Dummy injury assessment method based on simplified model deep learning algorithm

Abstract

The invention relates to the technical field of automobile safety assessment, in particular to a dummy injury assessment method based on a simplified model deep learning algorithm. Comprising the following steps: constructing a database of a one-dimensional simplified model: collecting vehicle body speed-time curve data, and adopting the calibrated one-dimensional simplified model to replace a whole vehicle model to carry out vehicle body speed-time curve data sampling; based on the calibrated one-dimensional simplified model, a vehicle body speed-time curve sample set is generated through discrete main force transmission path load distribution; substituting into a constraint system CAE model for finite element analysis, extracting damage curve data of each part of the dummy, and sorting the extracted damage curve data of the dummy into a database form; a CNN-LSTM architecture is adopted to build a deep learning model, and parameters and connection relations of all layers of the model are set; the dummy injury curve prediction module predicts dummy injury curves; and training of the CNN-LSTM prediction model is carried out. According to the technical scheme, the efficiency and accuracy of dummy injury curve prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile safety assessment, and in particular to a dummy injury assessment method based on a simplified model deep learning algorithm. Background Art

[0002] Vehicle modification and development based on mature architectures, platforms and models is one of the main processes for new product development in various automobile manufacturers. The development cycle and cost of modified vehicles are usually related to the scope of modification. For vehicle collision safety, when the scope of modification is limited to the exterior styling, cabin layout and local structural parts of the main force transmission path, the project team usually hopes to use the interior, restraint system configuration and parameters of the basic model to minimize the development cycle and cost. To ensure the achievement of the above strategy, during the product development process, in addition to using necessary CAE means to verify the crashworthiness indicators of the vehicle structure, it is also necessary to evaluate the performance of the restraint system of the modified vehicle in real time to ensure that the dummy injury indicators meet safety regulations and star development requirements.

[0003] Reference Figure 1 , the traditional dummy injury assessment method usually relies on a detailed whole vehicle collision model for simulation analysis. By inputting the vehicle body "acceleration-time" or "speed-time" curve into the basic vehicle model restraint system simulation model, the damage values ​​of various parts of the dummy are obtained to determine whether the design scheme of the modified vehicle is up to standard. This method relies on high-performance servers, finite element solvers and high-precision dummy models. Under complex and changeable collision conditions, the establishment and adjustment of the model often requires a lot of time and resources. However, during the product development process, the scheme changes frequently. On the one hand, this assessment process consumes a lot of manpower and software and hardware resources. On the other hand, due to the long CAE calculation time, the restraint system dummy injury assessment is delayed, which seriously affects the product development efficiency and makes it difficult for automobile manufacturers to respond quickly to market changes and launch modified models that meet safety performance and cost-effectiveness.

[0004] Although with the rapid development of artificial intelligence (AI) technology, automobile OEMs have increased their investment in AI technology in the research and development stage, and the application research of AI technology in the field of automobile safety has also increased. For example, in the field of automobile restraint system performance, there have been studies that apply the long short-term memory network LSTM model to whiplash performance prediction, and there are also studies that use time domain reduced-order model algorithms to predict dummy injury curves, as well as studies on the prediction accuracy of various deep learning algorithms in occupant injury assessment. However, these studies have not effectively solved the problem of rapid evaluation of restraint system dummy injury indicators in the development of modified models. Current AI application research has not yet been able to organically integrate different models and algorithms to meet the dual needs of efficiency and accuracy in restraint system dummy injury assessment in automobile modification development. Summary of the invention

[0005] The purpose of the present invention is to propose a dummy injury assessment method based on a simplified model deep learning algorithm, which can improve the efficiency and accuracy of dummy injury curve prediction. In order to achieve the above objectives, the present technical solution provides a dummy injury assessment method based on a simplified model deep learning algorithm, comprising: Construct a database of one-dimensional simplified models, including: Collect vehicle body speed-time curve data, and use the calibrated one-dimensional simplified model instead of the whole vehicle model to sample the vehicle body speed-time curve data; Based on the calibrated one-dimensional simplified model, generate a vehicle body speed-time curve sample set for restraint system CAE analysis input by discretizing the load distribution of the main force transmission path; Substitute the generated vehicle body speed-time curve sample set into the restraint system CAE model for finite element analysis, extract the damage curve data of each part of the dummy, and organize the extracted dummy damage curve data into a database form; Use CNN-LSTM architecture to build a deep learning model, set the parameters and connection relationships of each layer of the model; used to predict the dummy injury curve; Train the CNN-LSTM prediction model.

