Computer-implemented method for simulating accident of motor vehicle using artificial neural network
By transforming input data in the hidden layer of the artificial neural network and generating a single transform scalar or embedded signal using a transformer architecture or attention mechanism, the accuracy and reality in the transformation process of simulated motor vehicle accident signal in the prior art is solved. The problem is that higher precision and realistic simulation results are achieved, thereby improving accident safety.
Patent Information
- Application Number
- CN202410714047.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-06-15
- Filing Date
- 2024-06-04
- Publication Date
- 2024-12-17
AI Technical Summary
The existing methods for simulating motor vehicle accidents are difficult to accurately explain the way artificial neural networks process input signals during signal transformation, and it is difficult to evaluate whether the used data part is correct, affecting the accuracy and fidelity of the simulation.
By receiving and transforming input data in the hidden layer of an artificial neural network, generating a single transform scalar or embedded signal, using a transformer architecture or attention mechanism for data conversion, and inputting it into an appropriate hierarchy to generate more Accurate and realistic simulation results.
Improve the accuracy of the simulation and the understanding of the impact on the input data, and generate particularly realistic simulation results, which can more effectively evaluate the severity of the accident and the status of the components, thereby improving the accident safety of the motor vehicle.
Smart Images
Figure CN119150650A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a computer-implemented method for simulating an accident of a motor vehicle using an artificial neural network. Background Art
[0002] DE 10 2020 115 192 A1 discloses a method for simulating an accident of a motor vehicle using an artificial neural network. An input scalar related to the accident is transformed into a transformed signal in a hidden layer of the network. During the transformation, a mathematical convolution of the input scalar is performed. The accident is simulated using the transformed signal. Summary of the invention
[0003] With this in mind, a basic object of the present invention is to create an improved method for simulating accidents of motor vehicles.
[0004] This object is achieved at least by the method described below. Embodiments of the present invention are further explained in the method described below.
[0005] According to the method, input data related to the accident is received. The input data may be data describing, for example, characteristics of the motor vehicle and / or individuals in the motor vehicle. The input data may, for example, describe the position, length, width and / or height of a component and / or body part.
[0006] The input data is transformed into transformed data. This can be done, for example, in a hidden layer of an artificial neural network. The artificial neural network uses the transformed data to simulate an accident.
[0007] For example, the artificial neural network can have an encoder-decoder architecture. For example, convolutions of different dimensions can be used. Preferably, a one-dimensional convolution can be used for the encoder and a two-dimensional convolution can be used for the decoder.
[0008] This is a computer-implemented method. This may mean using, for example, a digital processing device and a digital data memory to perform this method. The digital data memory may store instructions that can be read and executed by the processing device. The instructions may be configured to cause the processing device to perform a method according to an embodiment of the present invention when the instructions are executed.
[0009] According to one embodiment of the present invention, the input data may be configured as an input signal. The transformed data may include a scalar. The scalar may be converted into a transformed signal before simulation. The simulation may be performed using the transformed signal.
[0010] This embodiment is advantageous because the scalars transformed from the input signal make it possible to interpret the way the artificial neural network processes the input signal during the simulation. For example, the scalars can be generated in the hidden layers of the neural network. The data expert can analyze these scalars to determine the impact of the input signal on the simulation. It can also be checked whether the appropriate part of the input signal is being used.
[0011] For example, a scalar may include information about an input signal.For example, a transformed signal may include several values related to the acceleration of a body part of a person in a motor vehicle during a simulated accident.
[0012] According to one embodiment of the present invention, the input data may be configured as an input scalar.
[0013] According to one embodiment of the invention, the transformation data may be configured as a single transformation scalar. The transformation scalar may include, for example, an indication of the severity of the accident. The transformation scalar may include, for example, an indication of the maximum acceleration of a body, a body part, a component of a motor vehicle, or a motor vehicle.
[0014] This embodiment is particularly advantageous because a single transform scalar includes less information than the input data, making it easier for the data expert to assess whether the correct portion of the input data is being used and how the simulation is affected by the input data. This improves both accuracy and understanding of how the transform scalar is affected by the input data.
[0015] According to one embodiment of the present invention, the transformation can be performed using a transformer architecture or an attention mechanism. In the context of this specification, the transformer architecture is specifically understood to be an architecture comprising an encoder connected in series and a decoder connected in series. The encoder and the decoder may have different dimensions. In this embodiment, convolution can be specifically omitted. Actual tests have shown that this embodiment produces particularly realistic results.
[0016] According to one embodiment of the invention, the input data may be configured as an input scalar. The transformed data may be configured as a first embedded signal and a second embedded signal. The first embedded signal may include, for example, an indication of the position and movement of a person's body or body part, and the second embedded signal may include an indication of the position and movement of a component of a motor vehicle or a motor vehicle. In the context of this specification, an embedded signal is specifically understood to be a signal that does not appear outside the artificial neural network. Practical tests have shown that this embodiment produces particularly realistic results.
[0017] According to one embodiment of the present invention, both the first embedded signal and / or the second embedded signal may be multi-dimensional.These signals may respectively include several values in several dimensions, for example 32 or more dimensions.
[0018] According to one embodiment of the present invention, the transformed data in the simulation can only affect the current or future state of the simulation. For example, this can be achieved through the structure of an artificial neural network. For example, the neurons of the artificial neural network can only receive information and / or data from the past or present.
