A method for constructing a machine learning-based neural network prediction model of a collision dummy muscle, a prediction method, a construction device, and a computer-readable storage medium.

Abstract

This invention relates to a method for constructing, predicting, and constructing a machine learning-based neural network prediction model for a collision dummy's muscles, as well as a computer-readable storage medium. The method for constructing the machine learning-based neural network prediction model for a collision dummy's muscles includes: S1, acquiring the coordinate data of muscle nodes in the collision dummy; S2, constructing the neural network prediction model; and S3, training the model. This invention proposes a method for constructing, predicting, and constructing a machine learning-based neural network prediction model for a collision dummy's muscles, as well as a computer-readable storage medium, capable of constructing a neural network prediction model that can quickly and effectively predict the coordinates of various muscle nodes in a collision dummy.

Description

Technical Field

[0001] This invention relates to the field of finite element collision simulation technology, and in particular to a method for constructing a collision dummy muscle neural network prediction model based on machine learning, a prediction method, a construction device, and a computer-readable storage medium. Background Technology

[0002] As regulations increasingly demand higher standards for vehicle crash safety, finite element crash simulation, a crucial tool for verifying vehicle designs, is being used more and more extensively in the automotive design phase. Crash dummies are an important component of finite element crash simulation; however, with the deepening of human anatomy research and the ever-increasing demands for the accuracy of simulation results, finite element models of crash dummies are evolving towards more precise components, smaller model sizes, more elements, and greater model complexity.

[0003] Under the aforementioned development trends, the difficulty of adjusting finite element collision dummies and the time required for each adjustment are gradually increasing. Currently, the main method for adjusting muscles in finite element collision dummies is to set the target rotation angle of the dummy's joints and then obtain the coordinates of the nodes of each muscle of the dummy at the target angle through finite element calculations. However, this method involves a long simulation calculation process, is cumbersome to operate, and requires significant hardware resources. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention proposes a method for constructing a neural network prediction model for a collision dummy muscle based on machine learning, a prediction method, a construction device, and a computer-readable storage medium. This model can construct a neural network prediction model, which can quickly and effectively predict the coordinates of each muscle node in a collision dummy.

[0005] Specifically, this invention proposes a method for constructing a collision dummy muscle neural network prediction model based on machine learning, including the following steps:

[0006] S1, Obtain the coordinate data of the muscle nodes of the collision dummy, including the following steps:

[0007] S11, Based on the different application conditions of the collision dummy, determine the adjustment characteristics of the collision dummy and the adjustment range of each adjustment characteristic;

[0008] S12, Based on the aforementioned adjustment characteristics and their adjustment range, conduct orthogonal simulation experiment design and establish an orthogonal simulation design table;

[0009] S13. Based on the orthogonal simulation design table, construct a finite element calculation model of the muscle nodes of the collision dummy, and solve the finite element calculation model to generate coordinate data of the muscle nodes of the collision dummy.

[0010] S2, Constructing the neural network prediction model, including the following steps;

[0011] S21, Organize the calculation results of the orthogonal simulation experiment to generate training dataset and validation dataset;

[0012] S22, Construct an equivalent small model, wherein the equivalent small model is a model with fewer nodes composed of muscle node coordinates of the parts of the collision dummy with large muscle deformation selected from the muscle node coordinate data of the collision dummy.

[0013] S23, using the training dataset, the neural network prediction algorithm is used to predict the muscle nodes in the equivalent small model. The determination coefficient of the neural network prediction model is calculated using the validation dataset. The fitting relationship between the hyperparameters of the neural network prediction algorithm and the determination coefficient is established. The fitting relationship is optimized to obtain the optimal combination of hyperparameters of the neural network prediction algorithm. The neural network prediction model corresponding to the neural network prediction algorithm is established based on the optimal combination of hyperparameters.

[0014] S3, Model Training;

[0015] S31, Determine the weight coefficients of the neural network prediction model based on the training dataset;

[0016] S32, Determine the determination coefficients of the neural network prediction model based on the validation dataset;

[0017] S33, determine whether the determination coefficient meets the set target accuracy. If not, return to step S12. If yes, the construction of the neural network prediction model is completed.

