Determining deformation parameters for a crash test by means of machine learning

Abstract

The invention relates to a computer-implemented method and system for indirectly measuring, by means of a trained machine learning model, deformation parameters for a crash test of a vehicle to be analyzed, having the following work steps: • detecting geometry parameters of a plurality of different vehicles, the parameters being measured by means of a sensor; • calculating a location probability distribution for each component and / or assembly, in particular each crash-relevant component and / or assembly, within a vehicle configuration space of the vehicle to be analyzed, the location probability distribution being derived from the detected geometry parameters; • determining boundary conditions with respect to the vehicle structure, in particular the arrangement of the components and the rigidities thereof, of the vehicle to be analyzed; • determining the vehicle structure, which is characterized by at least one geometry parameter, of a vehicle to be analyzed on the basis of the boundary conditions and the location probability distribution; • detecting at least one material parameter of individual portions of the vehicle structure, said parameters characterizing the mass, plastic properties, and elastic properties, in particular moduli of elasticity and flow curves, of the individual portions of the vehicle structure, the at least one geometry parameter and the at least one material parameter forming vehicle data relating to the vehicle to be analyzed; • calculating output data means of the machine learning model on the basis of input data, the input data being based on the vehicle data and the output data comprising a value for at least one first deformation parameter for the intrusion into the vehicle to be analyzed; and • outputting the at least one value for the first deformation parameter. The invention further relates to a computer-implemented method and a system for training the machine learning model.

Description

[0001] DETERMINATION OF DEFORMATION PARAMETERS FOR A CRASH TEST USING MACHINE LEARNING

[0002] The invention relates to a computer-implemented method and system for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed using a trained machine model. Furthermore, the invention relates to a computer-implemented method and system for training such a machine model.

[0003] Crash tests are crucial for vehicle safety and reliability. Analyzing deformation patterns during a crash event helps us understand the behavior of vehicle bodies under stress and minimize the risk of occupant injury through appropriate structural measures. However, traditional methods for measuring deformation parameters are complex, costly, and time-consuming, often resulting in the destruction of test specimens and equipment such as sensors. Vehicle manufacturers are therefore striving to minimize the number of actual crash tests and replace them with indirect measurement methods wherever possible.

[0004] Existing indirect measurement methods require significant effort for their application, including the geometric representation of vehicle components, the determination of their materials and mechanical properties, and the analysis of component joining techniques to ensure good predictive accuracy. This necessitates detailed knowledge, particularly of vehicle geometry, and appropriately trained development personnel for implementation. The time required to obtain results is also comparatively high (more than 10 hours).

[0005] Especially in the early concept phase of vehicle development, general decisions about the vehicle structure must be made, which are motivated by boundary conditions from various requirements.

[0006] The computer-implemented method presented here enables concept development personnel to verify the crashworthiness of different vehicle concepts in real time with good predictive accuracy and to consider different boundary conditions at a very early development phase, without having to carry out a detailed model build using conventional measurement methods. WO 2023 / 154442 A1 discloses a method for crash simulation of a vehicle structure, comprising:

[0007] Simulating a crash using a complete model of the vehicle structure;

[0008] Identifying a critical impact area of ​​the vehicle structure based on the simulated impact of the complete vehicle structure model;

[0009] Generate, using the critical impact area, a substructure model of the vehicle structure that is smaller than the full vehicle structure model; and simulate the crash using the substructure model of the vehicle structure.

[0010] US 2020 / 0394278 A1 discloses a computer-implemented method for optimizing vehicle design with regard to safety, comprising the following steps: a) collecting design data sets containing values ​​of design parameters of existing car models and values ​​of parameters of related analyzed data; b) collecting safety data sets of safety parameter values ​​of said vehicle models; c) generating an artificial neural network such that values ​​from the design data sets are used for the input layer and values ​​from the safety data sets are used for the target output layer, so that the neural network learns how to get from input to output within predefined tolerance values;d) Creating an optimizer where the input is a new design data set to be used for the design of a new vehicle, defining ranges of possible values ​​for such a new data set, defining safety targets based on the safety parameters, and using the neural network as a simulator to find optimized values ​​for the new design data set with respect to the expected safety of the new vehicle;and e) based on such optimized values ​​of the new design dataset, performing a iterative process of freezing parameter values ​​and updating the design of the new vehicle until all optimized values ​​are implemented in such a new design. Methods for indirectly measuring deformation parameters for a crash test of a vehicle to be analyzed are known from THEL S. et al. "Introducing Finite Element Method Integrated Networks (FEM IN)", Computer Methods in Applied Mechanics and Engineering, Vol. 427, pages 1170-73, July 1, 2024, and from CN 117332501 A.

[0011] It is an object of the invention to improve the simulation of crash tests. In particular, it is an object of the invention to provide a measurement method for estimating or predicting deformation parameters in crash tests.

[0012] These tasks are solved by the independent claims. Advantageous embodiments are claimed in the dependent claims.

[0013] A first aspect of the invention relates to a computer-implemented method for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed using a trained machine model, comprising the following steps:

[0014] • Acquisition of vehicle data of the vehicle to be analyzed, wherein the vehicle data includes at least one geometric parameter that characterizes a vehicle structure and material parameters of individual sections of the vehicle structure, which are based on a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure;

[0015] • Calculating output data using the machine-engine model based on input data, wherein the input data is based on the vehicle data and wherein the output data includes a value of at least one first deformation parameter for the intrusion into the vehicle to be analyzed; and

[0016] • Output at least one value of the first deformation parameter.

[0017] A second aspect of the invention relates to a computer-implemented method for training a machine learning model, in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, comprising the following steps:

[0018] • Acquisition of vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter characterizing a vehicle structure and material parameters of individual sections of the vehicle structure, which are based on a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space;

[0019] • Configuring vehicle models based on vehicle data, wherein each configured vehicle model defines volume elements of the respective vehicle, which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter at a respective position of a volume element;

[0020] • Simulating a single crash load case for the configured vehicle models using a finite element method with volume elements, generating simulation data that includes at least one first deformation parameter for the intrusion into the vehicle, and repeating the crash load case simulation for differently configured vehicle models; and

[0021] • Training a machine learning model, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a respective vehicle are assigned to the simulation data of that vehicle, wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle to be analyzed as input data.

[0022] Preferably, the machine learning model used in the procedure according to the first aspect is trained using the procedure according to the second aspect.

[0023] A third aspect of the invention relates to a system for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, using a trained machine model, comprising:

[0024] • a first interface for acquiring vehicle data of the vehicle to be analyzed, wherein the vehicle data includes at least one geometric parameter that characterizes a vehicle structure and material parameters of individual sections of the vehicle structure, which are based on a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections of the vehicle structure;

[0025] • Means for calculating output data using the machine-engine model based on input data, wherein the input data is based on the vehicle data and wherein the output data includes a value of at least one first deformation parameter for the intrusion into the vehicle to be analyzed; and

[0026] • A second interface for outputting at least one value of the first deformation parameter.

[0027] A fourth aspect of the invention relates to a system for training a machine learning model, in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, comprising:

[0028] • a first interface for acquiring vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter characterizing a vehicle structure and material parameters of individual sections of the vehicle structure, which are based on a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space;

[0029] • Means for configuring vehicle models based on vehicle data, wherein each configured vehicle model defines volume elements of the respective vehicle, to which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter at a respective position of a volume element;

[0030] • Means for simulating a, in particular a single, crash load case for the configured vehicle models using a finite element method with volume elements, wherein simulation data is generated which includes at least one first deformation parameter for the intrusion into the vehicle, wherein the simulation is repeated for differently configured vehicle models; and • Means for training a machine learning model, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a respective vehicle are mapped to the simulation data of that vehicle, wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle to be analyzed as input data.

[0031] A fifth aspect of the invention relates to a computer-implemented classifier, in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, wherein the classifier is generated by training a classification algorithm, wherein the classification algorithm was configured by the following steps, which are performed for each training input of a plurality of training inputs:

[0032] • Acquisition of vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter characterizing a vehicle structure and material parameters of individual sections of the vehicle structure, which are based on a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space;

[0033] • Configuring vehicle models based on vehicle data, wherein each configured vehicle model defines volume elements of the respective vehicle, which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter at a respective position of a volume element;

[0034] • Simulating a, in particular a single, crash load case for the configured vehicle models using a finite element method with volume elements, wherein simulation data is generated which includes at least one first deformation parameter for the intrusion into the vehicle, wherein the simulation of the crash load case is repeated for differently configured vehicle models; and • Training a machine learning model to generate the classifier, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a respective vehicle are mapped to the simulation data of that vehicle, and wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle to be analyzed as input data.

