Vibration waveform generating method and vibration waveform generating device

Abstract

To generate a plurality of simulated vibration waveforms having dispersion.SOLUTION: A vibration waveform generation method includes steps of: regressively analyzing the time history waveform of actual vibration on the basis of an analysis model that represents the amplitude characteristic of vibration and acquiring the amplitude spectrum of actual vibration as a target spectrum; generating a plurality of simulated vibration waveforms under a prescribed vibration occurrence condition by a trained model that has been trained by machine learning using the time history waveform of actual vibration and the vibration occurrence condition thereof; comparing the prescribed number of simulated vibration waveforms among the plurality of simulated vibration waveforms with the target spectrum in a prescribed frequency band to determine the degree of mutual fitness and selecting from the plurality of simulated vibration waveforms a combination of the prescribed number of simulated vibration waveforms to be outputted, on the basis of the degree of fitness.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a vibration waveform generating method and a vibration waveform generating device. [Background technology]

[0002] Various techniques for predicting and evaluating vibrations occurring in an object have been known (for example, Patent Document 1). Patent Document 1 discloses a method for generating an earthquake motion evaluation model, including: an acquisition step of acquiring, from a database storing a plurality of pieces of earthquake motion data in which earthquake motion characteristic parameters are associated with calculation results of earthquake motion indexes obtained when executing an earthquake motion simulation that calculates an earthquake motion index based on the earthquake motion characteristic parameters, a plurality of pieces of training data consisting of the feature quantities and the objective variables, with the earthquake motion characteristic parameters as feature quantities and the objective variables; and a generation step of generating an earthquake motion evaluation model as a trained model for machine learning by learning the correlation between the feature quantities and the objective variables based on the plurality of pieces of training data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-53155 Summary of the Invention [Problem to be solved by the invention]

[0004] The earthquake motion evaluation model in Patent Document 1 outputs a calculation result of earthquake motion indexes that includes at least one of the amplitude characteristics, period characteristics, and time characteristics of earthquake motion. In other words, this earthquake motion evaluation model outputs index values ​​that represent the characteristics of earthquake motion.

[0005] On the other hand, there is a demand for predicting and evaluating the waveform of vibrations acting on an object due to external forces under specific conditions. Even if the specific parameters of the external forces acting on the object are the same, the waveform of the vibrations actually occurring in the object will vary. Therefore, even when predicting the waveform of vibrations acting on an object, it is necessary to take this variation into account.

[0006] An object of the present invention is to generate a simulated vibration waveform that takes variations into consideration. [Means for solving the problem]

[0007] The present invention is a vibration waveform generation method for generating a simulated vibration waveform that can act on an object due to an external force generated under predetermined vibration generation conditions, and includes the steps of: performing regression analysis on the time history waveform of an actual vibration based on an analytical model that represents the amplitude characteristics of the vibration to obtain the amplitude spectrum of the actual vibration as a target spectrum; generating a plurality of simulated vibration waveforms under predetermined vibration generation conditions using a trained model constructed by machine learning the time history waveform of the actual vibration and the vibration generation conditions; and comparing a predetermined number of the plurality of simulated vibration waveforms with the target spectrum in a predetermined frequency band to determine their compatibility with each other, and selecting a combination of the predetermined number of simulated vibration waveforms to output from the plurality of simulated vibration waveforms based on the compatibility.

[0008] The present invention also provides a vibration waveform generating device that generates a simulated vibration waveform that can act on an object due to an external force generated under specified vibration generation conditions, and includes: a waveform generating unit that generates a plurality of simulated vibration waveforms under specified vibration generation conditions using a trained model in which the time history waveform of an actual vibration and the vibration generation conditions have been machine-learned; a target spectrum acquisition unit that performs regression analysis of the actual vibration based on an analytical model that represents the amplitude characteristics of the vibration to acquire the amplitude spectrum of the actual vibration as a target spectrum; and a waveform selection unit that compares a predetermined number of the plurality of simulated vibration waveforms with the target spectrum in a predetermined frequency band to determine their compatibility with each other, and selects a combination of the predetermined number of simulated vibration waveforms to output based on the compatibility. [Effects of the Invention]

