Method for evaluating simulation degree of human body model based on biomechanical limit value curve

Abstract

To provide a method for improving the accuracy of evaluation to a simulation degree of a human body model.SOLUTION: A method includes: acquiring body type characteristic information of a human body model; constructing a self-adaptive dimensionality reduction neural network and a regression neural network; verifying the accuracy of the self-adaptive dimensionality reduction neural network and the regression neural network; acquiring simulation biomechanical response low-dimensional data by inputting the body type characteristic information of the human body model in the regression neural network, simulation biomechanical response high-dimensional data by inputting the simulation biomechanical response low-dimensional data in an intermediate layer of the self-adaptive dimensionality reduction neural network to expand dimensionality, and biomechanical simulation limit value curve of the human body model; and evaluating a simulation degree of the human body model on the basis of the biomechanical simulation limit value curve.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to the technical field of crash simulation, and in particular to a method for evaluating the degree of simulation of a human body model based on a limit curve of biomechanical simulation. [Background technology]

[0002] In the prior art, a human body model is generally used to evaluate the injury to the human body during the collision process, and further evaluate the collision safety performance of the vehicle. As can be seen, the biomechanical simulation degree of the human body model directly affects the accuracy of the test. Summary of the Invention [Problem to be solved by the invention]

[0003] The conventional method for evaluating the biomechanical simulation degree of a human body model is to obtain a real biomechanical response signal through a crash test of a cadaver, an animal, or a volunteer, calculate the scaling ratio of the human body model to the body type information of the tester such as a cadaver, an animal, or a volunteer, and finally scale the human body model according to the real biomechanical response signal to obtain the limit value data of the simulation of the human body model, and evaluate the biomechanical response simulation degree of the human body model being studied. This method has problems such as large restrictions and the accuracy of the scaling method is not high. The main reasons are as follows: the body type information of the tester such as a cadaver, an animal, or a volunteer is limited, the obtained biomechanical response data has restrictions and does not have universality and representativeness; at the same time, the scaling method is used to make the biomechanical response and the body type change of the tester have a linear correlation, but the impact of the body type difference of the tester on biomechanics is quite complex; the method of only adopting linear scaling to obtain the limit value of the simulation of the biomechanical response is not highly accurate, and the accuracy of the evaluation of the simulation degree of the human body model is not high.

[0004] Therefore, there is an urgent need for a method for evaluating the simulation degree of a human body model based on the limit curve of biomechanical simulation. The present invention improves the accuracy of the limit curve of biomechanical simulation and improves the accuracy of evaluation of the simulation degree of a human body model. [Means for solving the problem]

[0005] In order to solve the above technical problems, the present invention provides a method for evaluating the simulation degree of a human body model based on the limit value curve of biomechanical simulation, thereby improving the accuracy of the limit value of biomechanical simulation and improving the accuracy of the simulation degree evaluation of the human body model.

[0006] The present invention provides a method for evaluating the simulation degree of a human body model based on a limit curve of biomechanical simulation, the evaluation method comprising the steps of: In S1, body shape characteristic information of a human body model is obtained, and the body shape characteristic information includes height, weight, distance from the top of the head to the pubic symphysis, shoulder width, arm length, and pelvic width; In S2, a self-adaptive dimensionality reduction neural network is constructed, the self-adaptive dimensionality reduction neural network includes a dimensionality reduction network, an intermediate layer and a dimensionality expansion network, and the dimensionality reduction network and the dimensionality expansion network are used as separate networks respectively; In S3, a recurrent neural network is constructed, and the number of neurons in each layer of the recurrent neural network is equal to the number of body shape feature information; In S4, verify the accuracy of the self-adaptive dimension reduction neural network and the regression neural network. If the accuracy meets the requirement, proceed to S5. If the accuracy does not meet the requirement, optimize the self-adaptive dimension reduction neural network and the regression neural network until the accuracy of the self-adaptive dimension reduction neural network and the regression neural network meets the requirement. In S5, the body shape feature information of the human body model is input into the regression neural network to obtain simulated biomechanical response low-dimensional data, and the simulated biomechanical response low-dimensional data is input into the intermediate layer of the self-adaptive dimension reduction neural network, and then the dimension expansion network is used to perform dimension expansion processing on the simulated biomechanical response low-dimensional data to obtain simulated biomechanical response high-dimensional data, and a limit value curve of the biomechanical simulation of the human body model is obtained according to the simulated biomechanical response high-dimensional data; In S6, the simulation degree of the human body model is evaluated based on the limit curve of the biomechanical simulation. Effect of the Invention

