Performance prediction device for seismic isolation device, performance prediction method for seismic isolation device, and performance prediction program for seismic isolation device

Abstract

To provide a performance prediction device for a seismic isolation device, a performance prediction method for the seismic isolation device and a performance prediction program for the seismic isolation device that can predict the performance of the seismic isolation device with less labor, time, and cost than conventional techniques, without the need to perform analysis in a finite element method using a finite element model for an actual product.SOLUTION: In a performance prediction device for a seismic isolation device using an elastic body, a variable generation unit 202 randomly generates variables including a part or all of the shape of the elastic body, material constants, and force application conditions, a finite element model generation unit 204 generates a finite element model on the basis of the generated variables, an analysis unit 206 executes analysis using the generated finite element model, a calculation unit 208 calculates mechanical properties of the seismic isolation device on the basis of a result of the executed analysis, and a learned model generation unit 210 generates a learned model for predicting the properties of the seismic isolation device on the basis of a correlation when the generated variables are used as input and the calculated mechanical properties are used as output.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to a performance prediction device for a seismic isolation device, a performance prediction method for a seismic isolation device, and a performance prediction program for a seismic isolation device.

Background Art

[0002] As a seismic isolation device, for example, a pair of connecting plates are integrally vulcanized to both upper and lower ends in the axial direction of a laminate in which rubber layers and rigid plates having higher rigidity than the rubber layers are alternately laminated, and rigid flanges integrated by assembly bolts are provided on each of the upper and lower connecting plates. After inserting mounting bolts through the mounting holes provided in the rigid flanges, the mounting bolts are screwed into the upper structure and the lower structure respectively to fix them to the lower structure and the upper structure (see Patent Document 1 and Patent Document 2). Note that the connecting plates may be absent, and flanges may be provided at the upper and lower ends instead of the connecting plates.

[0003] For the purpose of analyzing the behavior of such a seismic isolation device and appropriately setting the design values of the seismic isolation device and the structure, a displacement amount input step is executed in which information indicating a displacement amount that changes over time is input from a displacement amount input unit, an analysis step is executed to calculate a reaction force corresponding to the input displacement amount and the history of the displacement amount, and an analysis result output step is executed to output an analysis result indicating the reaction force calculated by the analysis step. Here, each of these displacement amount input steps is executed every time one displacement amount is input. In the analysis result output step, a technique of outputting a characteristic diagram of displacement amount versus reaction force by a printer or the like based on the reaction force calculated by the analysis step is known (see Patent Document 3).

[0004] In addition, when performing seismic response analysis of a seismic isolation structure supported through a seismic isolation device having a laminated rubber body formed by alternately laminating and vulcanizing and bonding a plurality of rubbers and steel plates, and a lead plug embedded in the central portion of the laminated rubber body, a history characteristic model of the seismic isolation device can accurately represent the rise of the initial displacement, and moreover, it can be simulated accurately with the same model regardless of the magnitude of the displacement, aiming to improve the accuracy of seismic response analysis of the seismic isolation structure. For the seismic isolation device, a history characteristic model is set in which a plurality of spring-slider elements in which a spring element and a slider element are connected in series, one spring element, and one dashpot element are connected in parallel to each other, and based on this history characteristic model, seismic response analysis of the seismic isolation structure is performed, whereby it is known that a history characteristic model similar to an actual seismic isolation device having both elastic and plastic characteristics can be obtained (see Patent Document 4).

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Summary of the Invention

Problems to be Solved by the Invention

[0006] Conventionally, when it is desired to analyze the buckling characteristics of a seismic isolation device, as a conceivable method, it is necessary to fabricate an actual device (actual product) of the seismic isolation device and conduct a horizontal loading test under high surface pressure on the fabricated actual product. For the evaluation of these actual products of the seismic isolation device, a lot of man-hours, time, and cost have been incurred. Instead of actually fabricating the actual product of the seismic isolation device, it is also possible to analyze a modeled seismic isolation device by a program. However, when modeling the seismic isolation device in this way, since it is necessary to create a plurality of seismic isolation device models corresponding to each of a plurality of parameters, although it is reduced compared to the case of fabricating and evaluating an actual product, still, a lot of man-hours, time, and cost are incurred for the analysis. The present invention provides a performance prediction device for a seismic isolation device using an elastic body, which randomly generates variables including the shape, material constants, and surface pressure of the elastic body, generates a finite element model based on the generated variables, executes an analysis of the seismic isolation device using the generated finite element model, calculates mechanical characteristics related to the seismic isolation device based on the result of the executed analysis, and generates a learned model for predicting the characteristics of the seismic isolation device based on the correlation when the generated variables are used as inputs and the calculated mechanical characteristics are used as outputs. The purpose is to provide a performance prediction device for a seismic isolation device, a performance prediction method for a seismic isolation device, and a performance prediction program for a seismic isolation device that can predict the performance of the seismic isolation device with less man-hours, time, and cost than conventional methods without the need to perform an analysis by the finite element method using a finite element model for an actual product.

Means for Solving the Problems

[0007] The present invention has been made in view of such problems, and in a performance prediction device for a seismic isolation device using an elastic body, a variable generation unit that randomly generates variables including the shape, material constants, and surface pressure of the elastic body; a finite element model generation unit that generates a finite element model based on the generated variables; an analysis unit that executes an analysis of the seismic isolation device using the generated finite element model; a calculation unit that calculates mechanical characteristics related to the seismic isolation device based on the result of the executed analysis; A seismic isolation device performance prediction apparatus, comprising: a learned model generation unit that generates a learned model for predicting the characteristics of the seismic isolation device based on a correlation when the generated variable is used as an input and the calculated mechanical characteristics are used as an output.

[0008] According to the above configuration, by generating a learned model in which the correlation when the generated variable is used as an input and the calculated mechanical characteristics related to the seismic isolation device are used as an output is learned, the characteristics of the seismic isolation device can be predicted, so that an optimal learned model can be generated. Therefore, it is not necessary to perform an analysis by the finite element method using a finite element model for the actual product, and the performance of the seismic isolation device can be predicted with less man-hours, time, and cost than before.