[0035] A sixth aspect of the invention relates to a computer program or computer program product, wherein the computer program or computer program product contains instructions, in particular stored on a computer-readable and / or non-volatile storage medium, which, when executed by a computer, cause the computer to perform the steps of a method according to the first aspect and / or according to the second aspect.

[0036] Other aspects of the invention are:

[0037] A computer-implemented method for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, using a trained machine model, comprising the following steps:

[0038] • Capturing geometric parameters of a large number of different vehicles measured by a sensor;

[0039] • Calculating a probability distribution for each component and / or assembly, especially crash-relevant ones, within a vehicle configuration space of the vehicle to be analyzed, where the probability distribution is derived from the recorded geometry parameters;

[0040] • Determining boundary conditions relating to the vehicle structure, in particular an arrangement of the components and their stiffnesses, of the vehicle to be analyzed;

[0041] • Determining a vehicle structure, characterized by at least one geometric parameter, of a vehicle to be analyzed based on the boundary conditions and the probability distribution; • Determining at least one material parameter of individual sections of the vehicle structure, which characterize a mass as well as plastic and elastic properties, in particular moduli of elasticity and flow curves, of the individual sections of the vehicle structure, wherein the at least one geometric parameter and the at least one material parameter constitute vehicle data of the vehicle to be analyzed;

[0042] • Calculating output data using the machine-engine model based on input data, wherein the input data is based on the vehicle data and wherein the output data includes a value of at least one first deformation parameter for the intrusion into the vehicle to be analyzed; and

[0043] • Output at least one value of the first deformation parameter.

[0044] A computer-implemented method for training a machine learning model, in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, comprising the following steps:

[0045] • Acquisition of vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter measured by means of a sensor, which characterizes a vehicle structure, and material parameters of individual sections of the vehicle structure, which characterize a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space;

[0046] • Configuring vehicle models based on vehicle data, wherein each configured vehicle model defines volume elements of the respective vehicle, to which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter of the individual sections of the vehicle structure, at a respective position of a volume element, wherein the sections of the vehicle structure of each vehicle are mapped by geometric volume areas in a vehicle configuration space, wherein dimensions and / or positions of the geometric volume areas in the vehicle configuration space are derived from the vehicle data, wherein an assignment of the volume elements to the respective sections of the vehicle structure is defined by the position of the volume elements in relation to the geometric volume areas;• Simulating a crash load case for the configured vehicle models using a finite element method with the volume elements, generating simulation data which includes at least one first deformation parameter for the intrusion into the vehicle, and repeating the crash load case simulation for differently configured vehicle models; and;

[0047] • Training the machine learning model, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a respective vehicle are assigned to the simulation data of that vehicle, wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle to be analyzed, measured at least partially by means of a sensor, as input data.

[0048] A system for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, using a trained machine model, comprising:

[0049] • a first interface for capturing geometric parameters of a large number of different vehicles measured by means of a sensor and of at least one material parameter of individual sections of the vehicle structure, which characterize a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure, wherein the at least one geometric parameter and the at least one material parameter constitute vehicle data of the vehicle to be analyzed;

[0050] • Means for calculating a probability distribution for each component and / or assembly, particularly crash-relevant components, within a vehicle configuration space of the vehicle to be analyzed, wherein the probability distribution is derived from the acquired geometric parameters, for determining boundary conditions relating to the vehicle structure, in particular an arrangement of the components and their stiffnesses, of the vehicle to be analyzed, for determining a vehicle structure characterized by at least one geometric parameter of a vehicle to be analyzed based on the boundary conditions and the probability distribution, and for calculating output data using the machine model based on input data.wherein the input data are based on the vehicle data and wherein the output data include a value of at least one first deformation parameter for the intrusion into the vehicle to be analyzed; and,

[0051] • A second interface for outputting at least one value of the first deformation parameter.

[0052] A system for training a machine learning model, in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, comprising:

[0053] • a first interface for acquiring vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter measured by means of a sensor, which characterizes a vehicle structure, and material parameters of individual sections of the vehicle structure, which characterize a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space;

[0054] • Means for configuring vehicle models based on vehicle data, wherein each configured vehicle model defines volume elements of the respective vehicle, to which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter of the individual sections of the vehicle structure at a respective position of a volume element, wherein the sections of the vehicle structure of each vehicle are mapped by geometric volume regions in a vehicle configuration space, wherein dimensions and / or positions of the geometric volume regions in the vehicle configuration space are derived from the vehicle data, wherein an assignment of the volume elements to the respective sections of the vehicle structure is defined by the position of the volume elements in relation to the geometric volume regions;

[0055] • Means for simulating a crash load case for the configured vehicle models using a finite element method with the volume elements, wherein simulation data is generated which includes at least one first deformation parameter for the intrusion into the vehicle, wherein the simulation is repeated for differently configured vehicle models; and

[0056] • Means for training the machine learning model, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a respective vehicle are assigned to the simulation data of that vehicle, wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data, based on vehicle data of a vehicle to be analyzed measured at least partially by means of a sensor as input data.

[0057] A computer-implemented classifier, in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, wherein the classifier is generated by training a classification algorithm, wherein the classification algorithm was configured by the following steps, which are performed for each training input of a multitude of training inputs:

[0058] • Acquisition of vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter characterizing a vehicle structure and material parameters of individual sections of the vehicle structure, characterizing a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space;

[0059] • Configuring vehicle models based on vehicle data, wherein each configured vehicle model defines volume elements of the respective vehicle, to which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter at a respective position of a volume element, wherein the sections of the vehicle structure of each vehicle are mapped by geometric volume regions in a vehicle configuration space, wherein dimensions and / or positions of the geometric volume regions in the vehicle configuration space are derived from the vehicle data, wherein an assignment of the volume elements to the respective sections of the vehicle structure is defined by the position of the volume elements in relation to the geometric volume regions;

[0060] • Simulating a single crash load case for the configured vehicle models using a finite element method with volume elements, generating simulation data that includes at least one first deformation parameter for the intrusion into the vehicle, and repeating the crash load case simulation for differently configured vehicle models; and

[0061] • Training a machine learning model to generate the classifier, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a given vehicle are associated with the simulation data of that vehicle, and wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle to be analyzed as input data.

[0062] Preferably, the stiffnesses and moments of inertia of the sections depend on the materials used and on the dimensions of the structural sections.

[0063] Preferably, the vehicle data is at least partially measurement data or derived from measurement data.

[0064] Preferably, the vehicle model is a fully parameterized model with respect to the vehicle data, in particular the geometry parameters and the material parameters.

[0065] Preferably, when simulating the crash load case, a table is generated in which values ​​of the parameters of the vehicle data of a respective vehicle are assigned to values ​​of the parameters of the simulation data, wherein the vehicle data and the generated simulation data are provided to the machine learning model by means of the table.

[0066] An indirect measurement within the meaning of the present disclosure preferably involves determining a desired measured value of a parameter using other available information from a system to be measured. More preferably, the desired measured value of a parameter is determined from at least one other physical parameter.

[0067] A machine learning model as defined in the disclosure is preferably based on a polynomial algorithm or an artificial neural network. Preferably, the machine learning model is a metamodel of a plurality of vehicle models. A crash load case as defined in the disclosure is preferably a frontal crash (in accordance with existing test regulations, e.g., (Euro) NCAP or II HS) against a rigid wall or a frontal crash against an offset deformable barrier (ODB), or any other arbitrary crash configuration.

[0068] An intrusion within the meaning of the present disclosure is preferably defined by the degree of vehicle structural deformation, represented as the penetration depth into the vehicle at a defined location of the vehicle structure.

[0069] A deformation parameter within the meaning of the disclosure is preferably a standardized measured quantity. Preferably, such a parameter indicates a directly measurable deformation, in particular as a displacement of nodes within a vehicle coordinate system, an acceleration, or a parameter derived from the acceleration, in particular a deformation.