[0009] According to the present invention, a plurality of simulated vibration waveforms with variations are generated. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a block diagram showing a configuration of a vibration waveform generating device according to an embodiment of the present invention. [Figure 2] FIG. 4 is a diagram showing an example of actual waveform data according to the embodiment of the present invention. [Figure 3] FIG. 1 is a flowchart illustrating a waveform generating method according to an embodiment of the present invention. [Figure 4] FIG. 10 is a flowchart illustrating a method for selecting an output waveform group in the vibration waveform generating method according to the embodiment of the present invention. [Figure 5] 1 is a graph showing a target spectrum and an average value of a group of simulated vibration waveforms according to an embodiment of the present invention, in which the vertical axis represents amplitude and the horizontal axis represents frequency. [Figure 6] 1 is a graph showing the standard deviation of a target spectrum and the standard deviation of a group of simulated vibration waveforms in an embodiment of the present invention, where the vertical axis represents the standard deviation and the horizontal axis represents the frequency. DETAILED DESCRIPTION OF THE INVENTION

[0011] A vibration waveform generating method and a vibration waveform generating device 100 according to an embodiment of the present invention will be described below with reference to the drawings.

[0012] This embodiment predicts and estimates vibrations that may act on an object due to external forces generated under predetermined vibration generation conditions, and generates a predetermined number of simulated vibration waveforms with variations. Below, an example will be described in which simulated vibration waveforms are generated for vibrations that act on an object, such as a structure on the ground, due to the force of earthquake motion (seismic force) as an external force.

[0013] First, the vibration waveform generating device 100 will be described with reference to FIG.

[0014] The vibration waveform generation device 100 is configured by a computer including a CPU (Central Processing Unit) that executes a control program and the like, a ROM (Read-Only Memory) that stores the control program executed by the CPU, a RAM (Random Access Memory) that stores the results of CPU calculations and the like, a communication device, etc. The vibration waveform generation device 100 performs various functions of the vibration waveform generation device 100 described in this specification by loading the control program stored in the ROM into the RAM and executing it on the RAM by the CPU. The vibration waveform generation device 100 may be configured by a single computer, or may be configured by multiple microcomputers and configured to perform each control in a distributed manner among the multiple computers.

[0015] As shown in FIG. 1, the vibration waveform generating device 100 has a memory unit 10 that stores a control program and various data, and a processing unit 20 that executes the control program stored in the memory unit 10 and generates a simulated vibration waveform.

[0016] The storage unit 10 stores time history waveform data of seismic motion (actual vibration) that actually occurred at an observation point (hereinafter simply referred to as "actual waveform data") in association with the vibration occurrence conditions. The stored data is used as a data set for machine learning, which will be described later. The actual waveform data may be the observed data of the time history waveform at the observation point itself, or may be an image obtained by converting the observed data into a spectrogram.

[0017] The plurality of actual waveform data are stored with a uniform sampling length and sampling interval from the start of sampling at the time of the earthquake occurrence to the end of sampling.

[0018] Furthermore, even if the vibration generating conditions are the same, the maximum amplitude and other values ​​of earthquake motion may differ. For this reason, the actual waveform data is stored in the storage unit 10 as waveforms in which the amplitude of each waveform is divided by the maximum amplitude, and the amplitude is normalized within the range of -1 to +1, as shown in Figure 2. The amplitude may be expressed by displacement, velocity, or acceleration.

[0019] In this embodiment, at least the magnitude of the earthquake and the distance between the observation point and the epicenter are set as vibration generation conditions. A plurality of condition values ​​are set for each of the magnitude and distance, thereby setting a plurality of combinations of magnitude and distance as vibration generation conditions. The memory unit 10 stores a plurality of actual waveform data corresponding to each vibration generation condition.

[0020] Furthermore, if the stored actual waveform data is data observed at different observation points, it is desirable to include the site characteristics of the ground at the observation points in the vibration generation conditions. By including the ground characteristics in the vibration generation conditions, the generated simulated vibration waveform can be adapted to the actual vibration waveform with greater accuracy.

[0021] 1, the processing unit 20 includes a waveform generation unit 21 that generates multiple simulated vibration waveforms under input vibration generation conditions using a trained model, a target spectrum acquisition unit 22 that performs regression analysis on actual vibrations that have actually occurred to acquire the amplitude spectrum of the actual vibration as a target spectrum, and a waveform selection unit 23 that selects a combination of a predetermined number of simulated vibration waveforms that match the target spectrum from the multiple simulated vibration waveforms. Note that the components of the processing unit 20 shown in Fig. 1 are shown as virtual units that represent the functions of the processing unit 20, and do not necessarily mean that they physically exist.

[0022] The waveform generation unit 21 generates a plurality of simulated vibration waveforms that can be generated under predetermined vibration generation conditions. The waveform generation unit 21 is configured by a trained model that is machine-learned from the actual waveform data stored in the storage unit 10 and the vibration generation conditions.