[0018] According to an embodiment of the present invention, in step S12, the adjustment feature is used as an orthogonal simulation design table factor, and the level of the adjustment feature is the orthogonal simulation design table level.

[0019] According to an embodiment of the present invention, step S13 includes the following steps:

[0020] S131, Based on the connection relationship of each joint of the collision dummy, classify each joint of the collision dummy into different levels;

[0021] S132, based on the joint level, calculates the target coordinates of each joint center point sequentially using the initial coordinates of each joint center point;

[0022] S133, establish a loading model for the collision dummy's joint center point to move from the initial coordinates to the target coordinates, corresponding to the collision dummy's limbs moving smoothly from the initial position to the target position;

[0023] S134, Solve the finite element calculation model using a solver, obtain the initial and target coordinates of the muscle nodes of the collision dummy in the calculation results, and generate the muscle node coordinate data of the collision dummy.

[0024] According to an embodiment of the present invention, step S21 includes the following steps:

[0025] S211, normalize the feature matrix corresponding to the orthogonal simulation test design table;

[0026] S212, split the x, y, z coordinate values ​​and unit numbers of each muscle node in the collision dummy muscle node coordinate data, and integrate them according to the x, y, z coordinate values ​​to generate a dataset.

[0027] S213, The dataset is randomly divided according to a set ratio to determine the training dataset and the validation dataset.

[0028] According to an embodiment of the present invention, in step S23, the fitting relationship is optimized using a particle swarm optimization algorithm. The optimization process includes setting the optimization objective as maximizing the coefficient of determination, the optimization variable as the number of neurons in each layer of the neural network prediction algorithm, and the constraints as constraints on the total number of neurons in the neural network prediction algorithm and correlation constraints.

[0029] According to an embodiment of the present invention, in step S31, an initial coordinate matrix of the muscle nodes of the collision dummy is constructed, the training dataset and the initial coordinate matrix are subtracted to obtain the training increment dataset, and the weight coefficients of the neural network prediction algorithm are calculated and saved through the training increment dataset.

[0030] In step S32, an initial coordinate matrix of the muscle nodes of the collision dummy is constructed. The difference between the verification dataset and the initial coordinate matrix is ​​calculated to obtain the verification increment dataset. The determination coefficient of the neural network prediction algorithm is calculated through the verification increment dataset.

[0031] This invention also provides a method for adjusting the muscles of a collision dummy, applicable to the neural network prediction model constructed by the aforementioned construction method, wherein the method for adjusting the muscles of a collision dummy includes:

[0032] T1, using the neural network prediction model to predict the coordinates of the muscle nodes of the collision dummy;

[0033] T2 transmits the prediction results to the existing finite element model.

[0034] According to an embodiment of the present invention, step T1 includes the following steps:

[0035] T11, create a vector containing the angles of each joint of the collision dummy in each direction, use this vector as the input of the neural network prediction model, and obtain the coordinate values ​​of the muscle nodes of the collision dummy through the calculation of the neural network prediction model, and save the coordinate values ​​of the muscle nodes as a matrix;

[0036] T12 extracts the ID number of the muscle node of the collision dummy in the form of a column vector, adds the column vector of the muscle node ID number before the muscle node coordinate value, and saves it as a matrix;

[0037] In step T2, the matrix is ​​saved as a text format, generating a standard format text. Using the existing finite element model, the skeleton of the collision dummy is adjusted to the target angle in the finite element preprocessing software. Then, the standard format text is imported into the finite element model.

[0038] The present invention also provides a device for constructing a collision dummy muscle neural network prediction model based on machine learning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for constructing a collision dummy muscle neural network prediction model based on machine learning provided by the present invention.

[0039] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for constructing a machine learning-based collision dummy muscle neural network prediction model provided by the present invention.

[0040] This invention provides a method for constructing a machine learning-based neural network prediction model for a collision dummy muscle, a prediction method, a construction device, and a computer-readable storage medium. Based on muscle node coordinate data under different finite element collision dummy joint angles, the invention obtains the functional relationship between the finite element collision dummy joint angles and muscle node coordinates through a machine learning algorithm. It then uses an equivalent small model to determine the hyperparameter combination of the neural network prediction algorithm and establishes the corresponding neural network prediction model. This neural network prediction model can quickly and effectively predict the coordinates of each muscle node in the collision dummy.