[0009] In the above configuration, the analysis of the seismic isolation device is a buckling analysis, The mechanical characteristic is a buckling strain, which is characterized.

[0010] According to the above configuration, by generating a learned model in which the correlation when the generated variable is used as an input and the calculated buckling strain is used as an output is learned, the characteristics of the seismic isolation device can be predicted, so that an optimal learned model can be generated. Therefore, it is not necessary to perform an analysis by the finite element method using a finite element model for the actual product, and the performance of the seismic isolation device can be predicted with less man-hours, time, and cost than before.

[0011] In the above configuration, the analysis of the seismic isolation device is a fracture analysis, The mechanical characteristic is a fracture strength, which is characterized.

[0012] According to the above configuration, by generating a learned model in which the correlation when the generated variable is used as the input and the calculated breaking strength is used as the output is learned, the characteristics of the seismic isolation device can be predicted. Therefore, an optimal learned model can be generated, eliminating the need to perform finite element method analysis using a finite element model for the actual product, and enabling the performance prediction of the seismic isolation device with less man-hours, time, and cost than before.

[0013] In the above configuration, the analysis of the seismic isolation device is tensile analysis, The mechanical property is characterized by being the tensile strength

[0014] According to the above configuration, by generating a learned model in which the correlation when the generated variable is used as the input and the calculated tensile strength is used as the output is learned, the characteristics of the seismic isolation device can be predicted. Therefore, an optimal learned model can be generated, eliminating the need to perform finite element method analysis using a finite element model for the actual product, and enabling the performance prediction of the seismic isolation device with less man-hours, time, and cost than before. 。

[0015] In the above configuration, The performance prediction device for the seismic isolation device further is a performance prediction device for a seismic isolation device having a characteristic prediction unit that predicts the characteristics of the seismic isolation device using the generated learned model.

[0016] According to the above configuration, since it has a characteristic prediction unit that predicts the characteristics of the seismic isolation device using the generated learned model, there is no need to perform finite element method analysis using a finite element model for the actual product, and the performance prediction of the seismic isolation device can be enabled with less man-hours, time, and cost than before.

[0017] The above configuration further In the performance prediction device for the seismic isolation device, The characteristics of the seismic isolation device are characterized by being the buckling strain of the seismic isolation device, which is a performance prediction device for the seismic isolation device.

[0018] According to the above configuration, in order to predict the buckling strain of the seismic isolation device as a characteristic of the seismic isolation device, it is not necessary to perform an analysis by the finite element method using a finite element model for the actual product, and it is possible to predict the performance regarding the buckling strain of the seismic isolation device with less man-hours, time, and cost than before.

[0019] The above configuration further In the performance prediction device of the seismic isolation device The characteristic of the seismic isolation device is a performance prediction device of the seismic isolation device that is one or more compressive limit strength diagrams of the seismic isolation device.

[0020] According to the above configuration, in order to predict one or more compressive limit strength diagrams of the seismic isolation device as a characteristic of the seismic isolation device, it is not necessary to perform an analysis by the finite element method using a finite element model for the actual product, and it is possible to predict the performance regarding the compressive limit strength diagrams of one or more seismic isolation devices with less man-hours, time, and cost than before.

[0021] In the above configuration The seismic isolation device is a performance prediction device of a seismic isolation device having a rubber body with a predetermined shape.

[0022] According to the above configuration, for a seismic isolation device having a rubber body with a predetermined shape, it is not necessary to perform an analysis by the finite element method using a finite element model for the actual product, and it is possible to predict the performance regarding the compressive limit strength diagram of the seismic isolation device with less man-hours, time, and cost than before.

[0023] In the seismic isolation device The predetermined shape is a performance prediction device that is a cylindrical shape.

[0024] According to the above configuration, for a seismic isolation device having a cylindrical rubber body, it is not necessary to perform an analysis by the finite element method using a finite element model for an actual product, and it is possible to predict the performance regarding the compression limit strength diagram of the seismic isolation device with less man-hours, time, and cost than before.

[0025] In the seismic isolation device, The predetermined shape is a performance prediction device characterized by being a shape formed by a combination of a flare shape and a cylindrical shape.

[0026] According to the above configuration, for a seismic isolation device having a rubber body with a shape formed by a combination of a flare shape and a cylindrical shape, it is not necessary to perform an analysis by the finite element method using a finite element model for an actual product, and it is possible to predict the performance regarding the compression limit strength diagram of the seismic isolation device with less man-hours, time, and cost than before.

[0027] The present invention further provides a method for predicting the performance of a seismic isolation device using an elastic body, a computer randomly generates variables including the shape, material constants, and surface pressure of the elastic body, generates a finite element model based on the generated variables, performs an analysis of the seismic isolation device using the generated finite element model, calculates the mechanical characteristics regarding the seismic isolation device based on the results of the performed analysis, and is a method for predicting the performance of a seismic isolation device, which generates a learned model for predicting the characteristics of the seismic isolation device based on the correlation relationship when the generated variables are used as inputs and the calculated mechanical characteristics are used as outputs.

[0028] According to the above configuration, a computer randomly generates variables including the shape, material constants, and surface pressure of the elastic body, generates a finite element model based on the generated variables, performs analysis of the seismic isolation device using the generated finite element model, calculates the mechanical properties of the seismic isolation device based on the results of the performed analysis, and generates a learned model for predicting the characteristics of the seismic isolation device based on the correlation relationship when the generated variables are input and the calculated mechanical properties are output. Therefore, an optimal learned model can be generated, eliminating the need to perform finite element method analysis using a finite element model for the actual product, and enabling performance prediction of the seismic isolation device with fewer man-hours, time, and costs than before.

[0029] The present invention further provides a performance prediction program for a seismic isolation device using an elastic body, wherein the computer is caused to randomly generate variables including the shape, material constants, and surface pressure of the elastic body, generate a finite element model based on the generated variables, perform analysis of the seismic isolation device using the generated finite element model, calculate the mechanical properties of the seismic isolation device based on the results of the performed analysis, and generate a learned model for predicting the characteristics of the seismic isolation device based on the correlation relationship when the generated variables are input and the calculated mechanical properties are output.