[0006] Beneficial effects of the basic solution: This technical solution uses a one-dimensional simplified model instead of a complex vehicle model to sample the vehicle body speed-time curve data, which greatly reduces the amount of calculation and time cost. This enables faster acquisition of key data during the vehicle design and safety assessment phase, thereby accelerating the entire assessment process.

[0007] After calibration, the one-dimensional simplified model can more accurately reflect the dynamic behavior of the vehicle in an actual collision. Combined with the load distribution of the discrete main force transmission path, the generated vehicle body speed-time curve sample set is closer to the actual situation, providing a reliable basis for subsequent CAE analysis.

[0008] By substituting the generated vehicle body speed-time curve sample set into the restraint system CAE model for finite element analysis, detailed injury curve data of various parts of the dummy can be extracted.

[0009] The deep learning model built with the CNN-LSTM architecture combines the spatial feature extraction capability of CNN and the temporal dependency capture capability of LSTM. It can take into account the multimodal characteristics, spatial features and temporal dependencies of the data, and can make full use of historical data and feature information to make high-precision predictions of the dummy injury curve. This can more accurately mine the potential information in the data when processing time series data related to vehicle collisions. Compared with traditional models, it has better prediction accuracy and can provide more reliable results for dummy injury assessment.

[0010] Compared with traditional physical tests, this method avoids a lot of physical testing and manpower costs. Through the prediction of the deep learning model, the injury of the dummy can be accurately assessed without conducting actual collision tests, thus reducing the overall assessment cost.

[0011] The application of this method helps promote the intelligent development of automobile safety assessment. By continuously training and optimizing the deep learning model, the prediction accuracy and generalization ability can be further improved, providing strong support for the improvement of vehicle design and safety performance.

[0012] In summary, this technical solution has shown significant beneficial effects in improving evaluation efficiency, enhancing data accuracy, building rich database resources, improving prediction capabilities, reducing evaluation costs, and promoting the improvement of intelligence and security.

[0013] As an implementable preferred solution, the one-dimensional simplified model includes multiple one-dimensional spring units, each one-dimensional unit represents a main force transmission path or component of the vehicle, and the dynamic response of the entire vehicle during a collision is simulated by setting the mechanical properties of each one-dimensional unit.

[0014] As an implementable preferred solution, the CAE model of the restraint system includes the body structure, interior parts, seat belts, airbags and dummy models, and finite element software is used for simulation analysis; and based on the specified signal processing requirements, the CAE results are batch post-processed to extract the dummy's head acceleration curve and chest compression curve as input data for the deep learning model.

[0015] As an implementable preferred solution, the mechanical properties of each one-dimensional unit are set to simulate the dynamic response of the entire vehicle during a collision, including the following: The vehicle acceleration curve is predicted by discretizing the load characteristics of the main force transmission path or the main force transmission path load value is reversely designed according to the target acceleration curve. When used in a modified vehicle, the main force transmission path load must be reversely calibrated based on the detailed finite element model calculation results of the whole vehicle. By extracting the force-displacement curve of the main force transmission path section along the collision direction of the detailed finite element model calculation results, the load of the one-dimensional spring unit of the simplified main force transmission path model is defined and fine-tuned to match the acceleration and speed-time curve of the whole vehicle.

[0016] As an implementable optimal solution, a deep learning model is built using the CNN-LSTM architecture, including the following: Each data in the data set is serialized and regularized according to the RNN data rules. The regularization calculation formula is as follows:

[0017] In the formula, is the original data; is the regularized data, and its value range is [0,1]; and They are respectively the maximum and minimum values ​​of the data on the time domain curve in all sample data.