[0041] It should be understood that the above general description and the following detailed description of the invention are exemplary and illustrative, and are intended to provide further explanation of the invention as described in the claims. Attached Figure Description

[0042] The accompanying drawings are included to provide further explanation of the invention. They are incorporated into and constitute a part of this application. The drawings illustrate embodiments of the invention and, together with this specification, serve to explain the principles of the invention.

[0043] In the attached image:

[0044] Figure 1 A flowchart illustrating a method for constructing a machine learning-based neural network prediction model for a collision dummy muscle according to an embodiment of the present invention is shown.

[0045] Figure 2 A schematic diagram of the hip joint adjustment of a finite element collision dummy according to an embodiment of the present invention is shown.

[0046] Figure 3 A schematic diagram of the knee joint adjustment of a finite element collision dummy according to an embodiment of the present invention is shown.

[0047] Figure 4 A flowchart of a collision dummy muscle adjustment method according to an embodiment of the present invention is shown. Detailed Implementation

[0048] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0050] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0051] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0052] Furthermore, it should be noted that the use of terms such as "first" and "second" to define components is merely for the purpose of distinguishing the corresponding components. Unless otherwise stated, these terms have no special meaning and therefore should not be construed as limiting the scope of protection of this application. In addition, although the terminology used in this application is selected from commonly known and used terms, some terms mentioned in this application's specification may have been chosen by the applicant according to his or her judgment, and their detailed meanings are explained in the relevant sections of this description. Moreover, this application should be understood not only through the actual terms used, but also through the meaning implied by each term.

[0053] It should be noted that the finite element collision dummy muscle adjustment involved in this invention refers to the process of adjusting the joint angles of the collision dummy by taking into account its placement position and its sitting posture in the vehicle, and ensuring that there is no interference between the dummy's skeleton and muscles.

[0054] Figure 1 A flowchart illustrating a method for constructing a collision dummy muscle neural network prediction model based on machine learning according to an embodiment of the present invention is shown. As shown, a method for constructing a collision dummy muscle neural network prediction model based on machine learning includes the following steps:

[0055] S1, Obtain the coordinate data of the muscle nodes of the collision dummy, including the following steps:

[0056] S11. Based on the different application conditions of the collision dummy, determine the adjustment characteristics of the collision dummy and the adjustment range of each adjustment characteristic. The adjustment characteristics of the collision dummy are the rotation angles of the dummy's joints in various directions, such as the rotation angles of the dummy's thighs and ankles. The adjustment range of the adjustment characteristics is determined based on the geometric constraint relationship of the finite element collision dummy and the possible adjustment angle range of the dummy in the working condition simulation.

[0057] S12. Based on the regulation characteristics and their regulation range, conduct orthogonal simulation experiment design and establish an orthogonal simulation design table;

[0058] S13, Based on the orthogonal simulation design table, construct the finite element calculation model of the muscle nodes of the collision dummy, and solve the finite element calculation model to generate the coordinate data of the muscle nodes of the collision dummy;

[0059] S2, Construct a neural network prediction model, including the following steps;

[0060] S21, Organize the calculation results of the orthogonal simulation experiment to generate training dataset and validation dataset;

[0061] S22. Construct equivalent small models. Equivalent small models are models with fewer nodes, selected from the muscle node coordinate data of the collision dummy, focusing on the muscle nodes in areas of significant muscle deformation. For example, extract the coordinate data of 10 rows of muscle nodes near each joint of the collision dummy's limbs to create multiple equivalent small models. Specifically, extract the node coordinates of the equivalent small models from the files "Total x-coordinate," "Total y-coordinate," and "Total z-coordinate," respectively, and place them into "Small Model x-coordinate," "Small Model y-coordinate," and "Small Model z-coordinate."