[0030] According to the above configuration, the computer is caused to randomly generate variables including the shape, material constants, and surface pressure of the elastic body, generate a finite element model based on the generated variables, execute analysis of the seismic isolation device using the generated finite element model, calculate the mechanical properties related to the seismic isolation device based on the results of the executed analysis, and generate a learned model for predicting the characteristics of the seismic isolation device based on the correlation when the generated variables are used as inputs and the calculated mechanical properties are used as outputs. As a result, an optimal learned model can be generated because the characteristics of the seismic isolation device can be predicted. Therefore, it is not necessary to perform analysis by the finite element method using a finite element model for the actual product, and the performance prediction of the seismic isolation device can be enabled with fewer man-hours, time, and costs than in the conventional case.

Advantages of the Invention

[0031] According to the present invention, by generating a learned model in which the correlation when the generated variables are used as inputs and the calculated buckling strain is used as an output is learned, an optimal learned model can be generated because the characteristics of the seismic isolation device can be predicted. Therefore, it is not necessary to perform analysis by the finite element method using a finite element model for the actual product, and the performance prediction of the seismic isolation device can be enabled with fewer man-hours, time, and costs than in the conventional case.

Brief Description of the Drawings

[0032]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Embodiments for Carrying Out the Invention

[0033] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. As an example of the seismic isolation device 1 whose performance is to be predicted in the present invention, it will be described based on FIG. 1. In the description of the present invention, the "central axis L" of the seismic isolation device 1 (hereinafter simply referred to as the "central axis L") is the central axis of the rubber body 2. The central axis L of the seismic isolation device 1 is directed so as to extend in the vertical direction. The "inner peripheral side", "outer peripheral side", "radial direction", and "circumferential direction" of the seismic isolation device 1 respectively refer to the "inner peripheral side", "outer peripheral side", "radial direction", and "circumferential direction" when centered on the central axis L of the seismic isolation device 1. Also, "up" and "down" respectively refer to "up" and "down" in the vertical direction.

[0034] As shown in Fig. 1, the seismic isolation device 1 includes a rubber body 2 having a conventional cylindrical structure, a pair of connecting plates 6 integrally formed on both sides of the central axis L of the rubber body 2, for example, and a pair of rigid flanges 10 that are assembled to each of the pair of connecting plates 6 by assembly bolts 15 and have a larger diameter than the connecting plates 6. Note that the connecting plates may be omitted, and flanges may be provided at the upper and lower ends instead of the connecting plates. The seismic isolation device 1 is disposed between an upper structure 20A and a lower structure 20B, and is fixed to each of the upper structure 20A and the lower structure 20B by mounting bolts 16. The upper structure 20A is, for example, a building such as a building or a bridge, and the lower structure 20B is, for example, the foundation of a building.

[0035] The seismic isolation device 1 includes a columnar rubber body 2 having a predetermined height and located at the vertical center of the seismic isolation device 1. The rubber body 2 of the present embodiment is formed in a cylindrical shape, but may be formed in a square column shape or a polygonal column shape. The rubber body 2 has a predetermined rigidity against vertical loads and can ensure a predetermined amount of deformation against horizontal loads.

[0036] Furthermore, a pair of connecting plates 6 made of metal, for example, steel, are arranged so as to cover the upper and lower ends of the rubber body 2. The rubber body 2 and the pair of connecting plates 6 are placed in a mold (not shown) and adhered by vulcanization molding to be firmly integrated.

[0037] As shown in Fig. 1, the assembly bolt 15 includes a shaft portion 15a and a head portion 15b having a larger diameter than the shaft portion 15a. The mounting bolt 16 includes a shaft portion 16a and a head portion 16b having a larger diameter than the shaft portion 16a.

[0038] As shown in Fig. 1, the assembly hole 11 is formed to a predetermined depth from the surface of the rigid flange 10 in contact with the connecting plate 6. On the side of the rigid flange 10 in contact with the structure 20, a head accommodation hole 12 for accommodating the head portion 15b of the assembly bolt 15 is formed so as to communicate with the assembly hole 11.

[0039] To attach the rigid flange 10 to the connecting plate 6, the assembly bolt 15 is inserted through the head receiving hole 12 of the rigid flange 10 toward the assembly hole 11. When the tip reaches the assembly bolt hole 7 of the connecting plate 6, the assembly bolt 15 is screwed into the assembly bolt hole 7. When the assembly bolt 15 is completely screwed into the assembly bolt hole 7, the head 15b of the assembly bolt 15 is received in the head receiving hole 12 having a diameter larger than that of the assembly hole 11. The lower surface of the head 15b of the assembly bolt 15 abuts against the step formed by the difference in diameter between the assembly hole 11 and the head receiving hole 12, and the rigid flange 10 is fixed to the connecting plate 6 by the assembly bolt 15. Since the head 15b of the assembly bolt 15 is completely received in the head receiving hole 12, it does not prevent the attachment of the rigid flange 10 to the structure 20.

[0040] FIG. 2 is a functional block diagram of a performance prediction device for a seismic isolation device. The performance prediction device 1 is a prediction device that predicts the performance of a seismic isolation device having a rubber body of a predetermined shape. Here, the performance prediction device for a seismic isolation device is generally an information processing device such as a computer having a central processing unit (CPU) and a main memory (memory) connected to the central processing unit.

[0041] The present invention relates to a performance prediction device for a seismic isolation device using an elastic body, and includes a variable generation unit that randomly generates variables including part or all of the shape, material constants, and loading conditions of the elastic body, a finite element model generation unit that generates a finite element model based on the generated variables, an analysis unit that performs an analysis of the seismic isolation device using the generated finite element model, a calculation unit that calculates mechanical characteristics related to the seismic isolation device based on the results of the executed analysis, and a learned model generation unit that generates a learned model for predicting the characteristics of the seismic isolation device based on the correlation when the generated variables are used as inputs and the calculated mechanical characteristics are used as outputs.