[0018] As an implementable preferred solution, when building a prediction model, the CNN-LSTM algorithm combines the spatial feature extraction capability of CNN and the temporal dependency capture capability of LSTM to take into account the multimodal characteristics, spatial characteristics and temporal dependencies of the data; no pooling is set in the CNN network and there is no full connection step.

[0019] As an implementable preferred solution, the training of CNN-LSTM prediction model includes the following contents: The data set is divided into training set and test set by random sampling without replacement, and a CNN-LSTM architecture is built based on the open source Python deep learning framework. The Adam optimizer and mean square error function MSE are set as training parameters for model training.

[0020] As an implementable optimal solution, the Adam optimizer in the Pytorch tool is selected for model training, including the following: Set the training parameters, including the learning rate of the Adam optimizer and the loss function MSE; the calculation formula is as follows:

[0021] Where: is the sample size, is the true value, Predict values ​​for the neural network model; The training set data is input into the model for iterative training until the loss function converges or the preset number of iterations is reached.

[0022] As an implementable preferred solution, the accuracy of the CNN-LSTM prediction model is also evaluated, including the following: The accuracy of the prediction curve is evaluated from four aspects: channel, phase, amplitude, and slope. The mathematical expression is as follows:

[0023] In the formula, R is the precision value, , , and They are the channel correlation, phase correlation, amplitude correlation and slope correlation of the benchmark curve and the predicted curve respectively.

[0024] As an implementable preferred solution, the accuracy of the CNN-LSTM prediction model is evaluated, which also includes the following: The prediction accuracy R value of the test set dummy head acceleration curve and chest compression curve; compare the test set prediction results with the CAE calculation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the CAE model structure of the traditional restraint system.

[0026] Figure 2 This is a logical diagram of a dummy injury assessment method based on a simplified model deep learning algorithm.

[0027] Figure 3 Schematic diagram of one-dimensional spring unit distribution.

[0028] Figure 4 Schematic diagram of the simplified one-dimensional (1D) model structure.

[0029] Figure 5 Schematic diagram of the comparison between the one-dimensional (1D) simplified model and the basic model results of the acceleration time curve.

[0030] Figure 6 Schematic diagram of the comparison between the one-dimensional (1D) simplified model and the basic model results of the velocity time curve.

[0031] Figure 7 Schematic diagram of the sample distribution of the speed-time curve.

[0032] Figure 8 This is a schematic diagram of the CNN-LSTM model architecture.

[0033] Fig. 9 Schematic diagram of the convergence curve of the CNN-LSTM loss function of the head acceleration model.

[0034] Fig.10 Schematic diagram of the convergence curve of the CNN-LSTM loss function of the chest compression model.

[0035] Fig.11 Schematic diagram of the comparison of R value prediction accuracy of the pre-test samples of the head acceleration model.

[0036] Fig.12 This is a schematic diagram for comparing the prediction accuracy R value of chest compression test samples.

[0037] Fig.13 Schematic diagram of the comparison curve between the head acceleration model prediction results and the CAE calculation results.

[0038] Fig.14 Schematic diagram of the comparison curve between the chest compression prediction results and the CAE calculation results.

[0039] Fig.15 FIG. 4 is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the technical solution and advantages of the present application clearer, the technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings. It will be understood that the specific embodiments described herein are only partial embodiments of the present invention, which are only used to explain the present application, rather than to limit the present application. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated, and they can be combined with each other to achieve better technical effects. The same reference numerals appearing in the drawings of the following embodiments represent the same features or components, which can be applied to different embodiments.

[0041] In addition, unless otherwise defined, the technical terms or scientific terms used in the description of the present invention should have the common meanings understood by those skilled in the art in the art to which the present invention belongs.

[0042] Reference numerals: electronic device 500 includes processor 501 , communication interface 502 , memory 503 , and bus 504 .