[0062] S23, using the training dataset, the neural network prediction algorithm is used to predict the muscle nodes in the equivalent small model. The determination coefficient of the neural network prediction model is calculated using the validation dataset. The fitting relationship between the hyperparameters of the neural network prediction algorithm and the determination coefficient is established. The fitting relationship is optimized to obtain the optimal combination of hyperparameters of the neural network prediction algorithm. Based on the optimal combination of hyperparameters, the neural network prediction model corresponding to the neural network prediction algorithm is established.

[0063] S3, Model Training;

[0064] S31, Determine the weight coefficients of the neural network prediction model based on the training dataset;

[0065] S32, Determine the determination coefficients of the neural network prediction model based on the validation dataset;

[0066] S33, determine whether the determination coefficient meets the set target accuracy. If not, return to step S12. If yes, the construction of the neural network prediction model is completed.

[0067] Preferably, in step S12, the adjustment feature is used as the factor of the orthogonal simulation design table, the level of the adjustment feature is the level of the orthogonal simulation design table, and the feature parameter matrix corresponding to the orthogonal simulation design table is generated.

[0068] Preferably, step S13 includes the following steps:

[0069] S131, based on the connection relationships of each joint of the collision dummy, the joints of the collision dummy are classified into different levels. Points whose coordinates do not change during the dummy's posture adjustment are set as fixed reference points. Bones connected to fixed reference points are called primary bones, joints between primary bones and fixed reference points are called primary joints, bones connected to primary bones through joints and capable of rotation are called secondary bones, joints between primary and secondary bones are called secondary joints, and so on, defining the dummy's bones and joints. Figure 2 A schematic diagram of the hip joint adjustment of a finite element collision dummy according to an embodiment of the present invention is shown. Figure 3 A schematic diagram of the knee joint adjustment of a finite element collision dummy according to an embodiment of the present invention is shown. As shown in the figure, the lines in the figure correspond to the lower limbs of the dummy, including the thigh and the lower leg. Point A is the rotation center of the hip joint, point B is the rotation center of the knee joint, point C is the rotation center of the ankle joint, line segment AB represents the femur of the dummy, and line segment BC represents the shinbone of the dummy. Since the position of point H of the dummy does not change when adjusting the posture of the finite element collision dummy, the coordinates of the rotation center of the hip joint of the collision dummy also do not change. According to the above definition, point A is the fixed reference point of the dummy, the femur AB is the primary skeleton of the dummy, and the shinbone BC is the secondary skeleton of the dummy. Therefore, the hip joint of the dummy can be defined as a primary joint, the knee joint as a secondary joint, and the ankle joint as a tertiary joint.

[0070] S132, based on the joint level, sequentially calculates the target coordinates of each joint's center point using the initial coordinates of each joint's center point. (Reference) Figure 2 α is the rotation angle of the dummy's hip joint around the Y-axis of the spatial coordinate system, β is the angle between the dummy's femur and the spatial XY plane before adjustment, γ is the angle between the dummy's femur and the spatial XY plane after adjustment, h1 is the distance between the center point of the dummy's knee joint and the spatial XY plane before adjustment, and h2 is the distance between the center point of the dummy's knee joint and the spatial XY plane after adjustment.

[0071] When adjusting the hip joint of a primary articulated dummy, the h1 value is first calculated. Then, based on the length AB, the angles β and γ are calculated using trigonometric functions. Next, h2 is calculated based on the angle γ, and finally, the B' coordinate value is determined. The same method is used to determine the C' coordinate value after point C rotates around point A.

[0072] Figure 3 Is Figure 2Based on this, the dummy's knee joint rotates, where α1 is the rotation angle of the dummy's hip joint around the Y-axis of the spatial coordinate system, β1 is the angle between the dummy's tibia and the spatial XY plane before adjustment, γ1 is the angle between the dummy's tibia and the spatial XY plane after adjustment, h3 is the distance between the center point of the dummy's ankle joint and the spatial XY plane before adjustment, and h4 is the distance between the center point of the dummy's ankle joint and the spatial XY plane after adjustment. Similarly, when adjusting joints of level two and above, the same calculation method as for level one joints is used to determine the corresponding joint coordinate values, which will not be elaborated here.