[0070] A vehicle structure within the meaning of the disclosure preferably refers to the construction and arrangement of the various assemblies and components that make up a vehicle. This further preferably includes the frame, the body, the chassis, the wheel suspension, the drivetrain, and other structural elements. The vehicle structure is preferably a three-dimensional configuration.

[0071] A geometric parameter within the meaning of the disclosure preferably specifies dimensions of the vehicle structure, its assemblies and / or the individual components. The geometric parameters further preferably include at least the dimensions of the primary energy-absorbing structures (length and cross-section), of longitudinal beams and chassis frames (subframes) and the upper load path.

[0072] A vehicle model within the meaning of the disclosure is a simulation model.

[0073] A finite element method using volume elements as defined in the disclosure is preferably a numerical technique for solving partial differential equations, in which the domain to be solved is divided into volume elements in which the fundamental equations are discretized. In particular, by applying conservation laws to these volume elements and assuming a specific discretization for the differential equations, algebraic equations for the unknowns within each volume element are obtained. Preferably, these equations are solved to obtain numerical approximate solutions for the desired variables across the entire domain. The finite element method exhibits conservation properties and can handle complex geometric configurations. In particular, the volume elements exhibit different material properties depending on their position.The finite element method is preferably used in various fields of engineering and physics, especially in fluid mechanics, heat transfer and other areas of continuum mechanics.

[0074] Preferably, the volume elements are cubes arranged at right angles. Furthermore, preferably, the element edge length depends on the specific application and is chosen to represent all characteristic vehicle features while simultaneously enabling minimal computation times.

[0075] Recording, as defined in the disclosure, preferably means storing data, at least temporarily.

[0076] For the purposes of this disclosure, data acquisition preferably involves reading measured operating data via a data interface. Alternatively or additionally, data acquisition includes determining a measurement signal using a sensor and / or post-processing a measurement signal to generate vehicle data.

[0077] Output, as defined in the disclosure, preferably means providing information. Preferably, this provision occurs via a data interface. However, it is also possible to output information via a user interface, such as a screen.

[0078] In the context of the disclosure, "connections with respect to the volume elements" preferably means that the volume elements used in the finite element method are directly connected to each other at the nodes that form the vertices of the volume elements.

[0079] A means as defined in the disclosure can be configured in hardware and / or software and, in particular, comprises a processing unit, preferably a microprocessor (CPU), preferably connected to a storage and / or bus system via data or signals, and / or one or more programs or program modules. The CPU can be configured to execute instructions implemented as a program stored in a storage system, to acquire input signals from a data bus, and / or to output signals to a data bus. A storage system can comprise one or more storage media, in particular different types, especially optical, magnetic, solid-state, and / or other non-volatile media. The program can be designed to embody the methods described herein.is capable of executing such procedures, so that the CPU can perform the steps of such procedures and thus, in particular, analyze at least one technical device or train a classification algorithm.

[0080] The term "means" as used herein encompasses all structures, materials, or actions set forth herein, as well as all equivalents thereof. Furthermore, the structures, materials, or actions and their equivalents include everything described in the abstract, the brief description of the figures, the detailed description, the summary, and the claims themselves. A system and / or its means may preferably take the form of a pure hardware variant, a pure software variant (including firmware, resident software, microcode, etc.), or a combination of software and hardware aspects, generally referred to as a "circuit," "module," or "system." Any combination of one or more computer-readable media may be used. The computer-readable medium may be a computer-readable signaling medium or a computer-readable storage medium.

[0081] The systems and methods according to the present disclosure can preferably be implemented in conjunction with a suitably configured computer, a programmed microprocessor or microcontroller and one or more peripheral integrated circuit elements, an ASIC or other integrated circuit, a digital signal processor, a hard-wired electronic or logic circuit, such as a circuit with discrete elements, a programmable logic device or gate arrangement, such as a programmable logic device (PLD), a programmable logic array (PLA), a field-programmable gate arrangement (FPGA), a programmable logic arrangement (PAL), or a comparable means.In general, any device or means capable of implementing the methodology presented herein may be used to implement the various aspects of this disclosure. Exemplary hardware includes computers, handheld devices, telephones (e.g., cellular, internet-enabled, digital, analog, hybrid, and others), and other hardware known in engineering. Some of these devices include processors (e.g., a single or multiple microprocessors), memory, non-volatile memory, input devices, and output devices. Furthermore, alternative software implementations, including but not limited to distributed processing or distributed processing of components / objects, parallel processing, or processing by virtual machines, may be developed to implement the procedures described herein.

[0082] The invention is based on the idea of ​​determining the deformation of a vehicle based on its geometry and the materials used in the vehicle. For this purpose, a machine learning model is trained using input data from a simulation and simulation results from a crash simulation. A finite element method with volume elements is used, i.e., a finite element method in which three-dimensional volumes, instead of two-dimensional surfaces, are used to model the simulated objects. The input data for the simulation includes at least the geometry and material properties of the vehicle structure. The simulation results from the crash simulation are a deformation of the vehicle structure, in particular a so-called intrusion of the vehicle.

[0083] The present invention provides an indirect measurement method for estimating deformation parameters in crash tests using computer-aided techniques. This approach enables fast and sufficiently accurate estimations of deformation parameters without requiring significant expertise in advanced algorithms or computer-aided design software.

[0084] The invention enables the calculation of deformation parameters for a wide variety of vehicles and crash scenarios without the need to create new models. The invention allows for the instantaneous calculation and / or prediction of deformation parameters, which is advantageous in the early development phase of vehicle types, even when structural details of the vehicle are not yet available, or when optimizing existing vehicle types. Overall, the computer-implemented method offers an efficient and accurate approach for predicting deformations in crash scenarios.

[0085] The trained machine learning model can predict the effects of crash loads on the vehicle, such as the strain, torsion, compression, or indentation of the vehicle structure. Simulation allows for the testing of alternative design solutions to improve crash performance or explore design limits. The model's ability to predict crash characteristics can reduce the number of necessary real-world crash tests, thereby lowering costs. Simulation also enables the early identification of potential weaknesses in the vehicle design during the development process, allowing for the implementation of safety-related improvements.

[0086] The modeling approach for the vehicle to be simulated is based on an initial model consisting of a cuboid-shaped area completely filled with volume elements of predefined edge lengths, forming a vehicle configuration space. The volume elements belonging to a configured vehicle model constitute the finite elements (FE) of the finite element method used for simulation. Because the volume elements are interconnected, the boundary conditions at the transitions between them are uniquely defined.

[0087] In an advantageous embodiment, the procedure, according to the first aspect, further comprises the following work steps:

[0088] • Reading in vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter characterizing a vehicle structure and material parameters of individual sections of the vehicle structure, based on mass as well as plastic and elastic properties, in particular moduli of elasticity and flow curves of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space; • Calculating a probability distribution for each component and / or assembly, in particular crash-relevant components, within a construction space of the vehicle to be analyzed, wherein the probability distribution is derived from the vehicle data;

[0089] • Calculating boundary conditions relating to the vehicle structure, in particular the arrangement of components and their stiffnesses, of the vehicle to be analyzed; and

[0090] • Determining the vehicle structure of a vehicle to be analyzed based on the boundary conditions.

[0091] This allows the vehicle structure of a vehicle under analysis to be determined based on known vehicle concepts. Not only the most probable position, but also the size of assemblies and components can be taken into account when vehicle data from databases is used. In particular, such a database can contain standard market contours of components. Based on these, package concepts—i.e., fundamental arrangements of assemblies—and limits / specifications for component stiffness can be determined for a vehicle under analysis.

[0092] A location probability distribution as defined in the disclosure is preferably a 3D density distribution, in particular with a resolution of about 20x20x20mm.

[0093] In a further advantageous embodiment, the method, according to the first aspect, further comprises the following step:

[0094] • Capturing a crash load case with respect to the vehicle data, wherein the crash load case is provided to the machine learning model as a further input parameter and wherein the output data includes a value of at least one first intrusion parameter for the intrusion into the vehicle to be analyzed in this crash load case.

[0095] This makes it possible to use a single trained machine learning model to predict vehicle structure deformations for a multitude of crash load cases. This reduces the number of machine learning models that need to be trained and stored. It also makes it possible to use a single machine learning model to predict vehicle structure deformations for a wide variety of possible vehicle data configurations. Due to the large number of variable vehicle data points accessible with this method, the deformation parameters of a crash load case can be predicted for a wide range of vehicle structure types with only a few trained and stored machine learning models. For example, vehicle structure types from the A-Class to the D-Class with different engine package configurations can be covered with just a few machine learning models.