[0042] In one embodiment of the present invention, an analysis unit of a seismic isolation device that analyzes using a finite element model generated by a finite element model generation unit based on generated variables performs buckling analysis, and a calculation unit that calculates based on the result of the executed buckling analysis predicts the performance of the seismic isolation device using buckling strain. In the performance prediction device of the seismic isolation device according to the present embodiment, the analysis is performed by buckling analysis, and the performance of the seismic isolation device is predicted with the mechanical characteristics of the seismic isolation device as buckling strain. However, for example, the analysis may be fracture analysis or tensile analysis, and the mechanical characteristics of the seismic isolation device may be fracture strength or tensile strength.

[0043] In FIG. 2, the variable generation unit 202 randomly generates a plurality of variable data 203 including material constants of rubber and surface pressure on the rubber, which will be described later, and outputs the generated plurality of variable data 203 to the finite element model generation unit 204. Next, the finite element model generation unit 204 inputs the variable data 203 generated and output by the variable generation unit 203, generates a finite element model 205 of the seismic isolation device corresponding to the variable data 203, and outputs the generated finite element model 205 of the seismic isolation device. Here, the finite element model 205 refers to a model used in the finite element method (FEM: Finite Element Method) that, when performing analysis or simulation of physical phenomena, does not derive and solve the equations of the entire model, but decomposes the model into a plurality of meshes, solves simpler equations corresponding to each of the plurality of meshes, and combines the solutions of each mesh to derive the final solution.

[0044] Then, the buckling analysis unit 206 executes buckling analysis using the finite element model 205 generated by the finite element model generation unit 204. Here, "buckling" means that when the load applied to a structure is gradually increased, the structure suddenly undergoes a large deflection or other change at a certain load. More specifically, with respect to buckling in a seismic isolation device, it refers to the phenomenon where the stress-strain curve becomes negatively sloped. Also, "buckling strain" is the amount of deformation (strain) per unit length [%] when buckling occurs in a structure, and the strain mainly occurs in the end region of the structure. Here, the buckling analysis unit 206 applies a compressive force corresponding to the surface pressure in the vertical direction to the finite element model 205 of the seismic isolation device from 0%, and outputs, as buckling analysis result 207, correlation data representing the correlation between the shear strain and the shear stress when buckling occurs in the finite element model 205. For example, the buckling analysis result 207 may be an SS (Shear Stain-Shear Stress) curve representing the correlation between the shear strain and the shear stress of the finite element model 205. Also, for the buckling analysis, the force is applied up to 400%, and for example, if no buckling strain occurs even when the force is applied up to 400%, it may be output as buckling 400%.

[0045] Next, the buckling strain calculation unit 208 calculates a buckling strain 209 based on the buckling analysis result 207, which is correlation data representing the correlation between the shear strain and the shear stress when buckling occurs in the finite element model 205 generated and output by the buckling analysis unit 206. For example, when the buckling strain calculation unit 208 represents the shear strain and the shear stress on an SS curve for the correlation data representing the correlation between the shear strain and the shear stress when buckling occurs in the finite element model 205 generated and output by the buckling analysis unit 206, the buckling strain 209 of the finite element model 205 may be calculated based on, for example, the slope of the SS curve.

[0046] More specifically, generally for the SS curve (not shown) of the rubber constituting the seismic isolation device, as the shear strain increases, the value of the shear stress increases monotonically, and the slope of the SS curve is a positive gradient up to a predetermined shear strain value. However, at a certain point of the shear strain value, the slope of the SS curve becomes 0, and thereafter the slope becomes a negative gradient. In this case, for example, the value of the shear strain at the point when the slope of the SS curve becomes 0, that is, the moment when the negative gradient of the SS curve occurs, may be calculated as the buckling strain 209. Even when there is no negative gradient in the SS curve, when a phenomenon is observed in which the gradient of the SS curve becomes small and then becomes large again, for example, the buckling strain 209 of the finite element model 205 may be calculated as if buckling occurred at the point when the gradient became small.

[0047] Then, the learned model generation unit 210 inputs the buckling strain 209 calculated by the buckling strain calculation unit 208 and performs machine learning on the feedforward neural network described later, thereby generating the learned AI model 212. Here, the feedforward neural network to be learned is merely an example of an AI model, and the AI model in the present invention is not limited to the feedforward neural network. As other methods, for example, gradient boosting, random forest, support vector machine, etc. may be used. Through the above variable generation unit 202, finite element model generation unit 204, buckling analysis unit 206, buckling strain calculation unit 208, and learned model generation unit 210, the generation of the learned AI model 212 as preprocessing is performed.

[0048] Next, after generating the learned AI model 212 as preprocessing, a method for predicting the characteristics of the seismic isolation device using the generated learned AI model 212 will be described. First, LSS test data 211 for a lap shear sample (hereinafter referred to as "LSS" (Lap Shear Sample)) in a laboratory (experimental equipment) is prepared in advance, and by inputting the prepared LSS test data 211 into the learned AI model 212, characteristic prediction result data 213 to be predicted can be obtained. Here, the LSS test data 211 prepared in advance may be prepared using a spreadsheet described later. Further, the obtained characteristic prediction result data 213 may include the buckling strain and / or one or more ultimate property diagrams (UPD) described later.

[0049] Figure 3 shows an example of a finite element model 205 (FEM model: Finite Element Method model) (mesh not shown for simplicity) with a flare structure that is a combination of the flare shape and the cylindrical shape generated by the finite element model generation unit 204 in FIG. 2 according to an embodiment of the present invention. As described above, the finite element model is a model used in the finite element method (FEM: Finite Element Method) that, when analyzing or simulating physical phenomena, does not derive and solve the equations for the entire model, but decomposes the model into a plurality of meshes, solves simpler equations corresponding to each of the plurality of meshes, and combines the solutions of each mesh to derive the final solution. Note that the finite element model in the present invention is not limited to the flare shape, and may be other shapes such as a cylindrical shape.