[0073] S133, establish a loading model for the collision dummy's movement from the initial coordinates of the joint center point to the target coordinates of the joint center point. The corresponding loading curve can be a sinusoidal loading curve, which corresponds to the collision dummy's limbs moving smoothly from the initial position to the target position.

[0074] S134: The finite element model is solved using a solver to obtain the initial and target coordinates of the muscle nodes of the collision dummy, generating muscle node coordinate data. Specifically, the finite element model can be solved sequentially using the Pamcrash solver. The post-processing software Animator4 reads the calculation results and obtains the initial and final coordinates of the dummy's muscle nodes. These coordinates are saved as .inc files in the Pamcrash software, named "Initial Coordinates" and "Final Coordinates". Non-node coordinate information in the .inc files saved by Animator4 can be removed using shell commands, and the files can be re-saved.

[0075] Preferably, step S21 includes the following steps:

[0076] S211, normalize the feature matrix corresponding to the orthogonal simulation test design table. Normalize the feature matrix in the form of row normalization, that is, divide the value of each row by the absolute value of the sum of the squares of all elements in each row.

[0077] S212 involves splitting the x, y, and z coordinate values ​​and unit numbers of each muscle node in the collision dummy muscle node coordinate data, and then integrating them according to the x, y, and z coordinate values ​​to generate a dataset. Specifically, for each muscle node coordinate data, a shell script is used to split the file "Final Coordinates" into four files: "Node ID," "x-coordinate," "y-coordinate," and "z-coordinate." These four files contain the ID (data number), x-coordinate value, y-coordinate value, and z-coordinate value of the dummy muscle node, respectively. A Python script is used to iterate through each calculation result (muscle node coordinate data) and combine the "x-coordinates" from all calculation results to generate an n*m dimensional data matrix, where n is the number of muscle nodes and m is the number of orthogonal matrices. The same method is used to combine the "y-coordinates" and "z-coordinates" from all calculation results. Finally, these three matrices are saved as text files named "Total x-coordinates," "Total y-coordinates," and "Total z-coordinates," respectively, thus generating the dataset according to the x, y, and z coordinate values.

[0078] S213, randomly divide the dataset according to a set ratio to determine the training dataset and the validation dataset. Typically, the set ratio can be 8:2.

[0079] Preferably, in step S23, the hyperparameter determination of the machine learning method based on the equivalent small model uses the variable matrix of the training set as the feature matrix. A neural network prediction algorithm is applied to the equivalent small model to predict the coordinates of muscle nodes. The determination coefficient of the neural network prediction model is calculated based on the validation dataset. Then, the hyperparameters of the prediction algorithm are changed multiple times to obtain the determination coefficients of the prediction model under multiple combinations of hyperparameters. Finally, a functional relationship (polynomial equation) between the hyperparameters and the determination coefficients is fitted. This fitting process includes generating an orthogonal experimental table of hyperparameters for the neural network prediction algorithm. First, the number of neurons in each layer of the prediction algorithm, n1, ..., ni, is selected as a factor, where i is the maximum number of layers in the prediction algorithm. The average determination coefficient of the prediction algorithm is used as the index. To reasonably balance the running time of subsequent optimization processes, the maximum number of layers i is recommended to be less than or equal to 5, and the number of factors in the selected orthogonal experimental table should be greater than 5. When setting the factor levels, the number of hidden layer neurons must be greater than or equal to 0, and the case of 0 must be included. Then, an orthogonal experimental table is selected, and the hyperparameter orthogonal table of the prediction algorithm is filled in according to the standard orthogonal experimental table. Since the number of neurons in each hidden layer must satisfy the correlation constraint (i.e., if the number of neurons in the preceding hidden layer is 0, then the number of neurons in the subsequent hidden layers should also be 0), the orthogonal experimental table of hyperparameters needs to be modified. Experiments that do not meet the correlation constraint should be deleted. If the number of remaining experimental groups is less than 80% of the number of groups in the initial orthogonal experimental table, the number of levels of the factor should be increased and the process of establishing the orthogonal experimental table of hyperparameters should be repeated until the number of remaining experimental groups is greater than or equal to 80% of the number of groups in the initial orthogonal experimental table.