[0096] In an advantageous embodiment of the method according to the first aspect and the second aspect, the sections of the vehicle structure of each vehicle are formed by geometric volume regions in a vehicle configuration space, wherein the dimensions and positions of the geometric volume regions in the vehicle configuration space are derived from the vehicle data.

[0097] A vehicle configuration space as defined in the disclosure is preferably a virtual, cuboid volume area that is limited by the maximum external dimensions of a vehicle.

[0098] The values ​​of the geometric parameters, together with the values ​​of the material parameters, in particular with the mechanical stiffness information, preferably describe a point of the vehicle configuration space for a section of the vehicle structure. Furthermore, the geometric volume regions preferably completely fill the vehicle configuration space.

[0099] Geometric volume regions within the meaning of the present disclosure are preferably cuboid-, cylinder- or prism-shaped geometric bodies, defined by the positions of their vertices in the three-dimensional coordinate system of the vehicle configuration space. In particular, the geometric volume regions can also be formed by combining several simple geometric bodies or by means of further geometric data provided via databases.

[0100] The geometric volume regions allow the vehicle data of any configuration to be represented as a 3D model, particularly in a software-independent manner. The geometric volume regions, whose dimensions and positions are derived from the vehicle data, are superimposed on the volume elements in the vehicle configuration space.

[0101] Using a vehicle configuration space is particularly advantageous for the modeling approach of a finite element method with volume elements.

[0102] In an advantageous embodiment of the method according to the first aspect, when configuring the vehicle models, the geometric volume areas and / or the intermediate volume elements are each represented with a substitute material, which is determined on the basis of the material properties of the respective sections of the vehicle structure, in particular the component or components of the respective section, and / or the respective space.

[0103] The material properties are preferably transformed into generic material properties, whereby the volume areas and spaces / connection technology between the sections of the vehicle structure represented by volume elements reflect the real deformation properties.

[0104] Preferably, each vehicle model defines geometric volume regions with a substitute material based on the real material properties of the respective section of the vehicle structure, in particular the component or components of the vehicle contained in that section of the vehicle structure.

[0105] This allows the substitute material to be easily changed as a parameter. Furthermore, fewer individual materials with their respective properties need to be considered. Using substitute materials simplifies the geometric volume regions. In particular, the resolution of the material structure is reduced compared to the material parameters of the individual sections of the vehicle structure. The geometric volume regions are preferably assigned substitute materials from a modular system.

[0106] By using a substitute material for geometric volume regions, the simulation model can represent the vehicle being simulated with greater numerical stability. Numerically stable means that the simulation models are less sensitive to extreme deformations, which inevitably occur with parameter variations, especially those related to material stiffness. The properties of the vehicle structure sections, particularly the components of the respective section, do not need to be reconstructed in detail. This reduces the complexity of the simulation model, and these details are usually not precisely known in an early concept phase. Furthermore, the use of the substitute material enables a better simulation of the global deformation behavior of the materials and structures.This results in a better-trained machine learning model by enabling the generation of training data across a broader range of variations. With conventional, more detailed models, simulation failures are more frequent when larger material parameter variations occur, thus reducing the amount of usable training data. Additionally, the increased simulation effort in conventional, more detailed models limits the number of computable configurations.

[0107] In a further advantageous embodiment of the method according to the second aspect, the allocation of the volume elements to the respective sections of the vehicle structure is defined by the position of the volume elements of the finite element method in relation to the geometric volume areas.

[0108] This allows for a clear determination of which component a volume element or which cavity an intermediate volume element is assigned to. The material parameters to be assigned to the individual volume elements can then be determined.

[0109] From a three-dimensional arrangement of geometric volume regions, those volume elements of the vehicle configuration space are identified that correspond to the respective geometric volume regions. These volume elements are then assigned the associated material parameters of the corresponding sections of the vehicle structure. This results in a coherent finite element vehicle structure that corresponds to the real vehicle structure described by the vehicle data. Additional finite elements or correction elements representing connecting elements such as screws, weld points, weld seams, etc., are not required. Therefore, compared to classical finite element methods, the complex positioning of such additional finite elements or correction elements is also eliminated.

[0110] In a further advantageous embodiment of the method according to the second aspect, each vehicle model defines spaces and / or components between the sections of the vehicle structure of the respective vehicle by means of intermediate volume elements with a substitute material, wherein the substitute material preferably has a lower stiffness and / or a higher compressibility than real material properties of the components of the respective vehicle.

[0111] Gaps can be cavities in the vehicle structure, in particular an engine compartment, or hollow components, in particular made of sheet metal, of the vehicle structure, in particular longitudinal beams.

[0112] Replacement materials as defined in this disclosure preferably exhibit generic material properties. These replacement materials preferably allow for greater deformation than the actual mechanical material properties of the components, taking into account the effects of the internal structure of the depicted sections of the vehicle structure.

[0113] The replacement materials are predefined and designed to be scalable for both components, especially structural components, and for spaces between them.

[0114] The stiffness of the replacement material for a cavity preferably depends on the package density of the cavity with components that are not represented by their own volume elements. Preferably, these are components that can no longer be explicitly represented with the selected dimensions of the volume elements, in particular connection technology, hoses, pipes, brackets, and / or cables.

[0115] By using a substitute material for intermediate volume elements, the simulation model can represent the vehicle being simulated with exceptional reliability, without having to reconstruct or define the properties of the cavities and small components in detail. This reduces the complexity of the simulation model, and these details are usually not precisely known in an early concept phase. Furthermore, their use enables a better simulation of the global deformation behavior of the materials and structures. This results in a better trained machine learning model.

[0116] In a further advantageous embodiment of the method according to the second aspect, the intermediate volume elements additionally compensate for a missing contribution to the stiffness of the vehicle components that are not considered with their own volume elements.

[0117] This could be, for example, a body part, especially a front flap / hood, which is not shown, but whose stiffness contribution is essentially taken into account by the replacement material.

[0118] Even cavities and relatively small components can contribute to the stiffness of a vehicle structure. Considering these therefore increases the accuracy of the simulation and thus of the machine learning model trained with it. Conversely, if the parameter range within which the stiffness is allowed to vary is known, the combined influence of many small components can be taken into account in order to derive limit values ​​for "filling" these cavities with components such as connectors, cables, etc.

[0119] The determination of the generic, scalable substitute materials that are assigned to the volume elements is carried out in separate variation calculations.

[0120] In a further advantageous embodiment, the procedure, according to the second aspect, therefore further comprises the following work steps:

[0121] • Simulating the deformation behavior of reference components, for example a rectangular hollow profile, such as a vehicle longitudinal beam, with varying dimensions and wall thicknesses, using a conventional, more detailed FE model for several elementary load cases, in particular axial compression, oblique compression (off-axis load), or 3-point bending tests, generating reference simulation data; and

[0122] • Validating the deformation behavior of the equivalent, but abstracted, representation of the components using volume elements, as characterized by the simulation data, with the reference simulation data. The associated stiffness scaling parameters are also among the vehicle data to be varied within the framework of the variational analysis.

[0123] The validated generic materials make it possible to realistically represent the global component deformation properties for a wide variety of reference cross-sections, sheet thicknesses or materials in the highly abstracted vehicle models via scaling parameters.

[0124] By translating the real material properties of 3D-formed two-dimensional sheet metal structures into generic properties of substitute materials for volume elements, greater abstraction is possible, particularly by selecting larger element edge lengths for the connected volume elements in the vehicle models. This allows for faster simulation times combined with higher numerical stability.

[0125] In a further advantageous embodiment, the procedure, according to the second aspect, further comprises the following work steps:

[0126] • Simulating the crash load case for some of the configured vehicle models, which represent actual existing vehicles, using a finite element method with surface elements, generating FE simulation data for each vehicle, which includes at least one first intrusion parameter for intrusion into the vehicle, and / or

[0127] • Import of measurement data from a real crash test, which includes at least one first intrusion parameter for intrusion into the vehicle; and

[0128] • Validating the configured vehicle models by comparing the simulation data with the FE simulation data and / or the measurement data.