[0050] Here, FIG. 4 corresponds to the finite element model 205 in FIG. 3 according to an embodiment of the present invention, and shows a list of the parameters to be generated and the random value generation intervals as an overview of the finite element model generated by the finite element model generation unit 204 in FIG. 2. S1 represents the first-order shape coefficient of the finite element model 205, and the random value generation range by the finite element model generation unit 204 is, for example, in the range of 25 to 40, but the present invention is not limited thereto. S2 represents the second-order shape coefficient of the finite element model 205, and the random value generation range by the finite element model generation unit 204 is, for example, in the range of 2.8 to 4.3, but the present invention is not limited thereto. R1 represents the thickness ratio of the rubber and the steel plate of the finite element model 205, and the random value generation range by the finite element model generation unit 204 is, for example, in the range of 0.24 to 1, but the present invention is not limited thereto. R2 represents the value obtained by dividing the difference between the central diameter and the maximum diameter of the finite element model 205 by the central diameter, and the random value generation range by the finite element model generation unit 204 is, for example, in the range of 0 to 0.35, but the present invention is not limited thereto. R3 represents the value obtained by dividing the difference between the central diameter and the maximum diameter of the finite element model 205 by the total rubber thickness of the upper and lower inclined portions. When there is no inclined portion, it is 0. The random value generation range by the finite element model generation unit 204 is, for example, in the range of 0 to 55, but the present invention is not limited thereto. R4 represents the ratio of the inner diameter to the central diameter of the finite element model 205. The random value generation range by the finite element model generation unit 204 is, for example, in the range of 0 to 0.55, but the present invention is not limited thereto.

[0051] σ represents the surface pressure of the finite element model 205. The random value generation range by the finite element model generation unit 204 is, for example, in the range of the reference surface pressure to twice the reference surface pressure, but the present invention is not limited thereto. Here, the "reference surface pressure" is the surface pressure corresponding to 10% to 30% of the compressive limit strength when the horizontal deformation is 0, and it is the reference surface pressure determined when measuring the basic performance of the seismic isolation device.

[0052] C 10 represents the material constant that designates the deviation portion (shape change) of the material in the deformation model deformed from the Yeoh model, which is the material model related to the finite element model 205. The random value generation range by the finite element model generation unit 204 is, for example, in the range of 0.13 to 0.26 [MPa], but the present invention is not limited thereto. C 30 represents the material constant that designates the deviation portion (shape change) of the material in the deformation model deformed from the Yeoh model, which is the material model related to the finite element model 205. The random value generation range by the finite element model generation unit 204 is, for example, in the range of 0.00005 to 0.0003 [MPa], but the present invention is not limited thereto. k represents the bulk modulus of elasticity in the deformation model deformed from the Yeoh model, which is the material model related to the finite element model 205. The random value generation range by the finite element model generation unit 204 is, for example, in the range of 700 to 2,100 [MPa], but the present invention is not limited thereto.

[0053] The above deformation model, which is a type of strain energy density function used in the analysis of elastic materials, is represented by the following equation (1).

[0054] [Number] ···(1) W: Strain energy density (strain energy per unit volume of the object) C 10 : Material constant I1: Strain invariant C 30 : Material constant k: Bulk modulus J: Jacobian

[0055] FIG. 5 is a schematic diagram showing the neural network structure of the learned AI model 210 generated by the learned model generation unit 210 of FIG. 2 in an embodiment of the present invention. Here, the AI model 301 is, for example, a four-layer feedforward neural network with three intermediate layers, but the neural network is not limited to this configuration, and other methods such as gradient boosting, random forest, support vector machine, etc. may also be used. More specifically, the input layer 303, intermediate layer 304, and output layer 305 of the AI model 301 each include a plurality of nodes 302, and when learning is performed on the AI model 301, weight values as the results of learning are set for each node 302.

[0056] FIG. 6 shows an overview of the AI model 301 of FIG. 5 in an embodiment of the present invention. The number of input units is, for example, S1, S2, R1, R2, R3, R4, σ, C 10, C 30, k, a total of 10 units, but the present invention is not limited to this number of units, and other shape coefficients, loading condition parameters, and material constants of other material models may also be used.

[0057] The number of units in the intermediate layers 1 - 3 is 35 units, but the present invention is not limited to this number of units. Also, as the activation function of the intermediate layers 1 - 3, for example, the ReLU (Rectified Linear Unit) function is used, but the present invention is not limited thereto, and other appropriate functions may be used. Here, the ReLU (Rectified Linear Unit) function is a function in which the output value becomes 0 when the input value is 0 or less, and the output value becomes the same value as the input value when the input value is greater than 0.

[0058] The number of units in the output layer is 21 units with a resolution of 0.1 increments from 2.0 to 4.0 corresponding to the value of the buckling strain, but the present invention is not limited to this increment width and / or the number of units. In one embodiment of the present invention, by using the learned AI model 210 generated by the learned model generation unit 210 in FIG. 2, it becomes possible to predict the buckling strain with a resolution of 0.1 increments in the range of 2.0 to 4.0. Note that the above - mentioned learned AI model may be a regression AI that outputs a continuous value as a prediction result instead of a classification AI that classifies into the above - mentioned predetermined range as a prediction result.

[0059] FIG. 7 shows various parameters and the like for generating a learned AI model regarding the learning of the AI model 301 in FIG. 5 in one embodiment of the present invention. The number of training data is the number of training data (teacher data) used for machine learning of the AI model 301. For example, 5,920 pieces of training data are used for learning, but the present invention is not limited to this number of training data. The number of test data is the number of data (test data) used for performance evaluation of the learned model 212 in FIG. 2. Generally, unknown data different from the training data is prepared. For example, 1,480 pieces of test data are used for performance evaluation, but the present invention is not limited to this number of test data. The total number of data is the sum of the above-mentioned number of training data and the number of test data. For example, in the case of the above-mentioned number of training data and the number of test data, the total number of data is 7,200, but the present invention is not limited to this total number of data.

[0060] The initial learning rate is a value set as the initial value of the learning rate. Specifically, it refers to a hyperparameter that represents how much the weight value is changed at one time in the optimization of machine learning. In this case, a predetermined initial learning rate is set as the initial learning rate. The learning rate reduction rate refers to the reduction rate multiplied by the initial learning rate in the learning rate decay method, which is a method used to improve the performance of machine learning. When learning has progressed to a certain extent, the learning rate is reduced for the purpose of improving learning accuracy, and a predetermined learning rate reduction rate is set as the learning rate reduction rate.