[0007] The embodiments of the present invention have the following technical effects. A functional relationship between the body shape feature information and the biomechanical response low-dimensional data is established by a regression neural network, high-dimensional biomechanical response data for the biomechanical response low-dimensional data is obtained by a self-adaptive dimensionality reduction neural network, a corresponding relationship between the biomechanical response data and the biomechanical response curve is established, a biomechanical simulation limit value curve of a human body model is generated by the regression neural network and the self-adaptive dimensionality reduction neural network, and the simulation degree of the human body model is evaluated based on the biomechanical simulation limit value curve, thereby generating a corresponding biomechanical simulation limit value curve for a human body model with any known body shape feature information, improving the accuracy of the biomechanical simulation limit value, and improving the accuracy of the evaluation of the simulation degree of the human body model. [Brief description of the drawings]

[0008] In order to more clearly describe the specific embodiments of the present invention or the technical solutions to the prior art, the following provides a brief description of the drawings that need to be used to describe the specific embodiments. The drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings based on these drawings without requiring any creative efforts. [Figure 1] 1 is a flowchart of a method for evaluating the simulation degree of a human body model based on a limit curve of a biomechanical simulation according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a structural schematic diagram of a self-adaptive dimension-reduced neural network in a method for enhancing the simulation degree of a human body model based on limit polarity of biomechanical simulation according to an embodiment of the present invention; [Diagram 3] FIG. 2 is a structural schematic diagram of a recurrent neural network in a method for enhancing the simulation degree of a human body model based on limit polarity of biomechanical simulation according to an embodiment of the present invention; [Figure 4] FIG. 2 is a structural schematic diagram of an inverse regression neural network in a method for enhancing the simulation degree of a human body model based on limit polarity of biomechanical simulation according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] In order to make the objectives, technical solutions and advantages of the present invention clearer, a clear and complete description of the technical solutions of the present invention is provided below. The embodiments described below are only some of the embodiments of the present invention, and are not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative labor are all within the scope of protection of the present invention.

[0010] 1 is a flowchart of a method for evaluating the simulation degree of a human body model based on a limit curve of biomechanical simulation provided by an embodiment of the present invention. Referring to FIG. 1, the method specifically includes the following steps:

[0011] In S1, body type characteristic information of a human body model is obtained, and the body type characteristic information includes height, weight, distance from the top of the head to the pubic symphysis, shoulder width, arm length, and pelvic width.

[0012] Specifically, based on the needs of research and development of human body models, p pieces of body type feature information of the human body model are obtained, and the body type feature information includes, but is not limited to, height, weight, distance from the top of the head to the pubic symphysis, shoulder width, arm length, pelvic width, etc. Since the body type feature information of the human body model is easy to obtain, the body type feature information is used as a training parameter for the neural network, and the relationship between the body type feature information and the biomechanical response data is obtained.

[0013] In S2, a self-adaptive dimensionality reduction neural network is constructed, which includes a dimensionality reduction network, a hidden layer and a dimensionality expansion network, and the dimensionality reduction network and the dimensionality expansion network are used as independent networks respectively.

[0014] Specifically, FIG. 2 is a structural schematic diagram of a self-adaptive dimension reduction neural network provided by an embodiment of the present invention. Referring to FIG. 2, the self-adaptive dimension reduction neural network includes a dimension reduction network, an intermediate layer and a dimension expansion network, and the dimension reduction network, the intermediate layer and the dimension expansion network are all composed of one or more layers of neural networks, and each layer of neural networks includes weight values, biases and multiple neurons, the dimension reduction network further includes an input layer, and the dimension expansion network further includes an output layer, and the dimension reduction network and the dimension expansion network are symmetrical with respect to the intermediate layer, and the dimension reduction network and the dimension expansion network of the self-adaptive dimension reduction neural network can be used as independent networks respectively. The number of layers from the input layer to the intermediate layer, the number of layers from the intermediate layer to the output layer, and the number of neurons in each layer can be adjusted according to the loss function value of the training of the neural network, the number of neurons in the intermediate layer of the self-adaptive dimension reduction neural network is equal to the number of body shape feature information, and the number of neurons in the input layer of the dimension reduction network and the output layer of the dimension expansion network is equal to the number of force value points collected from the biomechanical response curve obtained in the collision test, thereby reducing the degree of calculation complexity and facilitating the subsequent verification process. However, the more force value points collected in the biomechanical response curve, the more accurately the biomechanical response curve can be displayed, but in this case, the more the number of force points increases, the more complicated the calculation becomes, so the number of force value points collected can be set according to the requirements of accuracy or the requirements of the degree of calculation complexity.