[0129] Validation allows the selected properties of the substitute material to be verified. In particular, the validation of abstract vehicle models, by comparing simulation results using detailed vehicle finite element models with surface elements and / or measurement data from real crash tests, can be performed for a much wider range of vehicle geometries than in state-of-the-art analyses, thanks to the possibility of automated model generation, comparatively flexible parameterization, and fast computation times. However, the number of geometries is small compared to the total training data that can be generated with the simulation model. The abstract vehicle simulation model is only validated for individual control points in the configuration space.

[0130] In a further advantageous embodiment, the method, according to the second aspect, further comprises the following step:

[0131] • Adapting the properties of the replacement material in the vehicle models; and / or

[0132] • Adding additional components to the vehicle models, in particular engine mounts, or adjusting the material parameters of the components in the vehicle models, in particular a body-in-white front bulkhead; wherein the simulation of crash load cases for the defined number of configured vehicle models is repeated using a finite element method with volume elements until a deviation between the simulation data on the one hand and the FE simulation data and / or the measurement data on the other hand falls below a predefined threshold.

[0133] This can increase the accuracy of the vehicle models and thus of the trained machine model.

[0134] In a further advantageous embodiment, the method, according to the second aspect, further comprises the following step:

[0135] • Selecting the vehicle data used to generate the vehicle models from the variation space using a statistical design of experiments.

[0136] This allows for an increase in the quality of the vehicle model with respect to the vehicles being simulated. This, in turn, leads to an improvement in the trained machine learning model. The very short simulation time of the vehicle models used with volume elements allows for more comprehensive statistical experimental designs than classical finite element models.

[0137] In a further advantageous embodiment of the procedure according to the second aspect, the statistical design of experiments includes an automated adjustment of the vehicle data, in particular the geometry parameters and material parameters, by means of predefined mathematical relationships in order to fill the variation space with geometrically meaningful related vehicle structures that simultaneously fulfill predefined installation space and spacing requirements.

[0138] Installation space specifications and spacing specifications within the meaning of the present disclosure are preferably characterized by boundary conditions of the vehicle data, such as dependencies between linked geometry parameters and / or minimum and maximum value limits.

[0139] The predefined mathematical relationships can be derived from the arrangement and already known information regarding the vehicle structure.

[0140] Preferably following the statistical design of experiments, the vehicle data is used to verify compliance with installation space and clearance specifications using mathematical relationships, in order to eliminate "non-practical" or "unbuildable" vehicles (a "clean-up procedure"). Experience shows that after this, only < 0.3% of the previously generated configurations remain. Therefore, the initial number of generated configurations is preferably > 8,000,000 to ensure sufficient training data. Predefining and considering geometric constraints, for example, prevents components from overlapping due to an unfortunate selection of geometry parameter values ​​in a vehicle model. Automated adjustments can reduce or completely eliminate the need for manual user intervention.

[0141] In a further advantageous embodiment of the method according to the second aspect, the machine learning model is additionally provided with the respective simulated crash load case during training, whereby the respective simulated load case is assigned to the simulation data and the vehicle data.

[0142] This allows machine models to be generated that are valid for a majority of crash load cases.

[0143] In a further advantageous embodiment of the method according to the second aspect, the simulation is repeated for different load cases. In a further advantageous embodiment of the method according to the second aspect, the configuration of vehicle models comprises converting the vehicle data, in particular the dimensions of vehicle components, into three-dimensional coordinates, and reconstructing crash-relevant vehicle components in a simplified three-dimensional cuboid representation, which forms the volume regions.

[0144] In a further advantageous embodiment of the method according to the first aspect or the second aspect, the geometry parameter characterizes a global vehicle dimension, in particular a vehicle length, vehicle width, wheelbase, track width, cross-sectional area of ​​the vehicle front, center of gravity height, and / or dimensions of at least one assembly, in particular a wheel size, and / or dimensions of a component, in particular an insert, a sill (area of ​​the body shell at the bottom of the side, below the door openings), a seat cross member (stiffening on the vehicle floor panel at the point where the seat is bolted on).

[0145] In a further advantageous embodiment of the method according to the first aspect or the second aspect, the vehicle structure comprises the following sections: body, subframe, wheel suspension, assemblies and / or components of the vehicle.

[0146] In a further advantageous embodiment of the method according to the first aspect or the second aspect, the geometric parameters and material parameters of modules formed from predefined sections of the vehicle structure are grouped into subgroups of the vehicle data, wherein the vehicle data includes at least one module type parameter, in particular a package type parameter, which describes a specific composition of geometric parameters and material parameters that are treated jointly in a respective module, and wherein, when configuring the vehicle models, the vehicle data assigned to a respective module are changed exclusively jointly, in particular by means of higher-level module scaling parameters and translation parameters, in order to generate different configurations within the variation space, in particular within the framework of a statistical design of experiments.This modular strategy for changing parameters has the advantage that the number of free parameters for variations when configuring vehicle models can be kept as low as possible.

[0147] By varying several parameters simultaneously when configuring vehicle parameters, vehicle models can be configured particularly easily and also adapted particularly easily during validation.

[0148] In a further advantageous embodiment of the method according to the first aspect or the second aspect, the vehicle data further include chassis-specific material parameters, in particular a B-point strength parameter, an A-point strength parameter, or a package type parameter and / or a package density parameter.

[0149] In a further advantageous embodiment of the method according to the first aspect or the second aspect, at least one deformation parameter is selected from the group of the following parameters:

[0150] Acceleration curve as a function of time or intrusion curve as a function of time, mean impulse, impulse duration, mean penetration depth, occupant load criterion, maximum impact distance, inhibition energy, impulse efficiency, cabin penetration depth, yaw rate.

[0151] In a further advantageous embodiment of the method according to one of the aspects, the configuration of the vehicle models includes the conversion of vehicle data, which defines a vehicle structure with various components, into geometric volume areas consisting of connected volume elements.

[0152] In a further advantageous embodiment according to one aspect of the method, the vehicle data in the variation space define the subsequent dimensions of the volume areas in the vehicle configuration space.

[0153] In a further advantageous embodiment of the method according to one of the aspects, the vehicle data further specify predefined volume ranges, which correspond in particular to an assembly of components, for example, an engine assembly. In a further advantageous embodiment of the method according to one of the aspects, the assignment of the volume elements to the respective sections of the vehicle structure is defined by the position of the volume elements in relation to the geometric volume ranges.

[0154] In a further advantageous embodiment of the method according to one of the aspects, when configuring the vehicle models, the geometric volume areas and / or intermediate volume elements, which lie between the geometric volume areas, are each represented with a substitute material, which is determined on the basis of the material properties of the respective sections of the vehicle structure, in particular the component or components of the respective section, and / or the respective space.

[0155] In a further advantageous embodiment of the method according to one of the aspects, the intermediate volume elements additionally compensate for a missing contribution to the stiffness of the vehicle components not considered with their own volume elements.

[0156] In a further advantageous embodiment of the procedure according to one of the aspects, it further comprises the following work steps:

[0157] • Simulating the crash load case based on vehicle data from a multitude of different vehicles using a finite element method with surface elements, whereby FE simulation data are generated for each vehicle, which include at least one deformation parameter for the intrusion into the vehicle, and / or

[0158] • Importing measurement data from a real crash test, which includes at least one deformation parameter for intrusion into the vehicle; and

[0159] • Validating the configured vehicle models by comparing the simulation data with the FE simulation data and / or the measurement data.

[0160] In a further advantageous embodiment of the procedure according to one of the aspects, it further includes the following step:

[0161] • Adjusting the properties of the replacement material, whereby the simulation of crash load cases for the defined number of configured vehicle models is repeated using a finite element method with volume elements until a deviation between the simulation data on the one hand and the FE simulation data and / or the measurement data on the other hand falls below a predefined threshold.

[0162] In a further advantageous embodiment of the procedure according to one of the aspects, it further includes the following step:

[0163] • Selection of vehicle data from the variation space used to generate the vehicle models by means of a statistical design of experiments.

[0164] In a further advantageous embodiment of the procedure according to one of the aspects, the statistical design of experiments includes an automated adjustment of the vehicle data, in particular the geometry parameters and material parameters, by means of predefined mathematical relationships in order to fill the variation space with geometrically meaningful related vehicle structures that simultaneously meet predefined installation space and spacing requirements.

[0165] In a further advantageous embodiment of the method according to one of the aspects, the machine learning model is additionally provided with the respective simulated crash load case during training, whereby the respective simulated load case is assigned to the simulation data and the vehicle data.