[0061] The batch size is a hyperparameter that refers to the number of data input to the AI model 301 at one time, and a predetermined batch size is set as the batch size. However, the present invention is not limited to the value of this batch size. Here, generally, when the batch size is large, the AI model can learn more data at one time and the learning efficiency is high, but the calculation time will increase. On the other hand, when the batch size is small, the calculation time is shortened, but the learning efficiency of the AI model will decrease.

[0062] The number of epochs is one of the hyperparameters in machine learning. In the learning of the AI model 301, it refers to the number of learning repetitions of the entire training data, and a predetermined number of epochs is set as the number of epochs. In machine learning, it is necessary to predict a value from the training data and reduce the difference between the predicted value and the correct value. Therefore, when the number of epochs is small, there is a risk that learning will end before the parameters converge properly. On the other hand, when the number of epochs is large, there is a risk that a problem of overfitting will occur only for specific training data. Therefore, it is desirable to set an appropriate number of epochs.

[0063] FIG. 7 is a flowchart of a part for generating a learned AI model among a performance prediction method of a seismic isolation device and a flowchart of a performance prediction program of the seismic isolation device. Here, the performance prediction method of the seismic isolation device and the performance prediction program of the seismic isolation device are executed by the performance prediction device 201 of the seismic isolation device in FIG. 2, and generally, they are executed by an information processing device such as a computer having a central processing unit (CPU) and a main storage device (memory) connected to the central processing unit.

[0064] Regarding the operation of generating the learned AI model in the performance prediction method of the seismic isolation device and the performance prediction program of the seismic isolation device, the operation starts from the start of step S801 (S801). In step S802, the variable generation unit 202 in FIG. 2 randomly generates a plurality of variable data 203 (see FIG. 4) in FIG. 2 including material constants and surface pressure, and outputs the generated plurality of variable data 203 to the finite element model generation unit 204 in FIG. 2 (S802). Next, in step S803, the finite element model generation unit 204 inputs the variable data 203 generated and output by the variable generation unit 203, generates a finite element model 205 of the seismic isolation device in FIG. 2 corresponding to the variable data 203, and outputs the generated finite element model 205 of the seismic isolation device (S803).

[0065] Then, in step S804, the buckling analysis unit 206 in FIG. 2 executes buckling analysis using the finite element model 205 generated by the finite element model generation unit 204 (S804). Here, the buckling analysis unit 206 applies a compressive force corresponding to the surface pressure in the vertical direction to the finite element model 205 of the seismic isolation device from 0% and outputs, as the buckling analysis result 207 in FIG. 2, correlation data representing the correlation between the shear strain and the shear stress when buckling occurs in the finite element model 205. For example, the buckling analysis result 207 may be an SS (Shear Stain-Shear Stress) curve representing the correlation between the shear strain and the shear stress of the finite element model 205. Also, for the buckling analysis, the force is applied up to 400%, and for example, if no buckling strain occurs even when the force is applied up to 400%, it may be output as buckling 400%.

[0066] Next, in step S805, the buckling strain calculation unit 208 in FIG. 2 calculates the buckling strain 209 in FIG. 2 (S805) based on the correlation data, which is the buckling analysis result 207 representing the correlation between the shear strain and the shear stress when buckling occurs in the finite element model 205 generated and output by the buckling analysis unit 206. For example, when the shear strain and the shear stress are represented on an SS curve for the correlation data representing the correlation between the shear strain and the shear stress when buckling occurs in the finite element model 205 generated and output by the buckling analysis unit 206, the buckling strain 209 of the finite element model 205 may be calculated based on, for example, the slope of the SS curve.

[0067] More specifically, generally for the SS curve (not shown) of the rubber constituting the seismic isolation device, as the shear strain increases, the value of the shear stress increases monotonically, and the slope of the SS curve is a positive gradient. However, at a certain point, the slope of the SS curve becomes 0, and thereafter, the slope becomes a negative gradient. In this case, for example, the value of the shear strain at the point when the slope of the SS curve becomes 0, that is, the moment when the negative gradient of the SS curve occurs, may be calculated as the buckling strain 209. Even when the negative gradient does not occur in the SS curve, when a phenomenon is observed in which the gradient of the SS curve becomes small and then becomes large again, for example, the buckling strain 209 of the finite element model 205 may be calculated assuming that buckling has occurred at the point when the gradient has become small.

[0068] Then, in step S806, in FIG. 2, the learned model generation unit 210 inputs the buckling strain 209 calculated by the buckling strain calculation unit 208 and executes machine learning for the feedforward neural network, thereby generating the learned AI model 212 in FIG. 2 (S806). Here, the feedforward neural network to be learned is merely an example of an AI model, and the AI model is not limited to the feedforward neural network. As other methods, for example, gradient boosting, random forest, support vector machine, etc. may be used.

[0069] After the machine learning for the feedforward neural network is completed, the generation of the learned AI model 212 in FIG. 2 is completed, and the generation of the learned AI model 212 is completed (S807). As described above, based on the flowchart of FIG. 7, the learned AI model 212 is generated as preprocessing by the variable generation unit 202, the finite element model generation unit 204, the buckling analysis unit 206, the buckling strain calculation unit 208, and the learned model generation unit 210 in FIG. 2.

[0070] Next, after generating the learned AI model as preprocessing, a method for predicting the characteristics of the seismic isolation device using the generated learned AI model will be described. FIG. 8 is a flowchart of a part for predicting the characteristics of the seismic isolation device using the learned AI model among the flowchart of the seismic isolation device performance prediction method and the seismic isolation device performance prediction program. The operation of predicting the characteristics of the seismic isolation device using the learned AI model in the seismic isolation device performance prediction method and the seismic isolation device performance prediction program starts from the start of step S901 (S901).