[0080] Ideally, the coefficient of determination of the predictive model This is the average of the determination coefficients for the prediction results of individual coordinate values, and the determination coefficient for an individual coordinate value can be calculated as follows:

[0081]

[0082]

[0083]

[0084] In the above formula y represents the average coordinate values ​​of the muscle nodes in the equivalent small model in the validation dataset. i f represents the true values ​​of the muscle node coordinates in the validation dataset for the equivalent small model. i R represents the predicted coordinates of muscle nodes in the equivalent small model. 2 The coefficient of determination for a single coordinate value of the equivalent small model.

[0085] The functional relationship between the determination coefficients and hyperparameters of the prediction algorithm is fitted using the least squares method, resulting in the following formula:

[0086]

[0087] In the above formula, x n b represents the number of neurons in the nth layer of the neural network. m Let i be the coefficient of the function. n is the power of the number of neurons in the nth layer of the neural network, and j is the power of the highest-order term.

[0088] Preferably, in step S23, a particle swarm optimization algorithm is used to optimize the fitting relationship. This optimization process includes setting the optimization objective to maximize the coefficient of determination, the optimization variable to be the number of neurons in each layer of the neural network prediction algorithm, and the constraints to the total number of neurons in the neural network prediction algorithm and correlation constraints. The particle swarm optimization algorithm is used to optimize and solve the polynomial equation to obtain the optimal combination of hyperparameters. Specifically, the algorithm optimization objective is... Minimum, the optimization target is the number of neurons in each layer.

[0089] The constraint related to the number of neurons is that the number of neurons per layer is an integer between 0 and Nmax, which is represented by the following formula:

[0090] x n ∈{0,…,N max}

[0091] Due to limitations in computing resources, the total number of neurons should also be less than a set value T, as represented by the following formula:

[0092]

[0093] The correlation constraint states that if the number of neurons in the hidden layer of a preceding neural network is 0, then the number of neurons in the subsequent hidden layers will also be 0. This can be represented by the following formula:

[0094] if x i =0, then x j =0,j>i

[0095] The particle swarm optimization algorithm is used to optimize hyperparameters and obtain the optimal combination of hyperparameters.

[0096] Preferably, in step S31, an initial coordinate matrix of the muscle nodes of the collision dummy is constructed, and the training dataset and the initial coordinate matrix are subtracted from each other to obtain the training increment dataset. The weight coefficients of the neural network prediction algorithm are calculated and saved using the training increment dataset.

[0097] In step S32, an initial coordinate matrix of the muscle nodes of the collision dummy is constructed. The validation dataset and this initial coordinate matrix are subtracted from each other to obtain the validation incremental dataset. The determination coefficients of the neural network prediction algorithm are then calculated using the validation incremental dataset. Specifically, using the features of the validation incremental dataset as input to the prediction algorithm, a predicted value matrix is ​​constructed to store the calculation results, and a ground truth matrix is ​​constructed to store the true values ​​of the validation incremental dataset. The average determination coefficient of the prediction algorithm is obtained by summing the determination coefficients of individual coordinate values ​​and averaging them. The coefficient of determination for a single coordinate value can be calculated as follows:

[0098]

[0099]

[0100]

[0101] In the above formula y represents the average coordinate values ​​of the dummy muscle nodes in the validation set. i_d f represents the true values ​​of the dummy muscle node coordinates in the validation set. i_d The predicted values ​​represent the coordinates of the muscle nodes in the dummy. The coefficient of determination for a single coordinate value of the dummy.

[0102] In step S33, the determination coefficients are... The accuracy is compared with the target accuracy. If the accuracy is not met, return to step S12 to redesign the orthogonal simulation experiment and execute the subsequent steps. If the accuracy is met, the construction of the neural network prediction model is complete.

[0103] This invention also provides a method for adjusting the muscles of a collision dummy, applicable to neural network prediction models constructed using the aforementioned construction method. The method for adjusting the muscles of a collision dummy includes:

[0104] T1 uses the constructed neural network prediction model to predict the coordinates of the muscle nodes of the collision dummy;

[0105] T2 transmits the prediction results to the existing finite element model.