[0015] Specifically, for the i-th neuron in the l-th layer in the dimensionality reduction network, the calculation formula for the value of the i-th neuron in the l-th layer is as follows:

[0016] [Number 1] JPEG2025036074000002.jpg24170

[0017] or [Number 2] JPEG2025036074000003.jpg29170

[0018] JPEG2025036074000004.jpg42170

[0019] Equation (2) can be written in matrix form as follows:

[0020] [Number 3] JPEG2025036074000005.jpg14170

[0021] JPEG2025036074000006.jpg39170

[0022] Specifically, for the i-th neuron in the l-th layer in the dimension expansion network, the calculation formula for the value of the i-th neuron in the l-th layer is as follows:

[0023] [Number 4] JPEG2025036074000007.jpg24170

[0024] or [Number 5] JPEG2025036074000008.jpg42170

[0025] JPEG2025036074000009.jpg40170

[0026] Equation (5) can be written in matrix form as follows:

[0027] [Number 6] JPEG2025036074000010.jpg17170

[0028] JPEG2025036074000011.jpg41170

[0029] The activation function of the self-adaptive dimensionality-reduced neural network can be linear or nonlinear. By changing the nature of the activation function of the self-adaptive dimensionality-reduced neural network, the dimensionality reduction process can be performed on the biomechanical response data to obtain linear and nonlinear biomechanical response low-dimensional data.

[0030] Furthermore, after completing the construction of the self-adaptive dimensionality reduction neural network, a restriction method is added to the dimensionality reduction network and dimensionality expansion network of the self-adaptive dimensionality reduction neural network, and the restriction method includes weight value association, weight value matrix orthogonal, unrelated features and unitization.

[0031] Specifically, trainable parameters, such as weights and biases, are set between layers of the dimension reduction network and the dimension expansion network, and the weights are associated with each other so that the weight matrix of the dimension reduction network and the weight matrix of the dimension expansion network have a corresponding relationship. The weights and biases in the dimension reduction network are updated by training the self-adaptive dimension reduction neural network, and the weights and biases in the dimension expansion network are directly obtained by converting the weights and biases of the dimension reduction network through a formula. The weights are associated with each other so that the weights of the dimension reduction network and the weights of the dimension expansion network have the following relationship:

[0032] [Number 7] JPEG2025036074000012.jpg16170

[0033] [Number 8] JPEG2025036074000013.jpg17167

[0034] Specifically, the orthogonal restriction of the weight value matrix can make the weight value matrix of each layer in the self-adaptive dimension reduction neural network orthogonal to each other. The irrelevant feature restriction using the L1 regularization method can restrict the off-diagonal data of the covariance matrix of the output data of the dimension reduction network to approach 0. The unitization restriction can make the sum of squares of the weight values ​​of each layer in the dimension reduction network unit 1. Adding restrictions to the self-adaptive dimension reduction neural network reduces the degree of calculation complexity, removes the correlation between data, improves the independence of data, and is favorable to the training of the neural network and obtains higher accuracy.

[0035] In S3, a recurrent neural network is constructed, and the number of neurons in each layer of the recurrent neural network is the same as the number of body shape feature information.

[0036] Specifically, Fig. 3 is a structural schematic diagram of a regression neural network provided by an embodiment of the present invention. Referring to Fig. 3, the regression neural network includes an input layer, an output layer and a hidden layer, and the structure of the regression neural network is a bobbin network, and the number of neurons in each layer of the regression neural network is the same as the number of body shape feature information. For the i-th neuron in the l-th layer in the regression neural network, the calculation formula for the value of the i-th neuron in the l-th layer is as follows:

[0037] [Number 9] JPEG2025036074000014.jpg25170

[0038] or [Number 10] JPEG2025036074000015.jpg25170

[0039] JPEG2025036074000016.jpg53170

[0040] Equation (10) can be written in matrix form as follows:

[0041] [Number 11] JPEG2025036074000017.jpg23169

[0042] JPEG2025036074000018.jpg31170

[0043] Furthermore, constructing the recurrent neural network includes training the recurrent neural network, and the method for training the recurrent neural network includes the following steps.