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings: Reference Figure 2 , a dummy injury assessment method based on a simplified model deep learning algorithm, comprising: A dummy injury assessment method based on a simplified model deep learning algorithm, the method comprising the following steps: Step S100, constructing a one-dimensional (1D) simplified model database, including: Step S101, collect vehicle body speed-time curve data, and use the calibrated one-dimensional (1D) simplified model instead of the whole vehicle model to sample the vehicle body speed-time curve data. The one-dimensional (1D) simplified model has the advantage of high computational efficiency, and the single computation time is much lower than that of the whole vehicle collision model, which can significantly reduce the consumption of computing resources.

[0044] Simplify the vehicle collision model into a one-dimensional (1D) simplified model, refer to Figure 3 , including multiple one-dimensional (1D) spring units, rigid parts, tires and crossbeam placeholders. Each 1D unit represents a major force transmission path or component of the vehicle, such as the front longitudinal beam, engine compartment, passenger compartment, etc. By setting the mechanical properties of each 1D unit (such as stiffness, damping, etc.), the dynamic response of the whole vehicle during the collision is simulated. Compared with the detailed collision model of the whole vehicle, the single calculation time of the 1D simplified model is less than 1 minute under 8-core solver resources, which has extremely high computational efficiency.

[0045] Figure 4 The one-dimensional (1D) simplified model shown is mainly used for the frontal 100% rigid wall collision condition of the vehicle. The vehicle acceleration curve is predicted by discretizing the load characteristics of the main force path or the load value of the main force path is reversely designed according to the target acceleration curve. Therefore, when used in a modified vehicle, the main force path load must be reversely calibrated based on the calculation results of the detailed finite element model of the whole vehicle. In this process, the load of the one-dimensional (1D) simplified model main force path one-dimensional spring unit can be defined and fine-tuned by extracting the force-displacement curve of the main force path section along the collision direction of the detailed finite element model calculation results, matching the acceleration and speed time curve of the whole vehicle, and completing the calibration. In addition, under the premise of a certain mass distribution, there is a strict theoretical correlation between vehicle acceleration and load, and the one-dimensional (1D) simplified model is universal in passenger cars.

[0046] Specifically, based on the finite element calculation results of a passenger car's 56km / h frontal collision with a 100% overlapped rigid wall, the 1D model is simplified and calibrated. The input of the 1D simplified model (such as collision speed, collision angle, etc.) is kept consistent with the basic model to perform collision simulation calculations. Then, the acceleration curves of the 1D simplified model and the basic model are compared (refer to Figure 5 )、Speed-time curve (refer to Figure 6 ) and other key indicators, so that the 1D model acceleration curve, speed curve and the original vehicle detailed model are optimally consistent, and the error is within an acceptable range.

[0047] Occupant Load Carrier (OLC) is an indicator for evaluating vehicle acceleration and is often used to evaluate the difficulty of developing the performance of the vehicle restraint system. After integrating the vehicle acceleration curve and solving the OLC using the CAE post-processing software Meta, the OLC values ​​of the 1D simplified model and the detailed model vehicle acceleration curve were 32.1g and 31.5g, respectively, with a relative error of only 1.9%.

[0048] Based on the calibrated 1D simplified model, the load distribution of the main force transmission path is discretized to generate a sample set of vehicle body speed-time curves for the CAE analysis input of the restraint system. Specifically, the load distribution on the main force transmission path is discretized to obtain a series of vehicle body speed-time curve samples under different load distributions. In this embodiment, under the 56km / h frontal 100% overlap rigid wall collision condition, the load distribution of the main force transmission path 1D of the discretized frontal collision condition is calculated using the Lsdyna finite element software, and by extracting the output of the B-pillar position node of the rigid body passenger compartment, 200 sets of vehicle body "speed-time" curves for the CAE analysis input of the restraint system are generated, such as Figure 7 shown.