[0166] In a further advantageous embodiment of the method, according to one of the aspects, the simulation is repeated for different load cases. Further advantages and features will become apparent from the following description with reference to the figures. These figures show, at least partially schematically:

[0167] Figure 1: a functional block diagram of an embodiment of a system for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed;

[0168] Figure 2: a flowchart of an embodiment of a method for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed;

[0169] Figure 3: a functional block diagram of an embodiment of a system for training a machine tool model for the indirect measurement of deformation parameters for a crash test; Figure 4: a flowchart of an embodiment of a method for training a machine tool model for the indirect measurement of deformation parameters for a crash test;

[0170] Figure 5: an embodiment of a vehicle structure;

[0171] Figure 6: a view of an exemplary embodiment of a vehicle configuration space with geometric volume areas;

[0172] Figure 7: another view of the embodiment of the vehicle configuration space according to Figure 6; and

[0173] Figure 8: an enlarged view of section A from Figure 7.

[0174] The exemplary embodiments are explained below with reference to a passenger car 1. However, it is obvious to those skilled in the art that the teaching according to the invention is transferable to any other type of vehicle, in particular other motor vehicles, ships, aircraft and spacecraft. Furthermore, methods 100 and 200 are explained using a machine learning model. Here, too, it is known to those skilled in the art that any type of machine learning model can be used for methods 100 and 200.

[0175] In an application example of the teaching according to the invention, basic crash performance assessments based on initial, rough design information can be made possible in the early concept phase of a vehicle development with the help of a so-called configurator tool.

[0176] Based on existing database information on typical vehicle body styles, their package component arrangement and the associated crash performance ratings, the configurator tool is intended to predict the overall crash performance (OCP), expressed as a qualitative value between 1 and 10, for similar, freely selectable or freely scalable vehicle geometries.

[0177] This is achieved using the Machine Intelligence Model 2, which forms a so-called metamodel and which, in the final application, outputs the corresponding OCP value for a new vehicle concept, characterized by a specific set of component dimensions and package arrangements, without any noticeable delay, preferably in real time. Preferably, the crash performance prediction tool is a separate module of a more comprehensive vehicle configurator tool that also generates vehicle dynamics, powertrain, and performance predictions.

[0178] A computer-implemented method 100 for the indirect measurement of deformation parameters for a crash test of a vehicle 1 to be analyzed is described below with reference to Figures 1 and 2. Figure 1 shows a functional block diagram of a system 10 for executing a method 100. Figure 2 shows a flowchart of the method steps 100.

[0179] In a first step 101 of the process 100, vehicle data from a large number of different vehicles 1A, 1B, 1C are read in. The vehicle data characterizes at least one geometric parameter and material parameter of individual sections of a vehicle structure 3 of a vehicle 1 to be analyzed. The vehicle data can preferably be extracted from a so-called vehicle benchmark database, in particular of type a2mac1, and transferred to an Access database.

[0180] A material parameter preferably characterizes a mass as well as plastic and elastic properties, in particular E-moduli and flow curves of the individual sections of the vehicle structure.

[0181] A geometric parameter can be a global vehicle dimension, in particular a vehicle length, vehicle width, wheelbase, track width, cross-sectional area of ​​the vehicle front, or center of gravity height. Additionally or alternatively, geometric parameters can be dimensions of a vehicle structural assembly, for example, a wheel size and / or a component dimension, in particular an insert, a sill dimension, and / or a seat crossmember dimension.

[0182] The vehicle structure 3 preferably comprises various sections. Such sections include, for example, a body, subframe, wheel suspensions, assemblies with multiple components, and / or individual vehicle components. It is also preferably provided that the geometric and material parameters of defined modules of the vehicle structure 3, which form one or more sections, are grouped into subgroups of the vehicle data. These subgroups of the vehicle data for a module preferably form a single parameter, a so-called module type parameter or package type parameter.

[0183] In a second step 102, a probability distribution of the location of the crash-relevant components and / or assemblies within an installation space of the vehicle 1 to be analyzed is calculated, which is derived from the vehicle data of the large number of different vehicles 1A, 1B, 1C.

[0184] In a third step 103, boundary conditions relating to the vehicle structure 3 of the vehicle 1 to be analyzed are calculated. These boundary conditions can relate in particular to the arrangement of components or assemblies and their respective stiffness.

[0185] In a fourth step 104, the vehicle structure 3 of the vehicle 1 to be analyzed is finally determined on the basis of the calculated boundary conditions and the calculated probability distribution.

[0186] In a fifth step 105, vehicle data of the vehicle 1 to be analyzed is acquired. Preferably, this is accomplished via a data interface 11, as shown in Figure 1. Here, the vehicle data, as also shown in Figure 1, characterizes a vehicle structure 3 of the vehicle 1 to be analyzed. Furthermore, the vehicle data characterizes material parameters of individual sections 3A, 3B, 3C of the vehicle structure 3. These material parameters characterize, in particular, mass and / or plastic and elastic properties, such as moduli of elasticity and flow curves of the individual sections 3A, 3B, 3C of the vehicle structure 3. The vehicle structure 3 can be the one that was determined in the fourth step 104 described above. Preferably, the vehicle data, and thus also the vehicle structure 3, can be provided in another way.The vehicle data can be provided, for example, by user input or from a database. In a sixth step, a crash load case 9 is preferably recorded, as also shown in Figure 1. This crash load case 9 is preferably also provided to the machine learning model 2 as input data. In this case, the machine learning model 2 calculates the value of the at least one deformation parameter, in particular the intrusion, for this specific crash load case 9.

[0187] In a seventh step 107, the input data is provided to the machine learning model 2. This input data is based on, or is based on, the vehicle data. Preferably, the input data is furthermore based on, or is based on, a specific crash load case 9, as explained above. In the seventh step 107, the machine learning model 2 calculates output data that includes or characterizes at least one value of a first deformation parameter of an intrusion into the vehicle 1 to be analyzed. Such a deformation parameter can be a deformation of the vehicle structure 3 with respect to an initial state or an acceleration. Alternatively or additionally, the deformation parameters are time-dependent acceleration and / or intrusion curves.

[0188] Preferably, key performance indicators (KPIs) can be derived or calculated from the deformation parameters. Depending on the requirements, these KPIs are weighted and combined into a single value, the so-called Overall Crash Performance (OCP) value.

[0189] To perform the seventh work step 107, the system preferably comprises 10 computational means 12. To perform the first work step 101, the second work step 102, the third work step 103, and the fourth work step 104, the system 10 may preferably include further corresponding means, in particular means for reading data, means for calculating data, and means for determining data. More preferably, these means are formed by the computational means 12.

[0190] Finally, in an eighth work step 108, the value of the first deformation parameter is output, preferably via a second data interface 13.

[0191] The following section explains a method 200 for training an artificial neural network 2 as a machine learning model for the indirect measurement of deformation parameters for a crash test of a vehicle 1 to be analyzed, using Figures 3 and 4. Figure 3 shows a functional block diagram of a system 20 for executing the method 200, and Figure 4 shows a flowchart of the steps of the method 200 itself.

[0192] Steps 201 to 203 involve the automated creation of simplified, abstracted vehicle models using data from the Access database and / or a statistical design of experiments (for example, using the Latin Hypercube Sampling method). Step 204 involves simulating crash load cases 9 selected using experimental design with the abstracted vehicle models 4A, 4B, and 4C, and determining KPIs such as acceleration or intrusion. Steps 205 to 207 involve validating and adjusting the properties of the abstracted vehicle models 4A, 4B, and 4C using detailed finite element models. Based on the simulation results summarized in the simulation data, equivalence tables between features (input) and attributes (output) are preferably created.Step 208 involves training / fitting a metamodel 2 based on the simulation data, particularly the equivalence table data. Customer-specific boundary conditions (centers of gravity and restrictions of a variation space (customer-specific design space) for an equivalence table and a metamodel) can be taken into account. Using the trained metamodel 2, crash performance predictions can be made without the need for FE simulations, as described in Figures 1 and 2.

[0193] In a first step 201, vehicle data from a large number of different passenger cars 1A, 1B, 1C are recorded, the vehicle data comprising geometric parameters and material parameters. The geometric parameters characterize the vehicle structures 3 of the passenger cars 1A, 1B, 1C. The material parameters characterize individual sections of the vehicle structure 3 of the respective vehicle 1A, 1B, 1C and preferably specify a mass as well as plastic and elastic properties.