[0023] The trained model is a conditional generative adversarial network (CGAN) consisting of two networks: a generator and a discriminator. The generator learns real waveform data as training data, using the corresponding vibration generation conditions as labels. When it receives input data including the vibration generation conditions and random numbers, it generates waveform data having the same structure as the real waveform data corresponding to the vibration generation conditions. The discriminator receives the real waveform data, which is training data, and waveform data generated by the generator, and trains to be able to distinguish between the real waveform data and the generated waveform data. Meanwhile, the generator trains to generate waveform data that the discriminator cannot distinguish from real waveform data. By training in this manner, the trained model generates waveform data of simulated vibration waveforms similar to real waveform data under specified vibration generation conditions.

[0024] The target spectrum acquisition unit 22 performs regression analysis on each of a plurality of pieces of actual waveform data generated under the same vibration generating condition based on an analytical model that represents the amplitude characteristics of the vibration to acquire an amplitude spectrum. The acquired amplitude spectrum is set as the target spectrum (see FIG. 6) for the vibration generating condition.

[0025] The waveform selection unit 23 selects a predetermined number of simulated vibration waveforms to be output from the plurality of simulated vibration waveforms generated by the waveform generation unit 21. Specifically, the waveform selection unit 23 selects, from the plurality of simulated vibration waveforms, a combination of a predetermined number of simulated vibration waveforms having a representative amplitude spectrum whose conformity to the target spectrum is equal to or less than a predetermined tolerance. The representative amplitude spectrum is one amplitude spectrum obtained based on a predetermined number of simulated vibration waveforms extracted from the plurality of simulated vibration waveforms. In this embodiment, the representative amplitude spectrum is an amplitude spectrum obtained from the average value of the predetermined number of simulated vibration waveforms.

[0026] Next, the simulated vibration waveform generating method of this embodiment will be described in detail.

[0027] When predetermined vibration generation conditions (hereinafter referred to as "input conditions") are given as input, the vibration waveform generating device 100 executes the process shown in FIG. 3 and outputs a predetermined number of simulated vibration waveforms that may be generated by external forces under the input conditions.

[0028] First, in step S10, a plurality of simulated vibration waveforms under the input conditions are generated by the trained model of the waveform generation unit 21. The trained model is trained based on actual waveform data whose amplitudes have been normalized. Therefore, in step S20, the trained model generates a plurality of waveforms (reference simulated waveforms) whose amplitudes have been normalized between -1 and +1. In the following, it is assumed that N reference simulated waveforms are generated. N is a natural number equal to or greater than 3.

[0029] Next, in step S11, the amplitude spectrum of the actual vibration is acquired as a target spectrum from the plurality of actual waveform data under the input conditions. Specifically, the target spectrum is obtained by performing regression analysis on the waveform of the actual vibration based on a regression model that represents the amplitude characteristics.

[0030] In this embodiment, a spectrum inversion analysis expressed by the following equation is performed as a regression model on the waveform of an actual vibration to obtain an amplitude spectrum: Each parameter is set based on input conditions and empirical rules.

number

[0031] The method of obtaining an amplitude spectrum by spectral inversion analysis is well known, as described in, for example, "Estimation of heterogeneous attenuation structure, source characteristics, and site amplification characteristics focusing on domain division of seismic motion propagation path characteristics" (Architectural Institute of Japan, Journal of Structural Engineering, Vol. 84, No. 756, February 2019), and therefore a detailed explanation will be omitted.

[0032] After obtaining an amplitude spectrum for each real waveform data by spectral inversion analysis, the mean and standard deviation (variation index value) of each amplitude spectrum are calculated for each frequency. The calculated mean of the amplitude spectra is set as the target spectrum. The standard deviation of the target spectrum is the standard deviation (regression error) of the amplitudes at each frequency of the multiple real waveform data.

[0033] Next, in step S12, a predetermined number of simulated vibration waveforms (hereinafter referred to as a "simulated vibration waveform group") having a representative amplitude spectrum that matches the target spectrum acquired in step S11 are selected as an output waveform group from the plurality of simulated vibration waveforms generated in step S10. In step S12, the process shown in Fig. 4 is executed. The process of selecting the output waveform group executed in step S12 will be described below with reference to Fig. 4.

[0034] First, in step S20, a predetermined number of simulated vibration waveforms are extracted from the plurality of simulated vibration waveforms (reference simulated waveforms) generated in step S10 by any combination. Hereinafter, it is assumed that n simulated vibration waveforms are extracted as the predetermined number, where n is a natural number equal to or greater than 2 and smaller than N (N>n≧2).