[0049] Step S102, substitute the vehicle body speed-time curve sample set generated in step S11 into the restraint system CAE model for finite element analysis. The restraint system CAE model includes the vehicle body structure, interior parts, restraint system (such as seat belts, airbags, etc.) and dummy model, etc. In this embodiment, finite element software (such as LS-DYNA) is used for simulation analysis.

[0050] The CAE results are batch post-processed based on the signal processing requirements specified in Appendix A of the C-NCAP Management Rules (2024 Edition) for the 100% overlap frontal rigid barrier collision test procedure. The processing process includes extracting the injury curve data of various parts of the dummy (such as the head, chest, etc.), and performing filtering, smoothing and other processing to obtain more accurate and reliable injury curve data. In this embodiment, the dummy head acceleration curve and chest compression curve are mainly extracted as input data for the deep learning model.

[0051] The extracted dummy head acceleration curve and chest compression curve data are organized into a database to facilitate the subsequent training and testing of the deep learning model. The database includes two parts: input variables (vehicle body speed-time curve) and output variables (dummy head acceleration curve and chest compression curve).

[0052] Step S200, constructing a prediction model, includes: Reference Figure 8 , using the open source Python deep learning framework Pytorch, to build a convolutional neural network-long short-term memory network (CNN-LSTM) architecture. The CNN-LSTM algorithm combines the spatial feature extraction capability of CNN and the temporal dependency capture capability of LSTM. The long short-term memory network (LSTM) controls the flow of information in an adaptive manner by introducing forget gates, input gates, and output gates to capture long-term dependent information. Based on the traditional LSTM algorithm, the CNN-LSTM algorithm adds a convolution operation on one-dimensional input data, which can take into account the multimodal characteristics, spatial characteristics, and temporal dependencies of the data, and has good prediction accuracy.

[0053] The structural parameters of CNN-LSTM are shown in Table 1.

[0054] Table 1

[0055] In this embodiment, when processing sequence data, the time series value is the same as the size of the CNN convolution kernel, and no edge data padding is performed in the CNN network. The output data after convolution with a single convolution kernel is a scalar, so pooling is not set in the CNN network. Secondly, after the CNN convolution in this model, only the output data is spliced ​​and expanded and the negative value elimination operation based on the Relu function is performed, so there is no full connection step in the CNN network architecture.

[0056] Each data in the data set is serialized and regularized according to the RNN data rules. In this embodiment, the time range of the time domain data is 0-100ms, the data interval is 0.1ms, the time series parameter is 3, and the regularization calculation formula is as follows:

[0057] In the formula, is the original data; is the regularized data, and its value range is between [0-1]; and They are respectively the maximum and minimum values ​​of the data on the time domain curve in all sample data.

[0058] Step S300: After the database is built and the prediction model is selected, the CNN-LSTM prediction model is trained. The training process includes steps such as data partitioning, model building, parameter setting, and training iteration.

[0059] The original data set is divided into a training set and a test set according to a certain ratio. In the present invention, the 200 groups of data are divided into 150 training sets and 50 test sets by random sampling without replacement. The same method is used to select 10 groups, 50 groups, 100 groups and 150 groups of samples from the training data set to carry out model training.

[0060] Build a CNN-LSTM architecture based on the open source Python deep learning framework Pytorch. Set the parameters and connection relationships of each layer of the model according to the training parameters defined above.

[0061] Set the training parameters of the Adam optimizer, such as the learning rate and the loss function MSE. Select the Adam optimizer in the Pytorch tool for model training. Its learning rate has adaptive characteristics. Set the initial learning rate to 0.001 and the loss function to the mean square error function MSE. Generally, the smaller the MSE during training, the better the generalization ability of the model. Its calculation formula is as follows:

[0062] Where: is the sample size, is the true value, Predict values ​​for the neural network model.

[0063] The training set data is input into the model for iterative training until the loss function converges or the preset number of iterations is reached. During the training process, the loss value of each iteration is recorded to evaluate the model performance.

[0064] The MSE value curve of the loss function during the training process of the dummy head acceleration and chest compression model is as follows: Fig. 9 and Fig.10 As shown in Figure 2, it can be seen that in the later stage of the training process, MSE has a slight oscillation phenomenon.