[0071] First, in step S902, prepare in advance the LSS test data 211 for the lap shear sample (hereinafter referred to as "LSS" (Lap Shear Sample)) in the lab (experimental equipment), and input the prepared prediction target variable data into the learned AI model 212 in FIG. 2. Here, from the SS characteristics of the LSS of the input Ref rubber and New rubber, the material constants C 10 、C 30 、and the bulk modulus k are determined respectively (S902). Here, if the data of the Ref rubber is not required, the Ref rubber data may not be necessary. Furthermore, when analyzing the mechanical properties related to the tensile strength and breaking strength in the seismic isolation device, for the determination of this material constant, it may be determined based not only on the LSS test data 211 but also on a disk or the like. Here, the prediction target variable data 211 to be prepared in advance may be prepared by a spreadsheet or the like shown in FIGS. 9, 10, and 12 below. Also, for example, the breaking criteria value, which is the predetermined value in the case where breakage occurs when the principal strain reaches a predetermined value, can also be a material constant.

[0072] FIG. 9 is the measured value of the strength against shear of the LSS of the reference rubber (hereinafter referred to as "Ref rubber") in the lab, and it is a spreadsheet "Ref-LSS file" including the SS curve data with a force applied up to 400%, and the numerical values are merely examples. Also, FIG. 10 is the measured value of the strength against shear of the LSS when the material of the new rubber material newly considered for the above existing rubber product is changed, or when the shape of the seismic isolation device is changed from the conventional cylindrical structure to the flare structure which is a combination of the flare shape and the cylindrical shape shown in FIG. 3, etc., and it is a spreadsheet "New-LSS file" including the SS curve data with a force applied up to 400%, and the numerical values are merely examples. On the premise, it is assumed that the material models and material constants of the Ref rubber and New rubber in the actual product have been determined in advance from the test results. Figure 12 is an example of an SS curve corresponding to a Ref-LSS file or a New-LSS file.

[0073] Next, for the Yeoh model, which is a material model, the above-mentioned equation (1) is used, and its material constants are determined to match the results of the large deformation test of the product. Regarding the material constants and bulk modulus of elasticity of the Ref rubber in the actual product, they are respectively C 10 ref, product and C 30 ref, product and k ref, product Similarly, regarding the material constants and bulk modulus of elasticity of the New rubber in the actual product, they are respectively C 10 new, product and C 30 new, product and k new, product are used. Regarding the material constants and bulk modulus of elasticity of the Ref rubber in LSS, they are respectively C 10 ref, LSS and C 30 ref, LSS and k ref, LSS Similarly, regarding the material constants and bulk modulus of elasticity of the New rubber in LSS, they are respectively C 10 new, LSS and C 30 new, LSS and k new, LSS are used. That is, there is a gap in the material constants and bulk modulus of elasticity between the product and LSS, and it is assumed that their values are different between the two.

[0074] Here, based on the SS characteristics of the LSS of the input Ref rubber and New rubber, the C 10 and C 30 and k of the New rubber are determined respectively. The material constants C 10 and C30 When determining the bulk modulus k, it may be determined using an appropriate conversion formula.

[0075] Considering that there is a gap between the LSS and the material constants of the product, when there is such a gap, the material constant C of the New rubber material in the product 10 new, product 、C 30 new, product 、the bulk modulus k new, product is calculated using appropriate conversion formulas respectively.

[0076] Next, FIG. 12 is an example of a UPD size list file showing the size of the ultimate property diagram (UPD: Ultimate Property Diagram). Specifically, the input values are respectively the secondary shape coefficient S2, the value obtained by dividing the difference between the maximum diameter D2 and the minimum diameter D1 in the flare shape part of the flare structure by D1, i.e., (D2 - D1) / D1, the magnitude of the slope of the inclined part, represented by (D2 - D1) / (tr * n2), the reference surface pressure, and the rubber type No. representing the rubber type, but the present invention is not limited thereto. Here, tr is the rubber layer thickness, n2 is the number of laminated layers of the upper (or lower) inclined part, and "1" of the rubber type No. represents natural rubber type G4.

[0077] Also, S1, the inner - outer diameter ratio, etc. can also be input values, but currently, for simplicity, they are set as fixed values and average values. When the rubber shape is not a flare structure formed by a combination of a flare shape and a cylindrical shape but a conventional cylindrical structure, (D2 - D1) / D1 and (D2 - D1) / (tr * n2) are set to 0. In the example of FIG. 12, the Ref rubber is the above - mentioned natural rubber type G4, and for S2, there are 2 cases of 3.5 and 4, and the structure outputs a total of 4 cases including 2 cases of the conventional structure with a cylindrical shape and 2 cases of the flare structure formed by a combination of a flare shape and a cylindrical shape.

[0078] Returning to FIG. 8, in step S903, for each size described in the UPD size list file of FIG. 12, the UPD of the New rubber is predicted using the learned AI model 212 in FIG. 2 (S903). Then, in step S904, for each size described in the UPD size list file of FIG. 12, by outputting the numerical values of the predicted compression limit strength diagram of the New rubber in one or more spreadsheets, the characteristic prediction result data 213 in FIG. 2 can be obtained (S904). Here, the obtained characteristic prediction result data 213 may include buckling strain.

[0079] Then, in step S905, when the output of the numerical values of the compression limit strength diagram in one or more spreadsheets is completed, the prediction of the characteristic prediction result data 213 in FIG. 2 is completed, and the characteristic prediction result data 213 of the prediction target by the learned AI model 212 in FIG. 2 can be obtained (S905).

[0080] FIGS. 15 to 16 are compression limit strength diagrams corresponding to a total of four cases in the example of FIG. 12, where the Ref rubber is the natural rubber - based G4, there are two cases for S2 which are 3.5 and 4, and the structure has two cases: the conventional structure with a cylindrical shape and the flare structure which is a combination of the flare shape and the cylindrical shape. Specifically, FIG. 13 shows an example of the compression limit diagram corresponding to the first set (a) in FIG. 12, where the rubber type is N3, S2 = 3.5, and the conventional structure. Next, FIG. 14 shows an example of the compression limit diagram corresponding to the second set (b) in FIG. 12, where the rubber type is N3, S2 = 4.0, and the conventional structure. Then, FIG. 15 shows an example of the compression limit diagram corresponding to the third set (c) in FIG. 12, where the rubber type is N3, S2 = 3.5, and the flare structure. Furthermore, FIG. 16 shows an example of the compression limit diagram corresponding to the fourth set (d) in FIG. 12, where the rubber type is N3, S2 = 4.0, and the flare structure.