[0106] Preferably, step T1 includes the following steps:

[0107] T11: Create a vector containing the angles of each joint of the collision dummy in each direction. Use this vector as the input to the neural network prediction model. The neural network prediction model will calculate the coordinate values ​​of the muscle nodes of the collision dummy. Keep the relative order of the coordinate values ​​of each node unchanged and save the coordinate values ​​of the muscle nodes as a matrix.

[0108] T12 extracts the ID numbers of the muscle nodes of the collision dummy in the form of column vectors, adds the column vector of muscle node ID numbers before the muscle node coordinate values, and saves it as a matrix.

[0109] In step T2, the matrix obtained in the previous step is saved as a text file, generating a standard format text. Using the existing finite element model, the skeleton of the collision dummy is adjusted to the target angle in the finite element preprocessing software. Then, the standard format text is imported into the finite element model. For example, in step T11, when recording muscle nodes, in addition to recording the muscle node ID number, x, y, and z coordinates, descriptive information in character form is added. For example, PamCrash records muscle nodes in the following format:

[0110] NODE / 90061349 38.900002 40.099998 93.400002

[0111] For this type of finite element file, "NODE++ / +" can be uniformly added before the matrix containing the muscle node data numbers and coordinate values ​​recorded in step T12. This matrix is ​​then saved as text. Finally, a shell script is used to replace all the "+" signs with spaces, generating a standard PamCrash format text. Next, the skeleton of the collision dummy is adjusted to the target angle in the finite element preprocessing software. Then, the standard PamCrash format text is imported, and duplicate node coordinates are replaced, thus completing the transfer of the dummy's muscle node coordinates to the finite element model.

[0112] The collision dummy muscle adjustment method utilizes the constructed neural network prediction model to directly predict the muscle nodes of the collision dummy. Therefore, in the subsequent collision dummy adjustment process, there is no need to perform dummy pre-simulation calculations, saving computing resources and time.

[0113] The present invention also provides a device for constructing a collision dummy muscle neural network prediction model based on machine learning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned method for constructing a collision dummy muscle neural network prediction model based on machine learning.

[0114] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the aforementioned method for constructing a machine learning-based collision dummy muscle neural network prediction model.

[0115] The specific implementation methods and technical effects of the device for constructing the collision dummy muscle neural network prediction model based on machine learning and the computer-readable storage medium can be found in the embodiments of the method for constructing the collision dummy muscle neural network prediction model based on machine learning provided by the present invention, and will not be repeated here.

[0116] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0117] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0118] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0119] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0120] It will be apparent to those skilled in the art that various modifications and variations can be made to the exemplary embodiments described above without departing from the spirit and scope of the invention. Therefore, it is intended that this invention cover modifications and variations falling within the scope of the appended claims and their equivalents.