[0044] In S31, physical characteristic information of the subject is acquired.

[0045] Specifically, the subjects may include cadavers, animals, volunteers, etc. According to the requirements of the research and development of the human body model, p pieces of body shape characteristic information of h pieces of subjects are recorded.

[0046] In S32, a biomechanical response curve of the subject in the crash test is obtained.

[0047] Specifically, h biomechanical response curves are obtained for the same body part for h subjects in the same type of crash test.

[0048] In S33, biomechanical response data is collected based on the biomechanical response curve.

[0049] In S331, the biomechanical response curve in which an unreasonable force value appears is eliminated. Furthermore, the force value on the biomechanical response curve is obtained by detecting the force value of the strain gauge. During the crash test, when the part where the strain gauge is installed is subjected to the force, if the pressure is a positive value, all the force values ​​from the start to the end of the force application should be positive. If a negative force value appears, the force value is an unreasonable force value, which indicates that the strain gauge or the test equipment is broken. This biomechanical response curve does not match the actual situation. Therefore, the biomechanical response curve in which an unreasonable force value appears needs to be eliminated.

[0050] In S332, for all the biomechanical response curves from which the irrational force values ​​have been removed, the time of collision occurrence is set as time zero, and the biomechanical response curves with a time length of interval t from time zero are cut out, and the biomechanical response curves with the time length of interval t include the maximum force value point of the biomechanical response curve. In addition, time zero is used to align each biomechanical response curve, so that the same time period is cut out, and the maximum force value is used as the limit value on the biomechanical response curve.

[0051] In S333, s force value points are collected from the biomechanical response curve for each of the intervals of time t to obtain biomechanical response data.

[0052] In S34, a random distribution method is used to divide the body shape characteristic information and the biomechanical response data into three data sets, which are a training set, a test set and a validation set, respectively.

[0053] In S35, the biomechanical response data in the training set and the test set are input into the dimension reduction network of the self-adaptive dimension reduction neural network to perform dimension reduction processing, and linear and / or nonlinear biomechanical response low-dimensional data corresponding to the biomechanical response data in the training set and the test set are output from the intermediate layer of the self-adaptive dimension reduction neural network.

[0054] In S36, the body shape feature information in the training set is used as input, and the biomechanical response low-dimensional data is used as output to train a recurrent neural network.

[0055] Specifically, the body type feature information in the training set is used as input, and the biomechanical response low-dimensional data output by the self-adaptive dimensional reduction neural network based on the biomechanical response data in the training set is used as output to train a regression neural network, so that the biomechanical response low-dimensional data output by the regression neural network based on the body type feature information in the training set approaches the biomechanical response low-dimensional data output by the self-adaptive dimensional reduction neural network based on the biomechanical response data in the training set.

[0056] In S37, the data in the test set is used to test whether the accuracy of the recurrent neural network meets the requirements.

[0057] Specifically, after completing each training of the regression neural network, the regression neural network is tested once to test whether the accuracy of the regression neural network meets the requirement. Whether the requirement is met is determined by comparing the biomechanical response low-dimensional data output by the regression neural network based on the test set with the biomechanical response low-dimensional data output by the self-adaptive dimensionality reduction neural network based on the biomechanical response data in the test set, and calculating whether the error value, which is the difference between the two, is smaller than a preset value.

[0058] In S38, if the accuracy of the recurrent neural network meets the requirement, the training of the recurrent neural network is stopped and a trained recurrent neural network is obtained.

[0059] In S39, if the accuracy of the recurrent neural network does not meet the requirement, the recurrent neural network is trained continuously.

[0060] In S4, the accuracy of the self-adaptive dimensionality reduction neural network and the regression neural network is verified. If the accuracy meets the requirement, proceed to S5. If the accuracy does not meet the requirement, optimize the self-adaptive dimensionality reduction neural network and the regression neural network until the accuracy of the self-adaptive dimensionality reduction neural network and the regression neural network meets the requirement.

[0061] In S41, the body shape feature information in the validation set is input into the trained recurrent neural network to obtain corresponding biomechanical response low-dimensional data.