[0035] In the next step S21, in order to scale the normalized amplitude to the amplitude of the actual vibration, a correction coefficient is calculated for each of the simulated vibration waveforms extracted in step S20 by which to multiply the amplitude of the simulated vibration waveform. Specifically, if the waveform obtained by multiplying the simulated vibration waveform, which is the reference simulated waveform, by the correction coefficient is taken as the correction waveform, the correction waveform of the simulated vibration waveform group is compared with the target spectrum, and a correction coefficient that minimizes the degree of fit between the two is calculated individually for each of the extracted simulated vibration waveforms.

[0036] The degree of fit is the sum of the residuals for each frequency between the simulated vibration waveform group (corrected waveform) and the target spectrum, and in this embodiment, is expressed as the sum of the sum of the residuals between the average value of the spectrum of the simulated vibration waveform group and the target spectrum, and the sum of the residuals between the standard deviation of the simulated vibration waveform group and the standard deviation of the target spectrum. In other words, in this embodiment, the smaller the degree of fit, the more compatible the simulated vibration waveform group and the target spectrum are. Furthermore, the average value of the spectrum of the simulated vibration waveform group is obtained by spectrally converting each waveform in the simulated vibration waveform group and averaging (geometric mean) the spectra of the waveforms for each frequency.

[0037] More specifically, the conformance of the simulated vibration waveform group to the target spectrum is calculated while updating the correction coefficient for each simulated vibration waveform at predetermined intervals within a predetermined numerical range between the initial value and the upper limit value. The correction coefficient is updated independently for each simulated vibration waveform. Then, while independently updating the correction coefficient for each simulated vibration waveform, the conformance as a simulated vibration waveform group is minimized, that is, the correction coefficient for each simulated vibration waveform that provides a combination in which the average of the corrected waveforms of the simulated vibration waveforms is most conformable to the target spectrum, is the correction coefficient used for the reference simulated waveform. In this way, the conformance is calculated while independently updating the correction coefficients to be multiplied by the simulated vibration waveform, which is the reference simulated waveform, within a predetermined range, and the conformance of each simulated vibration waveform that minimizes the conformance as a simulated vibration waveform group (in other words, the combination of correction coefficients for the simulated vibration waveforms that minimizes the conformance) is obtained. The initial value and upper limit value of the correction coefficient are set in advance based on empirical rules, etc.

[0038] The comparison between the simulated vibration waveform group and the target spectrum is performed within a predetermined frequency range (fitting range). The fitting range is set taking into consideration noise in the actual waveform data. For example, as shown in FIG. 5, when errors due to noise occur in a relatively low frequency range, the fitting range is set to a high frequency range. Note that the fitting range is not limited to a certain frequency range, and may be set to the entire frequency range in which data is recorded.

[0039] Next, as shown in FIG. 4, in step S22, it is determined whether the minimum degree of conformance calculated in step S21 (the degree of conformance calculated based on the adopted correction coefficient) is equal to or less than a predetermined tolerance. If the degree of conformance is not equal to or less than the tolerance, the process proceeds to step S23, where a predetermined number of combinations of simulated vibration waveforms extracted from the plurality of simulated vibration waveforms are updated (any simulated vibration waveform group is re-extracted), and step S21 is executed again. In this way, in step S22, step S21 is executed while updating the simulated vibration waveforms to be extracted until the degree of conformance becomes equal to or less than the tolerance, and a simulated vibration waveform group whose degree of conformance is equal to or less than the tolerance is extracted. If the degree of conformance is equal to or less than the tolerance in step S22, that simulated vibration waveform group is selected as an output waveform group in step S24. The tolerance of the degree of conformance is set depending on the degree of variation in the simulated vibration waveform group to be output, etc.

[0040] In this way, once the process of selecting the output waveform group in step S12 (FIG. 3) (the process shown in FIG. 4) has been performed, in step S13 (FIG. 3), the output waveform group selected in step S12 is output to the outside.

[0041] As a result, a predetermined number of simulated vibration waveforms (a group of simulated vibration waveforms) having a certain degree of variation are generated that match the target spectrum obtained from the actual waveform data.

[0042] According to the above embodiment, the following advantageous effects are achieved.