Claims

1. A method for constructing a machine learning-based crash dummy muscle neural network prediction model, comprising the steps of: S1, obtaining crash dummy muscle node coordinate data, comprising the steps of: S11, determining the adjustment characteristics of the crash dummy and the adjustment range of each adjustment characteristic according to different application conditions of the crash dummy; S12, based on the adjustment characteristics and their adjustment ranges, designing an orthogonal simulation experiment, and establishing an orthogonal simulation design table; S13, based on the orthogonal simulation design table, constructing a finite element calculation model of the crash dummy muscle node, solving the finite element calculation model to generate crash dummy muscle node coordinate data; S2, constructing the neural network prediction model, comprising the steps of: S21, organizing the calculation results of the orthogonal simulation experiment to generate a training data set and a validation data set; S22, constructing an equivalent small model, which is composed of muscle node coordinates of the joint near part of the crash dummy selected from the crash dummy muscle node coordinate data; S23, using the training data set to predict the muscle nodes in the equivalent small model by a neural network prediction algorithm, using the validation data set to calculate the determination coefficient of the neural network prediction model, establishing the fitting relationship between the hyperparameters of the neural network prediction algorithm and the determination coefficient, optimizing the fitting relationship to obtain the optimal hyperparameter combination of the neural network prediction algorithm, and establishing the neural network prediction model corresponding to the neural network prediction algorithm based on the optimal hyperparameter combination; S3, model training; S31, determining the weight coefficient of the neural network prediction model based on the training data set; S32, determining the determination coefficient of the neural network prediction model based on the validation data set; S33, determining whether the determination coefficient meets the set target accuracy, if not, returning to step S12, and if yes, completing the construction of the neural network prediction model. 2.The method of claim 1, wherein, In step S12, the adjustment characteristics are used as the factors of the orthogonal simulation design table, and the levels of the adjustment characteristics are the levels of the orthogonal simulation design table. 3.The method of claim 1, wherein the method further comprises: training the neural network model using the training data; and determining a performance of the neural network model using the validation data. Step S13 comprises the steps of: S131, based on the connection relationship of each joint of the crash dummy, classifying each joint of the crash dummy; S132, based on the joint level, sequentially calculating the target coordinates of each joint center point through the initial coordinates of each joint center point; S133, establishing a loading model of the movement of the initial coordinates of the joint center points of the crash dummy to the target coordinates of the joint center points, corresponding to the smooth movement of the limbs of the crash dummy from the initial position to the target position; S134, solving the finite element calculation model by a solver to obtain the initial coordinates and target coordinates of the muscle nodes of the crash dummy in the calculation results, and generating the crash dummy muscle node coordinate data. 4.The method of claim 1, wherein the method further comprises: training the neural network model using the training data; and determining a performance of the neural network model using the validation data. Step S21 comprises the steps of: S211, normalizing the feature matrix corresponding to the orthogonal simulation experiment design table; S212, splitting the x, y, z coordinate values and unit numbers of each muscle node in the crash dummy muscle node coordinate data, and integrating to generate a data set according to the x, y, z coordinate values; S213, determining a training data set and a validation data set according to a set proportion by randomly dividing the data set. 5.The method of claim 1, wherein the method further comprises: training the neural network model using the training data; and determining a performance of the neural network model using the validation data. In step S23, the fitting relationship is optimized by using a particle swarm algorithm, and the optimization process includes setting an optimization target as a maximum determination coefficient, setting an optimization variable as a number of neurons of each layer of the neural network prediction algorithm, and setting a constraint condition as a constraint on a total number of neurons of the neural network prediction algorithm and a relevance constraint. 6.The method of constructing a machine learning based crash dummy muscle neural network prediction model according to claim 1, wherein, In step S31, an initial coordinate matrix of muscle nodes of a crash dummy is constructed, a training incremental data set is obtained by performing difference processing on the training data set and the initial coordinate matrix, and weight coefficients of the neural network prediction algorithm are calculated and saved through the training incremental data set. In step S32, an initial coordinate matrix of muscle nodes of a crash dummy is constructed, a validation incremental data set is obtained by performing difference processing on the validation data set and the initial coordinate matrix, and a determination coefficient of the neural network prediction algorithm is calculated through the validation incremental data set.

7. A method for adjusting the muscle of a crash test dummy, which is suitable for the neural network prediction model constructed by the construction method of any one of claims 1 to 6, characterized in that, The muscle adjustment method of the crash dummy includes: T1, predicting coordinates of muscle nodes of the crash dummy by using the neural network prediction model; T2, transmitting the prediction result to an existing finite element model.

8. The crash test dummy muscle adjustment method of claim 7, wherein, Step T1 includes steps of: T11, creating a vector including angles of each joint of the crash dummy in each direction, taking the vector as an input of the neural network prediction model, obtaining a muscle node coordinate value of the crash dummy through operation of the neural network prediction model, and saving the muscle node coordinate value as a matrix; T12, extracting an ID number of the muscle node of the crash dummy in the form of a column vector, adding the column vector of the muscle node ID number before the muscle node coordinate value, and saving as a matrix; In step T2, the matrix is saved as a text format to generate a standard format text, the skeleton of the crash dummy is adjusted to a target angle in a finite element pre-processing software by using the existing finite element model, and then the standard format text is imported into the finite element model. 9.A device for constructing a machine learning based crash dummy muscle neural network prediction model, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the construction method of the muscle neural network prediction model of the crash dummy based on machine learning according to any one of claims 1-6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the construction method of the muscle neural network prediction model of the crash dummy based on machine learning according to any one of claims 1-6.

Images