[0194] The vehicle data spans a variation space in which the parameters defined in the vehicle data, in particular the geometry parameters, the material parameters and the parameters already mentioned in relation to the description of Figures 1 and 2, can be changed.

[0195] Preferably, the system 20 for training the machine learning model 2 has a first interface 21 for acquiring vehicle data. This first interface 21 can be a data interface or one or more sensors or measuring devices by means of which the respective parameters contained in the vehicle data can be determined.

[0196] The state of the art includes various methods and devices for the three-dimensional acquisition of vehicles or vehicle components. These typically employ optical or tactile measuring instruments that enable a precise digital representation of the vehicle geometry. Established technologies include, in particular, portable or stationary 3D scanners, which in this case serve as the sensors for determining the vehicle data. Such 3D scanners are based on methods such as structured light projection, laser triangulation, or photogrammetry. Preferably, the sensor is therefore a 3D scanner.

[0197] Well-known suppliers of such measuring equipment offer, for example, high-resolution, high-speed handheld 3D scanners that allow complex vehicle contours to be digitized directly on the physical object, without the need for special reference marks or rigid fixtures. These measuring systems can be used to derive comparatively precise CAD data from physical objects. One example of a supplier of such measuring equipment is Creaform Inc.

[0198] Preferably, the geometric and material parameters of modules formed by a predefined section or sections of the vehicle structure are grouped into subgroups of the vehicle data. The vehicle data includes at least one module type parameter, in particular a package type parameter, which describes a specific composition of geometric and material parameters that are jointly processed in a given module. When configuring the vehicle models 4A, 4B, and 4C, the vehicle data assigned to each module is modified exclusively together, in particular by means of higher-level module scaling parameters and translation parameters. In a second step, 202, sets of values ​​are selected from the vehicle data using a statistical design of experiments.These sets of values ​​consist of value constellations of the various parameters, in particular geometry parameters, material parameters, module type parameters, module scaling parameters and / or translation parameters, and are finally used in a third step 103 to configure vehicle models 2.

[0199] In the third step 203, the vehicle models 4A, 4B, 4C are created. Here, the vehicle models are configured using the selected sets of parameter values ​​that span the configuration space. Each configured vehicle model 4A, 4B, 4C defines volume elements 5A, 5B, 5C, 5D of the respective vehicle 1A, 1B, 1C. These volume elements 5A, 5B, 5C, 5D are interconnected, and each volume element 5A, 5B, 5C, 5D is assigned a material property, in particular a stiffness, depending on the geometry parameter and the material parameter at a specific position of the volume element 5A, 5B, 5C, 5D. Accordingly, the system 20 provides means 22 for configuring vehicle models 4A, 4B, 4C.Configuring the vehicle models 4A, 4B, 4C includes, in particular, converting vehicle data, which defines a vehicle structure 3 with various components 3A, 3B, 3C, into geometric volume regions 6A, 6B, 6C consisting of connected volume elements 5A, 5B, 5C, 5D. This methodology for creating vehicle models 4A, 4B, 4C is explained below with reference to Figures 5 to 8.

[0200] Figure 5 shows a vehicle structure 3 of a vehicle with various components. Three components—two reinforcements 3B and 3C of the front section of a vehicle 1 and a radiator 3A—are shown as exemplary sections of the vehicle structure 3 and are labeled with reference numerals. Other sections of the vehicle structure 3 or components visible in Figure 5 are not labeled. For example, Figure 5 shows the front part of the vehicle frame and various engine components such as a car battery, as well as various cables and hoses. The cables extend through cavities in the engine compartment of the vehicle 1. A steering column in the passenger compartment is also shown in the upper right part of Figure 5. The various components of the vehicle 1 form the vehicle structure 3. Figure 6 shows an embodiment of a vehicle configuration space 7 with geometric volume regions.The vehicle configuration space 7 is a virtual, in particular cuboid, volume region bounded by the maximum external dimensions of the respective passenger car 1. The outer edges of the vehicle configuration space 7 are shown in Figure 7. Within the vehicle configuration space 7, the geometric volume regions 6A, 6B, 6C of various components are represented. As can also be seen in Figure 6, the geometric volume regions 6A, 6B, 6C are cuboid, cylindrical, or prismatic, oblique, or rectangular geometric bodies, which together abstractly represent the vehicle structure 3. The individual geometric volume regions 6A, 6B, 6C are preferably defined by the positions of their vertices in a three-dimensional coordinate system of the vehicle configuration space 7.As can also be seen from Figure 6, the geometric volume regions 6A, 6B, 6C can also be formed by a combination of several elementary geometric bodies. The dimensions and positions of the geometric volume regions 6A, 6B, 6C shown in the vehicle configuration space 7 are derived from the vehicle data, in particular the vehicle structure 3, whereby the values ​​of the geometric parameters together with the values ​​of the material parameters each describe a point in the vehicle configuration space 7, which can be assigned to a section 3A, 3B, 3C of the vehicle structure 3. Those points in the vehicle configuration space 7 that can be assigned to a single vehicle structure 3 together form a geometric volume region 6A, 6B, 6C.

[0201] Figure 7 differs from Figure 6 in particular in that the outer edges of the vehicle configuration space 7 are shown and that the vehicle configuration space 7 is discretized by volume elements 5A, 5B, 5C, 5D. The vehicle configuration space 7 forms a kind of finite element initial model on the basis of which the vehicle models are configured.

[0202] The vehicle configuration space 7 is a cubic region consisting of connected volume elements with predefined edge lengths. The connections between these elements and their material properties are undefined. Only during the configuration process 203 are these finite element volume elements linked to substitute materials by their position within the geometric volume regions 6A, 6B, 6C, so that the volume elements 5A, 5B, 5C, 5D are assigned to the vehicle 1A, 1B, 1C, in particular its components and cavities. The geometric volume regions 6A, 6B, 6C can, in turn, preferably be reconstructed in the vehicle configuration space 7 from geometries stored in the vehicle data in a voxelized form.

[0203] In the configured vehicle model 4A, 4B, 4C, the volume elements 5A, 5B, 5C, 5D are assigned, depending on their position, to geometric volume regions 6A, 6C, 6C or to other geometric volume regions (not marked with reference numerals) through which components are represented, or to cavities. As can be seen in Figure 7, all components 3A, 3B, 3C of the vehicle structure 3, or rather their geometric volume regions 6A, 6C, 6C, consist of volume elements 5A, 5B, 5D. The volume elements 5A, 5B, 5D form a discretized representation of the geometric volume regions 6A, 6C, 6C, which is suitable for a finite element method with volume elements. By way of example, Figure 7 shows the volume elements 5A, 5B, 5C of volume region 6C of the lower reinforcement 3C of the vehicle front area.

[0204] Figure 8 shows an enlarged section A from Figure 7. Figure 8 again depicts the geometric volume regions 6A, 6C, 6C, which correspond to the radiator, the upper reinforcement of the vehicle front section, and the lower reinforcement of the vehicle front section of a passenger car 1. Furthermore, other geometric volume regions are visible, such as parts of the vehicle frame and a wheel (each without a reference numeral). With regard to the lower front reinforcement of a passenger car 1, or rather the geometric volume region 6C corresponding to this component, three volume elements 5A, 5B, 5C are shown with reference numerals as examples. Naturally, the lower front reinforcement has further volume elements, which, however, are not labeled with reference numerals. All volume elements of the vehicle configuration space 7 are designed as cubes.Volume elements of a single geometric volume region 6A, 6C, 6C are adjacent to each other. Volume elements of two geometric volume regions corresponding to connected vehicle components are also adjacent to each other.

[0205] Furthermore, in Figure 8, the reference numeral 5D indicates a so-called intermediate volume element. This intermediate volume element 5D lies between the wheel, or the volume area representing the wheel, and the engine compartment.

[0206] Intermediate 5D volume elements are defined by a substitute material. This substitute material exhibits generic material properties that depend on the nature of the cavity or the components represented by the intermediate volume elements. Preferably, the substitute material has lower stiffness and / or higher compressibility than the actual material properties of the original components. The substitute materials, or intermediate volume elements containing substitute materials, are used in spaces that can be cavities in the vehicle structure or hollow components. Further spaces are formed by components that cannot be represented as geometric volume regions due to excessively large edge lengths of the volume elements.The stiffness of the replacement material depends in particular on the degree to which the cavity is filled with components that are not, or cannot be, represented by separate volume elements. By providing the intermediate volume elements, the vehicle configuration space 7 can be completely filled with volume elements 5A, 5B, 5C, 5D, whereby specific material properties and plastic and elastic properties can be assigned to each volume element 5A, 5B, 5C, 5D.