[0062] In S42, the biomechanical response low-dimensional data is input to the intermediate layer of a self-adaptive dimensionality reduction neural network, and dimensionality is expanded by a dimensionality expansion network to obtain biomechanical response high-dimensional data.

[0063] In S43, the biomechanical response high-dimensional data is compared with the biomechanical response data in the validation set, and it is calculated whether the error value is smaller than a preset value.

[0064] In S44, if the error value is smaller than the preset value, the accuracy of the self-adaptive dimensionality reduction neural network and the regression neural network meets the requirement.

[0065] In S45, if the error value is greater than or equal to the preset value, the accuracy of the self-adaptive dimension reduction neural network and the regression neural network does not meet the requirements, and the self-adaptive dimension reduction neural network and the regression neural network need to be optimized.

[0066] Specifically, the method for optimizing the self-adaptive dimension reduction neural network includes the following steps: input the biomechanical response data in the validation set into the input layer of the self-adaptive dimension reduction neural network, obtain the corresponding biomechanical response high-dimensional data from the output layer, compare the output biomechanical response high-dimensional data with the input biomechanical response data, and calculate whether the error value is smaller than a preset value. If the error value is smaller than the preset value, the accuracy of the self-adaptive dimension reduction neural network meets the requirements, and there is no need to optimize. If the error value is equal to or larger than the preset value, retrain the self-adaptive dimension reduction neural network by the method of expanding the training set and changing adjustable parameters until the error value is smaller than the preset value.

[0067] Specifically, FIG. 4 is a structural schematic diagram of an inverse regression neural network provided in an embodiment of the present invention. Referring to FIG. 4, the method for optimizing a regression neural network includes constructing an inverse regression neural network, and the structure of the inverse regression neural network is completely reversed from the structure of the regression neural network. The input layer of the original regression neural network is the output layer of the inverse regression neural network, and the output layer of the original regression neural network is the input layer of the inverse regression neural network, and the inverse function of the activation function of the original regression neural network is calculated as the activation function of the inverse regression neural network.

[0068] The biomechanical response low-dimensional data obtained by the regression neural network based on the body type feature information in the validation set is input to the inverse regression neural network, and the body type feature information corresponding to the biomechanical response low-dimensional data obtained based on the body type feature information in the validation set is obtained. Since the structure of the inverse regression neural network is completely reversed to the structure of the regression neural network, the l-th layer of the regression neural network is the (n+1-l)-th layer of the inverse regression neural network, where n is the total number of layers of the regression neural network, and n is also the total number of layers of the inverse regression neural network. The body type feature information input to the l-th layer of the regression neural network is compared with the body type feature information output from the (n+1-l)-th layer of the inverse regression neural network, and it is calculated whether the error value of each layer is smaller than a preset value. When the error values ​​of each layer are all smaller than the preset value, the accuracy of the regression neural network meets the requirements; if there is a neural network layer whose error value is equal to or greater than the preset value, the action region of the activation function of the regression neural network is restricted, for example, only the region where the nonlinear characteristics of the activation function are maximum is allowed to participate in the calculation; and the regression neural network is retrained by changing adjustable parameters until the error value is smaller than the preset value.

[0069] In S5, the body shape feature information of the human body model is input into the trained regression neural network to obtain simulated biomechanical response low-dimensional data, and the simulated biomechanical response low-dimensional data is input into the intermediate layer of the trained self-adaptive dimensional reduction neural network, and then a dimensional expansion process is performed on the simulated biomechanical response low-dimensional data by the dimensional expansion network to obtain simulated biomechanical response high-dimensional data, and a limit value curve of the biomechanical simulation of the human body model is obtained based on the simulated biomechanical response high-dimensional data.

[0070] Specifically, the body shape feature information of the human body model is input into a trained regression neural network to obtain simulated biomechanical response low-dimensional data. The simulated biomechanical response low-dimensional data is input into the intermediate layer of the trained self-adaptive dimension reduction neural network, and then the dimension expansion network performs dimension expansion processing on the simulated biomechanical response low-dimensional data to obtain simulated biomechanical response high-dimensional data. For the obtained simulated biomechanical response high-dimensional data, time is shown as the abscissa and force value is shown as the ordinate to form a simulated biomechanical response curve, and a time length of the simulated biomechanical response curve with an interval of t from the zero time when the collision occurs, including the maximum force value, is cut out, and this simulated biomechanical response curve is the limit value curve of the biomechanical simulation of the human body model.