[0065] Step S400, CNN-LSTM prediction model accuracy analysis, including: After training is completed, the accuracy of the CNN-LSTM prediction model needs to be evaluated. The evaluation method includes comparing the prediction results of the test set with the CAE calculation results, calculating the prediction accuracy index, etc.

[0066] In this embodiment, the ISO / TS 18571 standard is used to evaluate the accuracy of the prediction curve from four aspects: channel, phase, amplitude, and slope. The mathematical expression is as follows:

[0067] In the formula, R is the precision value, , , and They are the channel correlation, phase correlation, amplitude correlation and slope correlation of the benchmark curve and the predicted curve respectively.

[0068] The prediction accuracy R value of the test set dummy head acceleration curve and chest compression curve is shown in Fig.11 and Fig.12 The closer the R value is to 1, the higher the prediction accuracy. The average R value of each prediction model is shown in Table 2.

[0069] Table 2 R means of each model

[0070] By comparing the prediction results of the test set with the CAE calculation results, it is found that the R values ​​of the CNN-LSTM prediction model on the two responses (head acceleration and chest compression) are at a high level except for sample point 2. Among them, the average R value of each model of the head acceleration curve is greater than 0.85, and the average R value of the chest compression curve is greater than 0.89. This shows that the CNN-LSTM prediction model has high prediction accuracy and generalization ability.

[0071] In addition, the effect of the training sample size on the generalization performance of the model is also studied. By comparing the prediction accuracy and loss function convergence of the model under different numbers of training samples, it is found that when the training samples reach a certain scale (such as more than 100 groups), continuing to increase the number of training samples has limited effect on improving the generalization ability of the model. Therefore, in practical applications, the appropriate number of training samples can be selected according to the specific situation to balance computing resources and prediction accuracy.

[0072] In order to better demonstrate the impact of the training sample size on the model prediction effect, further analysis is carried out in combination with the prediction curve graph. Select a sample point in the test sample data set, and its curve prediction effect is as follows: Fig.13 and Fig.14 As shown in the figure, the model prediction results of the two curve responses are in good agreement with the CAE calculation results. After evaluation, the trained model has basically met the development and use requirements of the modified model based on this model.

[0073] The disclosed embodiment also provides a dummy injury assessment system based on a simplified model deep learning algorithm, which uses a dummy injury assessment method based on a simplified model deep learning algorithm.

[0074] An embodiment of the present disclosure also provides a storage medium, in which a computer program is stored. When the computer program is executed by a processor, all steps of the above-mentioned dummy injury assessment method based on a simplified model deep learning algorithm can be implemented.

[0075] A person of ordinary skill in the art can understand that the implementation of all or part of the process in a dummy injury assessment method based on a simplified model deep learning algorithm can be completed by instructing related hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the process of each embodiment of a dummy injury assessment method based on a simplified model deep learning algorithm. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0076] The embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the above-mentioned dummy injury assessment method based on a simplified model deep learning algorithm are implemented. In the embodiment of the present application, the processor is the control center of the computer system, which can be a processor of a physical machine or a processor of a virtual machine.

[0077] Reference Fig.15 , the electronic device 500 includes: at least one processor 501, at least one communication interface 502, at least one memory 503 and at least one bus 504. The bus 504 is used to realize the connection and communication between these components, the communication interface 502 is used to communicate signaling or data with other node devices, and the memory 503 stores machine-readable instructions executable by the processor 501. When the electronic device 500 is running, the processor 501 communicates with the memory 503 through the bus 504, and when the machine-readable instructions are called by the processor 501, the steps of the above-mentioned dummy injury assessment method based on a simplified model deep learning algorithm are executed.

[0078] The above contents are only embodiments of the present invention. The common sense such as the known specific structures and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field to which the invention belongs before the application date or the priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for the ordinary technicians in the relevant field to implement this application. It should be pointed out that for the technicians in this field, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to explain the content of the claims.

