of solidification, such as the behavior of molten metal flow until it is filled into the mold cavity and mold deformation.

[0060] (7) Step (s7) of creating training data In the step (s7) of creating training data, a set of data related to the extended shape model 4 obtained up to step s6 is created as training data. That is, in step s7, training data having a structure is created, including shape data based on the extended shape model 4, analysis condition data applied to the coagulation analysis, and coagulation analysis data for all elements constituting the analytical model. Here, the shape data based on the extended shape model 4 may be data on the extended shape model 4 itself, or data (mesh data) of the analytical model created in step s5 based on the extended shape model 4. The format of the set of data constituting the training data including the shape data may be text format. Using training data written in text format is preferable because it increases versatility, such as enabling application to various machine learning devices. This training data is then created for all extended shape models 4. These sets of training data according to the present invention can be used to construct surrogate models.

[0061] [3] Third embodiment A third aspect of the present invention is a program that causes a computer to execute the series of operations described in the second embodiment. [Example]

[0062] (1) Step (s1) of preparing a basic shape model A basic shape model 3 shown in Fig. 1 was prepared. Specifically, it had a product section 30 equivalent to an exhaust manifold, with a longitudinal length of approximately 355 mm, a maximum dimension in the axial direction from the inlet flange to the outlet flange of approximately 85 mm, and a maximum dimension in the thickness direction of approximately 90 mm, and five approximately columnar feeder sections 311 to 315 integrally connected to the product section 30. The feeder sections 311 to 315 were aligned along a reference plane S perpendicular to the height direction z. B 1~S B 5 are defined, and the riser sections 311 to 315 and the reference surface S B 1~S B 5 and the intersecting surface, that is, the reference surface S B 1~S B 5 itself is the upper surface of the basic shape model 3. In this embodiment, the reference surface SB 1~S B The basic shape model 3 was created so that it was located close to the parting surface (parting surface, not shown) of the product part 30, where the effect of the feeder head is not substantially obtained in actual casting. The data format of the basic shape model 3 was STL format. The basic shape model 3 was then loaded into casting analysis software manufactured by MAGMA (MAGMASOFT, registered trademark; hereinafter also referred to as casting analysis software), and the following steps s2 to s7 were carried out using this casting analysis software as necessary.

[0063] (2) Step (s2) of determining the upper limit value and the interval value The upper limit values ​​H of the heights of the feeder sections 311 to 315 were set as H1 = 40 mm, H2 = 40 mm, H3 = 40 mm, H4 = 20 mm, and H5 = 40 mm, and the interval value h common to all feeder sections 31 was set to 20 mm.

[0064] (3) Step (s3) to calculate the dimension value In the casting analysis software, the quotient values ​​obtained by dividing the upper limit values ​​H1 to H5 by the interval value of 20 mm were 2 for feeder 311, 2 for feeder 312, 2 for feeder 313, 1 for feeder 314, and 2 for feeder 315. Therefore, the integer values ​​less than these quotients but greater than or equal to 0 were 0, 1, and 2 for feeder 311, 0, 1, and 2 for feeder 312, 0, 1, and 2 for feeder 313, 0, 1, and 2 for feeder 314, and 0, 1, and 2 for feeder 315. These integer values ​​were then multiplied by the interval value h (20 mm) to find dimensional values, and a combination table consisting of 162 combinations of deformed feeders 41 and their dimensional values ​​was created, including cases where the dimensional values ​​of all deformed feeders 41 were 0 mm.

[0065] (4) Step (s4) of creating an extended shape model The modified feeders 41 were created by referring to the 162 combinations of dimension values, and an extended shape model 4 was then created. In this example, one combination was created by referring to the combination table, and steps s5 to s7 were then repeated sequentially for the 162 combinations. Therefore, in the following steps s5 to s7, a description of this repetitive operation will be omitted. In this example, as described above, step s3 included a case where the dimension values ​​of all modified feeders 41 were 0 mm, and so one of the 162 combinations resulted in the creation of an extended shape model 4 equivalent to the basic shape model 3, in which the dimension values ​​of all modified feeders 41 were 0 mm.

[0066] (5) Step (s5) of creating an analytical model The extended geometric model 4 was divided into elements to create an analysis model (mesh model) with 7 million elements (mesh count).

[0067] (6) Step (s6) of acquiring coagulation analysis data As the specified analysis conditions, various physical property values ​​including the flow limit solid phase ratio of cast stainless steel as the assumed metallic material, 1550°C as the initial temperature of the molten metal, and 20°C as the initial temperature of the mold were set in the casting analysis software. Then, a solidification analysis was performed under these specified analysis conditions for the analysis model created in step s5, and solidification analysis data for all elements that make up the analysis model was obtained.