[0071] In S6, the simulation degree of the human body model is evaluated based on the limit curve of the biomechanical simulation.

[0072] Specifically, the simulation degree of the human body model can be evaluated by the method of calculating the cross-correlation coefficient, and the specific steps include the following steps:

[0073] In S61, a biomechanical response curve of a human body model obtained in a crash test is obtained.

[0074] Specifically, the same crash test as the crash test in step S32 is performed on the human body model to obtain a biomechanical response curve of the human body model.

[0075] In S62, the biomechanical response curve is compared with the limit curve of the biomechanical simulation, and a cross-correlation coefficient between the biomechanical response curve and the limit curve of the biomechanical simulation is calculated.

[0076] Specifically, the cross-correlation coefficient is used to evaluate the similarity of two waveforms. For example, the cross-correlation coefficient has a value range of [-1, 1]. The closer the cross-correlation coefficient is to 1, the more similar the two waveforms are, and the higher the degree of simulation of the limit curve of the biomechanical simulation. The closer the cross-correlation coefficient is to 0, the more dissimilar the two waveforms are, and the lower the degree of simulation of the limit curve of the biomechanical simulation. In addition, the cross-correlation coefficient can be further quantified in the form of a percentage to more intuitively indicate the similarity of the two waveforms. For example, when the cross-correlation coefficient is 0.9, it is quantified to 90%, and the similarity of the two waveforms is 90%.

[0077] In S63, if the cross-correlation coefficient is greater than a preset value, the simulation degree of the human body model meets the requirement; if the cross-correlation coefficient is equal to or less than the preset value, the simulation degree of the human body model does not meet the requirement.

[0078] Specifically, a preset value is set according to the accuracy requirement of the simulation degree of the human body model. For example, when the accuracy requirement is 0.9 or 90%, if the cross-correlation coefficient or the quantified value of the cross-correlation coefficient is greater than the preset value of 0.9 or 90%, the simulation degree of the human body model meets the requirement; if the cross-correlation coefficient or the quantified value of the cross-correlation coefficient is equal to or lower than the preset value of 0.9 or 90%, the simulation degree of the human body model does not meet the requirement.

[0079] Specifically, the degree of simulation of the human body model can be evaluated by a method for identifying a limit value interval of the simulation, and the specific steps include the following steps:

[0080] In S61', an evaluation coefficient is set based on the required accuracy of the simulation of the human body model.

[0081] Specifically, the evaluation coefficient β% is set based on the accuracy requirement of the required degree of simulation of the human body model. For example, the higher the accuracy requirement, the smaller the evaluation coefficient; for example, if the accuracy requirement is 90%, the evaluation coefficient should be set to 5%, and if the accuracy requirement is 99%, the evaluation coefficient should be set to 0.5%.

[0082] In S62', a limit interval of the simulation is determined based on the evaluation coefficient and the limit curve of the biomechanical simulation.

[0083] Specifically, the coefficient β% of the evaluation interval is added to and subtracted from the limit curve of the biomechanical simulation to obtain two curves, +β% and -β%, and the interval formed by these two curves is the limit interval of the simulation.

[0084] In S63', a biomechanical response curve of the human body model obtained in the crash test is obtained.

[0085] Specifically, the same crash test as the crash test in step S32 is performed on the human body model to obtain a biomechanical response curve of the human body model.

[0086] In S64', the biomechanical response curve is compared with a limit value interval of the simulation. If the biomechanical response curve is within the limit value interval of the simulation, the simulation degree of the human body model meets the requirement. If the biomechanical response curve is not within the limit value interval of the simulation, the simulation degree of the human body model does not meet the requirement.

[0087] Specifically, if all points on the biomechanical response curve are within the simulation limit range, the simulation degree of the human body model meets the requirements; if there are any points on the biomechanical response curve that are not within the simulation limit range, the simulation degree of the human body model does not meet the requirements.

[0088] In an embodiment of the present invention, a functional relationship between body type feature information and low-dimensional biomechanical response data is established by a regression neural network, high-dimensional biomechanical response data for the low-dimensional biomechanical response data is obtained by a self-adaptive dimension reduction neural network, a corresponding relationship between the biomechanical response data and the biomechanical response curve is established, a biomechanical simulation limit value curve of a human body model is generated by the regression neural network and the self-adaptive dimension reduction neural network, and the simulation degree of the human body model is evaluated by the biomechanical simulation limit value curve. For any human body model with known body type feature information, a corresponding biomechanical simulation limit value curve can be generated, thereby improving the accuracy of the biomechanical simulation limit value and improving the accuracy of the evaluation of the simulation degree of the human body model.