[0019] According to one embodiment of the invention, the transformed data can be input into an input layer of an artificial neural network. This is particularly advantageous if the artificial neural network has an architecture as described in DE 10 2020 132 042 A1. Practical tests have shown that this embodiment produces particularly realistic results.
[0020] According to one embodiment of the present invention, the transformed data may be input into different layers of the artificial neural network, specifically hidden layers, and / or used at different times during the execution of the simulation. For example, a first subset of the transformed data may be used in a first layer and / or at a first time during the execution of the simulation, while a second subset of the transformed data may be used in a second layer and / or at a second time during the execution of the simulation.
[0021] It is also particularly advantageous if the artificial neural network has an architecture as described in DE 10 2020 132 042 A1. This embodiment can also produce particularly realistic results.
[0022] The execution of the simulation generates information and / or data which can be used, for example, to improve the accident safety of the motor vehicle by modifying one or more components. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Other features and advantages of the present invention become apparent from the following description of preferred design examples with reference to the accompanying drawings. The same reference numerals are used for the same or similar features and features having the same or similar functions. The accompanying drawings show:
[0024] Figure 1 is a schematic diagram of an artificial neural network according to one embodiment of the present invention;
[0025] Figure 2 is a schematic diagram of an artificial neural network according to one embodiment of the present invention; and
[0026] Figure 3 is a schematic diagram of an artificial neural network according to one embodiment of the present invention. DETAILED DESCRIPTION
[0027] Figure 1 The artificial neural network 1 shown in comprises an input layer 3, a plurality of hidden layers 4, 5 and 6, and an output layer 2. Each of the layers 2-6 comprises a plurality of artificial neurons shown as circles.
[0028] exist Figure 1 In , t represents the time that has passed during the execution of the simulation. In the direction from the input layer 3 via the hidden layers 4-6 to the output layer 2, data is transmitted exclusively from neurons to neurons in the corresponding next layer that simulate the simulation at the same time point or simulate the simulation at a later time point. This ensures that later time points in the simulation do not affect earlier time points in the simulation.
[0029] For example, if data relating to the movement and / or position of a motor vehicle or a part of a motor vehicle are used as input data and information on the movement and / or position of a human body or a part of a human body is output in the output layer 2, a later state of a part of a motor vehicle or of a motor vehicle will not affect an earlier state of a part of a body or of a body. Thus, the simulation is performed in a more physically realistic manner.
[0030] Figure 2 The artificial neural network shown in comprises an input layer 7, a plurality of hidden layers 8, 9 and 10 and an output layer 11. For example, the artificial neural network can be constructed as described in DE 10 2020 132 042 A1. The input data 14 are transformed into transformed data 12 in a dense layer. The transformed data 12 are input into the input layer 7 of the artificial neural network together with a one-dimensional vector 13. For example, the one-dimensional vector 13 can relate to the position and / or movement of a motor vehicle or a part of a motor vehicle. The processing of the one-dimensional vector 13 and the transformed data 12 can be carried out as described in DE 10 2020 132 042 A1. By inputting the transformed data 12 into the input layer 7, a particularly realistic simulation is achieved.
[0031] The output data 11 output by the artificial neural network may for example include an indication of the severity of the simulated accident. For example, this may be a one-dimensional vector. The severity of the accident may for example be indicated by the severity of the injury to a body part.
[0032] Figure 3 The artificial neural network shown in is similar to Figure 2 . However, the transformed data 12 are not input into the input layer 7, but into different hidden layers 9 and 15. In some cases, depending on the type of accident being simulated, this can enable a particularly realistic simulation, since the transformed data 12 can be input where they have a particularly high influence on the realism of the simulation.
Claims
1. A computer-implemented method for simulating an accident of a motor vehicle using an artificial neural network (1), comprising the following steps: - receiving input data (14) related to the incident; - transforming the input data (14) into transformed data (12); and - using the transformed data (12) to simulate the accident with the aid of the artificial neural network (1).
2. The method according to claim 1, characterized in that The input data (14) is configured as an input signal, wherein the transformed data (12) includes a scalar, wherein the scalar is converted into a transformed signal before the simulation, and wherein the simulation is performed using the transformed signal.
3. The method according to claim 1, characterized in that The input data (14) is configured as an input scalar.
4. The method according to claim 3, characterized in that The transform data (12) is configured as a single transform scalar.
5. The method according to claim 1, characterized in that The transformation is performed using a transformer architecture or an attention mechanism.
6. The method according to claim 1, characterized in that The input data (14) is configured as an input scalar, wherein the transformed data (12) is configured as a first embedded signal and a second embedded signal.
7. The method according to claim 6, characterized in that The first embedded signal and / or the second embedded signal are both multi-dimensional.
8. The method according to any one of the preceding claims, characterized in that In the simulation, the transformation data (12) affects only the current or future state of the simulation.
9. The method according to any one of the preceding claims, characterized in that The transformed data (12) is input into the input layer (3; 7) of the artificial neural network (1).
10. The method according to any one of claims 1 to 8, characterized in that The transformation data (12) are input into different layers (9; 15) of the artificial neural network (1) and / or are used at different times during execution of the simulation.