[0068] (7) Step (s7) of creating training data A data set including a series of data related to the extended shape model 4 used or obtained up to step s6, i.e., the specified analysis conditions set in step s6, the coagulation analysis data obtained in step s6, and the shape data based on the extended shape model 4 created in step s4, was created as training data.

[0069] The solidification analysis data consisted of the following: the time required for the elements that constitute the hot spot in the product portion 30 to form a hot spot (hot spot time with the flow limit solid fraction as a threshold), the solidification completion time of each element that constitutes the product portion 30, the temperatures of all elements that constitute the product portion 30 at the maximum of these solidification completion times (the time required for all elements that constitute the product portion 30 to complete solidification), and the porosity values ​​of all elements that constitute the product portion 30. [Explanation of symbols]

[0070] 1: Training data creation device (creation device) 10: Main body 100: Housing 101:CPU 102:RAM 103:ROM 104:STR 105(108~110):I / F 106: GPU 107: Internal bus 12: Input device 13: Output device 151: Acquisition Department 152: Setting section 153: Arithmetic section 154: Shape model extension part 155: Analysis model creation department 156:Analysis Department 157: Training Data Creation Department 158: Output section 159: Adjustment section 160: Storage section 200: Training data creation system (creation system) 210: Communication line 220: Terminal 230: Server 3: Basic shape model 30:Product Department 31 (311~315): Riser section 4: Extended shape model 41 (412, 4120-4123): Deformed riser section

Claims

1. a product portion and at least one substantially columnar riser portion integrally connected to the product portion; A reference plane perpendicular to the height direction of the riser portion is defined, an acquisition unit that acquires a basic shape model in which a plane where the feeder portion intersects with the reference plane is the upper surface of the feeder portion; A setting unit that determines an upper limit value of the height of the feeder portion, with the upward direction from the upper surface being positive, and an interval value that is equal to or less than the upper limit value; a calculation unit that calculates a quotient by dividing the upper limit value by the step value, and calculates a dimension value by multiplying the step value by an integer that is equal to or less than the quotient and is equal to or greater than 0; a shape model extension unit that creates an extended shape model having a deformed feeder portion in which the upper surface of the feeder portion is extended upward by the dimension value; an analytical model creation unit that divides the extended geometric model into elements to create an analytical model; an analysis unit that performs coagulation analysis on the analysis model under predetermined analysis conditions and acquires coagulation analysis data for all of the elements that constitute the analysis model; For the augmented shape model, the predetermined analysis conditions; The coagulation analysis data; Shape data based on the extended shape model A teacher data creation unit that creates teacher data including an output unit that outputs the teacher data; A device for creating training data having the above structure.

2. a product portion and at least one substantially columnar riser portion integrally connected to the product portion; A reference plane perpendicular to the height direction of the riser portion is defined, The plane where the feeder portion and the reference plane intersect is the upper surface of the feeder portion. preparing a basic shape model; determining an upper limit value of the height, with the upward direction from the upper surface being positive, and an interval value that is equal to or less than the upper limit value; a step of obtaining a quotient by dividing the upper limit value by the step value, and multiplying an integer equal to or less than the quotient and equal to or greater than 0 by the step value to obtain a dimension value; creating an extended geometric model having a deformed feeder portion in which the top surface of the feeder portion is extended upward by the dimension value; creating an analytical model by dividing the plurality of extended geometric models into elements; performing a coagulation analysis on the analytical model under predetermined analytical conditions to obtain coagulation analysis data for all of the elements constituting the analytical model; Including, For the augmented shape model, the predetermined analysis conditions; The coagulation analysis data; The shape data based on the extended shape model A method for creating training data having the following structure.

3. The method for generating training data according to claim 2 , wherein the extended shape model includes the basic shape model.

4. a product portion and at least one substantially columnar riser portion integrally connected to the product portion; A reference plane perpendicular to the height direction of the riser portion is defined, The plane where the feeder portion and the reference plane intersect is the upper surface of the feeder portion. obtaining a base shape model; acquiring an upper limit value of the height, with the upward direction from the upper surface being positive, and an interval value that is equal to or less than the upper limit value; a step of obtaining a quotient by dividing the upper limit value by the step value, and multiplying the step value by an integer equal to or less than the quotient and equal to or greater than 0 to obtain a dimension value; creating an extended geometric model having a deformed feeder portion in which the top surface of the feeder portion is extended upward by the dimension value; creating an analytical model by dividing the extended geometric model into elements; performing a coagulation analysis on the analytical model under predetermined analytical conditions to obtain coagulation analysis data for all of the elements constituting the analytical model; For the augmented shape model, the predetermined analysis conditions; The coagulation analysis data; Shape data based on the extended shape model Creating training data including: A program that causes a computer to execute the following.

Images