[0089] It should be noted that the terms used in the present invention are intended only to describe specific embodiments and are not intended to limit the scope of the present application. As shown in the present specification, unless the context clearly provides an exception, terms such as "a", "one", "a kind" and / or "the" do not specify the singular but may include the plural. The terms "comprise", "comprise", or any other variation thereof are intended to cover a non-exclusive inclusion, whereby a process, method, or apparatus that includes a set of elements includes not only those elements but also other elements not expressly listed or elements inherent to such process, method, or apparatus. Unless further limited, an element defined by the phrase "comprise one" does not exclude the presence of other identical elements in the process, method, or apparatus that includes the element.

[0090] It should be further explained that the orientation or positional relationships such as "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", and "outer" are based on the orientation or positional relationships shown in the attached drawings, and are only for the purpose of making the present invention easy to explain and simplify the description, and do not indicate or imply that the indicated device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Unless otherwise clearly specified and limited, the terms "attached", "coupled", "connected", etc. should be understood in a broad sense, for example, may be fixedly connected, detachably connected, or connected together, may be mechanically connected, may be electrically connected, may be directly connected, may be indirectly connected through an intermediary, or may be communication between two elements. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific situation.

[0091] Finally, it should be noted that the above embodiments are only used to describe the technical solutions of the present invention, and are not limited thereto. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand the following: It is still possible to make modifications to the technical solutions described in the above embodiments, or to make equivalent replacements for some or all of the technical features. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the simulation degree of a human body model based on a limit curve of a biomechanical simulation, comprising: Step S1 of acquiring body type characteristic information of a human body model, the body type characteristic information including height, weight, distance from the top of the head to the pubic symphysis, shoulder width, arm length, and pelvic width; Step S2 of constructing a self-adaptive dimension reduction neural network, the self-adaptive dimension reduction neural network includes a dimension reduction network, an intermediate layer and a dimension expansion network, and the dimension reduction network and the dimension expansion network are used as separate networks respectively; Step S3 of constructing a recurrent neural network, in which the number of neurons in each layer of the recurrent neural network is the same as the number of the body type feature information; In step S4 of verifying the accuracy of the self-adaptive dimension reduction neural network and the recurrent neural network, if the accuracy meets the requirement, proceed to step S5; if the accuracy does not meet the requirement, optimize the self-adaptive dimension reduction neural network and the recurrent neural network until the accuracy of the self-adaptive dimension reduction neural network and the recurrent neural network meets the requirement; In step S5, input the body shape feature information of the human body model into the regression neural network to obtain simulated biomechanical response low-dimensional data, input the simulated biomechanical response low-dimensional data into the intermediate layer of the self-adaptive dimension reduction neural network, and then perform dimension expansion processing on the simulated biomechanical response low-dimensional data by the dimension expansion network to obtain simulated biomechanical response high-dimensional data, and obtain a limit value curve of the biomechanical simulation of the human body model according to the simulated biomechanical response high-dimensional data; The method further includes the step of: evaluating the simulation degree of the human body model based on the limit curve of the biomechanical simulation; The step S3 of constructing a recurrent neural network comprises training said recurrent neural network; The method for training the recurrent neural network includes: A step S31 of acquiring body type characteristic information of a subject; A step S32 of obtaining a biomechanical response curve of a subject in a crash test; a step S33 of collecting biomechanical response data based on the biomechanical response curve; splitting the body shape characteristic information and the biomechanical response data into three data sets using a random distribution method, the three data sets being a training set, a test set and a validation set, respectively; Step S35: inputting the biomechanical response data in the training set and the test set into a dimension reduction network of the self-adaptive dimension reduction neural network to perform dimension reduction processing, and outputting linear and / or nonlinear biomechanical response low-dimensional data corresponding to the biomechanical response data in the training set and the test set from an intermediate layer of the self-adaptive dimension reduction neural network; training the recurrent neural network using the body shape feature information in the training set as input and the biomechanical response low-dimensional data as output (S36); Step S37: testing whether the accuracy of the recurrent neural network meets the requirements using data in a test set; If the accuracy of the recurrent neural network meets the requirement, stop training the recurrent neural network in step S38 to obtain a trained recurrent neural network; If the accuracy of the recurrent neural network does not meet the requirement, the step S39 of continuing to train the recurrent neural network is included. A method for evaluating the degree of simulation of a human body model based on a limit curve of a biomechanical simulation, comprising:

2. The step S33 of collecting biomechanical response data based on the biomechanical response curve includes: A step S331 of removing a biomechanical response curve in which an irrational force value appears from the obtained biomechanical response curves; Step S332: for all the biomechanical response curves from which the irrational force values ​​have been removed, cut out the biomechanical response curves of a time length that is an interval t from the time zero, with the time when the collision occurred being set as time zero, where the biomechanical response curves of the time length that is the interval t include the maximum force value point of the biomechanical response curve; The method for evaluating the simulation degree of a human body model based on a limit value curve of a biomechanical simulation according to claim 1, further comprising: a step S333 of collecting s force value points from the biomechanical response curve of each of the intervals having a time length of t as biomechanical response data.

3. 2. The method for evaluating the simulation degree of a human body model based on a limit value curve of a biomechanical simulation according to claim 1, wherein in step S2, the number of hidden layer neurons of the self-adaptive dimension reduction neural network is equal to the number of the body shape feature information, and the number of neurons in the input layer of the dimension reduction network and the output layer of the dimension expansion network is equal to the number s of force value points collected from each of the biomechanical response curves with a time length of interval t.

4. The method for evaluating the simulation degree of a human body model based on a limit curve of biomechanical simulation as described in claim 1, characterized in that step S2 further includes adding restriction methods to the dimensionality reduction network and the dimensionality expansion network of the self-adaptive dimensionality reduction neural network, and the restriction methods include weight value association, weight value matrix orthogonality, irrelevant features and unitization.

5. The step S4 of verifying the accuracy of the self-adaptive dimensionality reduction neural network and the recurrent neural network includes: Step S41: inputting the body shape feature information in the validation set into the trained recurrent neural network and outputting biomechanical response low-dimensional data; Step S42: inputting the low-dimensional biomechanical response data into an intermediate layer of the self-adaptive dimensional reduction neural network, and performing dimensional expansion by a dimensional expansion network to obtain high-dimensional biomechanical response data; a step S43 of comparing the biomechanical response high-dimensional data with the biomechanical response data in the validation set and calculating whether the error value is smaller than a preset value; If the error value is smaller than the preset value, determining that the accuracy of the self-adaptive dimension reduction neural network and the regression neural network meets the requirements in step S44; The method for evaluating the simulation degree of a human body model based on a limit curve of biomechanical simulation according to claim 1, further comprising: determining (S45) that, if the error value is greater than or equal to the preset value, the accuracy of the self-adaptive dimension reduction neural network and the regression neural network does not meet the requirements and the self-adaptive dimension reduction neural network and the regression neural network need to be optimized.

6. Step S6 of evaluating the simulation degree of the human body model based on the limit curve of the biomechanical simulation includes: Step S61: acquiring a biomechanical response curve of the human body model obtained in a crash test; a step S62 of comparing the biomechanical response curve with the limit curve of the biomechanical simulation and calculating a cross-correlation coefficient between the biomechanical response curve and the limit curve of the biomechanical simulation; and determining (S63) that the simulation degree of the human body model meets the requirement when the cross-correlation coefficient is greater than a preset value, and that the simulation degree of the human body model does not meet the requirement when the cross-correlation coefficient is equal to or less than the preset value.

7. Step S6 of evaluating the simulation degree of the human body model based on the limit curve of the biomechanical simulation includes: A step S61' of setting an evaluation coefficient based on the required accuracy of the simulation of the human body model; A step S62' of determining a limit value interval of a simulation based on the evaluation coefficient and the limit value curve of the biomechanical simulation; A step S63' of acquiring a biomechanical response curve of the human body model obtained in a crash test; and comparing the biomechanical response curve with a limit value interval of the simulation, and determining that the simulation degree of the human body model meets the requirement if the biomechanical response curve is within the limit value interval of the simulation, and determining that the simulation degree of the human body model does not meet the requirement if the biomechanical response curve is not within the limit value interval of the simulation.

Images