[0043] In this embodiment, a predetermined number of simulated vibration waveforms are selected from the plurality of simulated vibration waveforms generated by the trained model so that the average amplitude spectrum of the predetermined number of simulated vibration waveforms matches a target spectrum obtained by regression analysis of actual vibrations using an existing analysis model. Therefore, it is possible to generate a plurality of simulated vibration waveforms that match the target spectrum with a certain degree of variation.

[0044] In this embodiment, the compatibility of the simulated vibration waveform with respect to the target spectrum is the sum of the residual between the target spectrum and the average of the simulated vibration waveform and the residual between the target spectrum and the standard deviation of the simulated vibration waveform. In this way, the compatibility is set based on the standard deviation in addition to the residual of the spectrum, so that a simulated seismic intensity waveform that is more compatible with the target spectrum can be selected.

[0045] In this embodiment, machine learning of a trained model is performed using a standardized waveform, and a simulated vibration waveform is generated by correcting the reference waveform generated by the trained model. Specifically, a correction coefficient is changed within a predetermined numerical range at predetermined numerical intervals and multiplied by the amplitude of the reference simulated waveform to obtain a corrected waveform. A correction coefficient is calculated for each simulated vibration waveform so that the degree of match between the corrected waveform of the simulated vibration waveform and the target spectrum is minimized, and the corrected waveform using the correction coefficient is selected as the simulated vibration waveform. In this way, by using a standardized waveform, the accuracy of machine learning can be improved by eliminating the influence of the amplitude, even when differences in amplitude occur under the same vibration generation conditions.

[0046] Although the present embodiment has been described above, the following modifications are also within the scope of the present invention. Furthermore, it is also possible to combine the configuration shown in the modifications with the configuration described in the above embodiment, or to combine the configurations described in the following different modifications.

[0047] In the above embodiment, a target spectrum is acquired from real waveform data by spectral inversion analysis expressed by equation (1). However, the analysis method (analysis model) for acquiring a target spectrum from real waveform data is not limited to that in the above embodiment. At least, an amplitude spectrum can be acquired by performing regression analysis on the real waveform data. For example, block inversion analysis may be used as the same inversion analysis.

[0048] In the above embodiment, the goodness of fit is the sum of the residual between the target spectrum and the average of the simulated vibration waveform and the residual between the target spectrum and the standard deviation of the simulated vibration waveform. In other words, in the above embodiment, the spectral residual and the standard deviation residual are weighted the same. Alternatively, the spectral residual and the standard deviation residual may be weighted to determine the goodness of fit. Furthermore, the goodness of fit is not limited to being based on the spectral residual and the standard deviation residual. For example, the goodness of fit may be based only on the spectral residual, such as the sum of the residuals between the target spectrum and the average of the simulated vibration waveform or the sum of squared residuals, without using the standard deviation.

[0049] Furthermore, in the above embodiment, the standard deviation is used as the variation index value, but other index values ​​such as variance may be used as long as they are index values ​​that represent the degree of variation in the target spectrum or simulated vibration waveform.

[0050] Furthermore, in the above embodiment, machine learning was performed using waveform data of actual vibrations whose amplitudes were normalized in the range of −1 to +1. In contrast, the data sets used in machine learning are not limited to those normalized in the range of −1 to +1, and may be normalized or standardized data, or data that has not been normalized or standardized may be used.

[0051] In addition, in the above embodiment, when a predetermined number of simulated vibration waveforms whose compatibility is equal to or less than a tolerance value are extracted while updating a predetermined number of simulated vibration waveforms extracted from a plurality of simulated vibration waveforms, the predetermined number of simulated vibration waveforms are selected as an output waveform group. This eliminates the need to calculate compatibility for all combinations of extracted N to n waves, making it possible to output a simulated vibration waveform group while suppressing the processing load. In contrast, when selecting a predetermined number (n waves) of simulated vibration waveforms from a plurality (N waves) of simulated vibration waveforms, compatibility may be calculated for all combinations of extracted n waves, and the simulated vibration waveform group with the combination having the smallest compatibility may be selected as the output waveform group. Furthermore, it is not necessary to determine the degree of compatibility for all combinations. For example, as disclosed in a publicly known technique (Baker, JW, AND Lee, C. (2018). An Improved Algorithm for Selecting Ground Motions to Match a Conditional Spectrum. Journal of Earthquake Engineering, 22(4), 708-723.), a predetermined number of simulated vibration waveforms that match the target spectrum may be selected from multiple simulated vibration waveforms using a calculation algorithm made more efficient by a statistical method.