Claims

1. A dummy injury assessment method based on a simplified model deep learning algorithm, characterized by: include: Construct a database of one-dimensional simplified models, including: Collect vehicle body speed-time curve data, and use the calibrated one-dimensional simplified model instead of the whole vehicle model to sample the vehicle body speed-time curve data; Based on the calibrated one-dimensional simplified model, the vehicle body speed-time curve sample set for the restraint system CAE analysis input is generated by discretizing the load distribution of the main force transmission path; Substitute the generated vehicle body speed-time curve sample set into the restraint system CAE model for finite element analysis, extract the damage curve data of each part of the dummy, and organize the extracted dummy damage curve data into a database form; Use CNN-LSTM architecture to build a deep learning model, set the parameters and connection relationships of each layer of the model; used to predict the dummy injury curve; Train the CNN-LSTM prediction model.

2. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: The one-dimensional simplified model includes multiple one-dimensional spring units, each of which represents a main force transmission path or component of the vehicle. The dynamic response of the entire vehicle during a collision is simulated by setting the mechanical properties of each one-dimensional unit.

3. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: The CAE model of the restraint system includes the vehicle body structure, interior parts, seat belts, airbags and dummy models, and finite element software is used for simulation analysis. The CAE results are batch post-processed based on the specified signal processing requirements, and the dummy head acceleration curve and chest compression curve are extracted as input data for the deep learning model.

4. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: The mechanical properties of each one-dimensional unit are set to simulate the dynamic response of the vehicle during a collision, including the following: Predict the vehicle acceleration curve by discretizing the load characteristics of the main force transmission path or reversely design the load value of the main force transmission path according to the target acceleration curve. When used in a modified vehicle, the main force transmission path load must first be reversely calibrated based on the calculation results of the detailed finite element model of the vehicle; By extracting the force-displacement curve of the main force transmission path section along the collision direction from the detailed finite element model calculation results, the load of the one-dimensional spring unit of the main force transmission path simplified model of the one-dimensional simplified model is defined and fine-tuned to match the acceleration and speed time curve of the whole vehicle.

5. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: The deep learning model is built using the CNN-LSTM architecture, including the following: Each data in the data set is serialized and regularized according to the RNN data rules. The regularization calculation formula is as follows: In the formula, is the original data; is the regularized data, and its value range is [0,1]; and They are respectively the maximum and minimum values ​​of the data on the time domain curve in all sample data.

6. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: When building the prediction model, the CNN-LSTM algorithm combines the spatial feature extraction capability of CNN and the temporal dependency capture capability of LSTM to take into account the multimodal characteristics, spatial features, and temporal dependencies of the data. No pooling is set in the CNN network and there is no full connection step.

7. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: Training of the CNN-LSTM prediction model includes the following: The data set is divided into training set and test set by random sampling without replacement, and a CNN-LSTM architecture is built based on the open source Python deep learning framework. The Adam optimizer and mean square error function MSE are set as training parameters for model training.

8. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 7, characterized in that: Select the Adam optimizer in the Pytorch tool for model training, including the following: Set the training parameters, including the learning rate of the Adam optimizer and the loss function MSE; the calculation formula is as follows: Where: is the sample size, is the true value, Predict values ​​for the neural network model; The training set data is input into the model for iterative training until the loss function converges or the preset number of iterations is reached.

9. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 1, characterized in that: It also includes an evaluation of the accuracy of the CNN-LSTM prediction model, including the following: The accuracy of the prediction curve is evaluated from four aspects: channel, phase, amplitude, and slope. The mathematical expression is as follows: In the formula, R is the precision value, , , and They are the channel correlation, phase correlation, amplitude correlation and slope correlation of the benchmark curve and the predicted curve respectively.

10. The method for dummy injury assessment based on a simplified model deep learning algorithm according to claim 9, characterized in that: The accuracy of the CNN-LSTM prediction model is evaluated, and the following are also included: The prediction accuracy R value of the test set dummy head acceleration curve and chest compression curve; compare the test set prediction results with the CAE calculation results.

Images