[0081] The performance prediction device for a seismic isolation device, the performance prediction method for a seismic isolation device, and the performance prediction program for a seismic isolation device have been described. However, the present invention is not limited to the above-described embodiments, and various design changes are possible without departing from the gist thereof. Needless to say, the present invention includes those implemented in various aspects within the scope of the gist of the present invention. Contribution to the Sustainable Development Goals (SDGs) led by the United Nations

[0082] The SDGs have been proposed for the realization of a sustainable society, and the present invention is considered to be a technology that can contribute to a town where people can continue to live.

Explanation of Signs

[0083] 1... Seismic isolation device, 2... Rubber body, 6... Connecting plate, 7... Assembly bolt hole, 10... Rigid flange, 11... Assembly elongated hole, 11a... End portion, 11b... Outer peripheral portion, 11b... Inner peripheral portion, 13... Mounting hole, 15... Assembly bolt, 16... Mounting bolt, 20A... Superstructure, 20B... Substructure 201... Performance prediction device for a seismic isolation device, 202... Variable generation unit, 203... Variable data, 204... Finite element model generation unit, 205... Finite element model, 206... Buckling analysis unit, 207... Buckling analysis result, 208... Buckling strain calculation unit, 209... Buckling strain, 210... Trained model generation unit, 211... LSS test data, 212... Trained AI model, 213... Characteristic prediction result data 301... AI model, 302... Node, 303... Input layer, 304... Intermediate layer, 305... Output layer

Claims

1. In a performance prediction device for a seismic isolation device using an elastic body, a variable generation unit that randomly generates variables including part or all of the shape, material constants, and loading conditions of the elastic body; a finite element model generation unit that generates a finite element model based on the generated variables; an analysis unit that performs analysis of the seismic isolation device using the generated finite element model; a calculation unit that calculates mechanical characteristics related to the seismic isolation device based on the results of the executed analysis; a learned model generation unit that generates a learned model for predicting the characteristics of the seismic isolation device based on the correlation when the generated variables are used as inputs and the calculated mechanical characteristics are used as outputs. A performance prediction device for a seismic isolation device having

2. The analysis of the seismic isolation device is buckling analysis, The performance prediction device for a seismic isolation device according to claim 1, wherein the mechanical characteristic is buckling strain.

3. The analysis of the seismic isolation device is fracture analysis, The performance prediction device for a seismic isolation device according to claim 1, wherein the mechanical characteristic is fracture strength.

4. The analysis of the seismic isolation device is tensile analysis, The performance prediction device for a seismic isolation device according to claim 1, wherein the mechanical characteristic is tensile strength.

5. The performance prediction device for the seismic isolation device further includes The performance prediction device for a seismic isolation device according to any one of claims 1 to 4, which predicts the characteristics of the seismic isolation device using the generated learned model.

6. In the performance prediction device for the seismic isolation device, The performance prediction device for a seismic isolation device according to claim 5, wherein the characteristic of the seismic isolation device is the buckling strain of the seismic isolation device.

7. In the performance prediction device for the seismic isolation device, The performance prediction device for a seismic isolation device according to claim 5, wherein the characteristic of the seismic isolation device is one or more compressive limit strength diagrams of the seismic isolation device.

8. The seismic isolation device has a rubber body with a predetermined shape. The performance prediction device for a seismic isolation device according to claim 1.

9. In the seismic isolation device, The performance prediction device for a seismic isolation device according to claim 8, wherein the predetermined shape is a cylindrical shape.

10. In the seismic isolation device, The performance prediction device for a seismic isolation device according to claim 8, wherein the predetermined shape is a shape formed by a combination of a flare shape and a cylindrical shape.

11. In a method for predicting the performance of a seismic isolation device using an elastic body, a computer randomly generates variables including part or all of the shape, material constants, and loading conditions of the elastic body, generates a finite element model based on the generated variables, Perform analysis of the seismic isolation device using the generated finite element model, calculate the mechanical properties of the seismic isolation device based on the results of the performed analysis, A performance prediction method for a seismic isolation device that generates a trained model for predicting the characteristics of the seismic isolation device based on the correlation relationship when the generated variables are input and the calculated mechanical properties are output.

12. The analysis of the seismic isolation device is a buckling analysis, The performance prediction method of the seismic isolation device according to claim 11, wherein the mechanical property is a buckling strain.

13. The analysis of the seismic isolation device is a fracture analysis, The performance prediction method of the seismic isolation device according to claim 11, wherein the mechanical property is a fracture strength.

14. The analysis of the seismic isolation device is a tensile analysis, The performance prediction method of the seismic isolation device according to claim 11, wherein the mechanical property is a tensile strength.

15. In a performance prediction program for a seismic isolation device using an elastic body, cause a computer to randomly generate variables including part or all of the shape, material constants, and loading conditions of the elastic body, generate a finite element model based on the generated variables, execute analysis of the seismic isolation device using the generated finite element model, calculate the mechanical properties of the seismic isolation device based on the results of the executed analysis, A performance prediction program for a seismic isolation device that generates a trained model for predicting the characteristics of the seismic isolation device based on the correlation relationship when the generated variables are input and the calculated mechanical properties are output.

16. The analysis of the seismic isolation device is a buckling analysis, The performance prediction program of the seismic isolation device according to claim 15, wherein the mechanical property is a buckling strain.

17. The analysis of the seismic isolation device is a fracture analysis, The performance prediction program of the seismic isolation device according to claim 15, wherein the mechanical property is a fracture strength.

18. The analysis of the seismic isolation device is a tensile analysis, The performance prediction program of the seismic isolation device according to claim 15, wherein the mechanical property is a tensile strength.

Images