[0052] Furthermore, if the compatibility is calculated for all combinations of extracting n waves from N waves, but no combination of simulated vibration waveforms of n waves with a compatibility below the allowable value is found, the vibration waveform generating device 100 can be configured to output error information and prompt the user to reset the compatibility, for example.

[0053] Furthermore, the above-described embodiment generates and outputs a plurality of simulated vibration waveforms (a group of simulated vibration waveforms) with variations. However, the vibration waveform generation device 100 and the simulated vibration waveform generation method using the same may be used to generate and output a single simulated vibration waveform. In other words, the number of waves to be output may be set to one (n=1).

[0054] In the above embodiment, a CGAN is used as the trained model. The trained model is not limited to a CGAN, and any other generative model that outputs an image or a waveform can also be used.

[0055] In the above embodiment, a case has been described in which a simulated vibration waveform of vibration acting on an object due to an earthquake force as an external force is generated. However, the present invention can also be applied to generating a simulated vibration waveform of vibration acting on an object due to, for example, wind force or wave force as an external force. Furthermore, the external force is not limited to kinetic energy such as earthquake motion, wind force, or wave force, and can also be applied to generating a simulated vibration waveform of vibration acting on an object due to, for example, temperature or humidity fluctuations (changes).

[0056] Although the embodiments of the present invention have been described above, the above embodiments merely illustrate some of the application examples of the present invention, and it is not intended that the technical scope of the present invention be limited to the specific configurations of the above embodiments. [Explanation of symbols]

[0057] 100 Vibration waveform generator 21 Waveform generator 22 Target spectrum acquisition unit 23 Waveform selection section

Claims

1. A vibration waveform generation method for generating a simulated vibration waveform that may act on an object due to an external force generated under predetermined vibration generation conditions, comprising: performing a regression analysis on a time history waveform of an actual vibration based on an analytical model that represents amplitude characteristics of the vibration to obtain an amplitude spectrum of the actual vibration as a target spectrum; generating a plurality of simulated vibration waveforms under predetermined vibration generation conditions using a trained model constructed by machine learning the time history waveform of the actual vibration and the vibration generation conditions; a step of comparing a predetermined number of the simulated vibration waveforms among the plurality of simulated vibration waveforms with the target spectrum in a predetermined frequency band to determine their mutual compatibility, and selecting a combination of the predetermined number of the simulated vibration waveforms to be output from the plurality of simulated vibration waveforms based on the compatibility, Vibration waveform generation method.

2. 2. The vibration waveform generating method according to claim 1, the degree of conformance is set based on a residual between an average value of amplitudes of the predetermined number of the simulated vibration waveforms and the target spectrum. Vibration waveform generation method.

3. 2. The vibration waveform generating method according to claim 1, the goodness of fit is set based on a residual between an average value of amplitudes of the predetermined number of the simulated vibration waveforms and the target spectrum, and a residual between a variation index value representing a variation in amplitude spectra of the predetermined number of the simulated vibration waveforms and the variation index value of the target spectrum. Vibration waveform generation method.

4. 4. The vibration waveform generating method according to claim 1, The time history waveform of the actual vibration that is machine-learned is a waveform in which the amplitude is normalized within a predetermined range, In the step of generating a plurality of simulated vibration waveforms, a reference simulated waveform having an amplitude standardized by the trained model is output, In the step of selecting the simulated vibration waveform, a correction coefficient is changed within a predetermined numerical range at predetermined numerical intervals and multiplied by the amplitude of the reference simulated waveform to obtain a correction waveform, the correction coefficient for each of the simulated vibration waveforms is calculated so that the degree of fit between the correction waveform of the simulated vibration waveform and the target spectrum is minimized, and the correction waveform based on the correction coefficient is selected as the simulated vibration waveform. Vibration waveform generation method.

5. A vibration waveform generating device that generates a simulated vibration waveform that can act on an object due to an external force generated under predetermined vibration generating conditions, a waveform generating unit that generates a plurality of simulated vibration waveforms under predetermined vibration generation conditions using a trained model that has been machine-learned based on a time history waveform of actual vibration and the vibration generation conditions; a target spectrum acquisition unit that performs regression analysis on the actual vibration based on an analysis model that represents amplitude characteristics of the vibration to acquire an amplitude spectrum of the actual vibration as a target spectrum; a waveform selection unit that compares a predetermined number of the simulated vibration waveforms among the plurality of simulated vibration waveforms with the target spectrum in a predetermined frequency band to determine their mutual compatibility, and selects a combination of the predetermined number of the simulated vibration waveforms to be output based on the compatibility; Vibration waveform generator.

Images