[0207] The totality of the volume elements 5A, 5B, 5C, 5D with their respective material properties and the values ​​of the respective geometry parameters in the vehicle configuration space form the vehicle models 4A, 4B, 4C.

[0208] Vehicle models 4A, 4B, and 4C are based on finite element methods, with LS-Dyna being the preferred (explicit) solver. These models are sufficiently complex to provide accurate load-case-specific predictions while remaining sensitive to parameter changes (e.g., different package arrangements should produce trends / effects comparable to those in a detailed vehicle model). Simultaneously, the models are sufficiently abstract to ensure short computation times (less than 20 minutes for a frontal crash). This allows for the most comprehensive experimental design possible.The creation of the vehicle model can be fully automated, preferably using C / C++ programs, based on sets of geometric and material parameter values ​​from user input or from a vehicle database. In particular, vehicle models 4A, 4B, and 4C are fully parameterized models. The connection technology between the individual components 6A, 6b, and 6C of vehicle models 4A, 4B, and 4C is represented by a substitute material, a kind of "filler material," which represents intermediate volume elements 5D with low stiffness and / or compressibility. These intermediate volume elements 5D fill, for example, the front engine compartment and are simultaneously used to compensate for the lack of stiffness in the vehicle components not shown separately (e.g., the front panel, headlights, hood, lines, etc.).

[0209] In a combined approach, it may be provided that, in addition to the volume elements for representing the arrangement and extent of the crash-relevant components (radiator, motor, DC converter, ...), beam (or spring elements) for structural components (longitudinal beams, CMS, ...) and for connecting the individual components in the vehicle models 4A, 4B, 4C are used.

[0210] In a fourth step 204 of procedure 200, a crash load case is simulated for the configured vehicle models 4A, 4B, 4C. A finite element method with volume elements is used, where the volume elements 5A, 5B, 5C, 5D in the vehicle configuration space 7 of vehicle models 4A, 4B, 4C serve as the grid for the finite element method. The simulation generates simulation data that specifies deformation parameters for the intrusion of the passenger car 1A, 1B, 1C, which is simulated in each case.

[0211] System 20 accordingly has means 23 for simulating a crash load case.

[0212] The crash load case simulation is repeated for differently configured vehicle models 4A, 4B, and 4C. Each of these vehicle models 4A, 4B, and 4C was preferably configured based on a selection of value constellations from the vehicle data using statistical experimental designs. As illustrated in Figure 3 using vehicle model 4A as an example, the simulation depicts a passenger car 1A, represented by vehicle model 4A, colliding with an obstacle 8. The upper section of box 23 in Figure 3 shows vehicle model 4A before the collision with obstacle 8, and the lower section of box 23 shows the same vehicle model 4A after the collision. The manner in which vehicle model 4A impacts obstacle 8 during the simulation and the nature of the obstacle 8 define the respective crash load case 9.Examples of crash load cases include a frontal crash against a rigid wall or a frontal crash against a plastically deformable obstacle.

[0213] In a fifth step 205a, the crash load case 9 is preferably simulated again using a conventional finite element method with surface elements, generating FE simulation data that also provide values ​​for deformation parameters for the intrusion of vehicle 1. Alternatively or additionally, in the fifth step 205b, measurement data from a real crash test are acquired, which also include at least one deformation parameter for the intrusion of vehicle 1.

[0214] In a sixth step 206, the configured vehicle models 4A, 4B, 4C are validated by comparing the simulation data from the finite element simulation with volume elements with the FE simulation data from the finite element simulation with surface elements and / or the measurement data. In particular, deviations between the simulation data and the FE simulation data are determined and preferably also evaluated.

[0215] In a seventh step (207), the properties of the substitute material are adjusted based on the validation. This adjustment is carried out to reduce differences between the simulation data, the FE simulation data, and the measurement data. Steps 204 to 207 are preferably repeated until the deviation between the simulation data on the one hand and the FE simulation data and / or the measurement data on the other falls below a predefined threshold. This ensures that the vehicle models 4A, 4B, and 4C adequately represent the vehicles according to the requirements.

[0216] Preferably, steps 205a to 207 are also performed by the simulation means 23. The machine learning model 2 is trained using the generated simulation data and the respective vehicle data of the respective vehicle 1A, 1B, 1C, using whose simulation model 4A, 4B, 4C the simulation was performed. For this purpose, the simulation data and the respective vehicle data are provided to the simulation model 2. In particular, the vehicle data of an input layer and the simulation data of a target output layer of an artificial neural network 2 are provided or read into them in a manner known per se. Based on this training, so-called weights and biases of the artificial neural network are preferably adjusted.

[0217] Preferably, the fourth step 204 of the simulation is repeated not only for different vehicle models 4A, 4B, 4C, but also for different crash load cases 9. In this case, the artificial neural network 2 is provided not only with the vehicle data relating to the simulation data for training, but also with the vehicle data relating to the crash load cases 9. An artificial neural network 2 trained in this way can output values ​​for deformation parameters for different passenger cars 1A, 1B, 1C, but also for different load cases 9.

[0218] Alternatively, a separate experimental design is carried out for each crash load case 9 (frontal crash against a rigid wall, or against an OBD, etc.), and a separate machine learning model 2 is trained for each case. The advantage achieved is higher predictive accuracy of the load case-specific machine learning models 2. Preferably, a separate configuration of vehicle models 4A, 4B, 4C is used for each additional crash load case 9 in order to carry out an experimental design and adapt the corresponding vehicle model 4A, 4B, 4C accordingly. The automatic model generation of the abstract models will / can therefore differ depending on the load case in order to represent the deformations in the most affected areas with sufficient accuracy while still maintaining the highest possible level of abstraction.

[0219] The respective machine models can also be chosen differently depending on the situation (neural network, polynomial, etc.) in order to achieve the best possible accuracy across the entire variation space while simultaneously providing real-time forecasting capability.

[0220] The training of the machine model 2 preferably takes place in system 20 by

[0221] Means 24 for training the machine learning model 2.

[0222] The trained machine learning model 2 outputs values ​​for deformation parameters for the respective crash load case 9 for which it was trained, when corresponding vehicle data of a vehicle 1 to be analyzed is input to it. In particular, the machine learning model 2 outputs time-dependent acceleration and / or intrusion curves. A machine learning model 2 trained using method 200 is particularly suitable for carrying out method 100 described with reference to Figures 1 and 2. Preferably, the two methods 100 and 100 are therefore executed sequentially.

[0223] Furthermore, preferably, the trained machine learning model 2 is output via a second data interface 25 by the system 20 in a ninth processing step 109, as shown in Figure 3.

[0224] The use of a finite element simulation with volume elements to generate the machine model 2, which forms a metamodel with respect to the vehicle models 4A, 4B, 4C used for the simulation, has the advantage that existing databases can be used to determine characteristic geometric features.

[0225] To demonstrate and utilize interrelationships and their effects. Through the construction of vehicle models 4A, 4B, 4C from fixed, interconnected components.

[0226] Volume elements 6A, 6B, 6C, 6D, which are automatically assigned position-dependent material properties, represent a vehicle 1 to be analyzed with sufficient accuracy to predict the correct trends in crash performance, corresponding to the variations in the input geometries. The term "fixedly connected" means, in the context of this disclosure, that the volume elements 5A, 5B, 5C, 5D at interfaces between the volume areas 6A, 6B, 6C are not equipped with a failure criterion, i.e., they cannot fail. The most general validity and predictive accuracy possible of the vehicle models 4A, 4B, 4C (possibly limited to individual vehicle classes and drive concepts) can be ensured by comparison with vehicles from different manufacturers in the database.It is also possible to consider customer-specific vehicle designs during the automated creation of vehicle models 4A, 4B, and 4C. In this case, the process is customized and run again. Individual geometric features that are difficult to capture using dimensions can be stored as scalable 3D templates in the model creation script. Vehicle models 4A, 4B, and 4C have short computation times and can be recalculated separately for individual input parameter sets (i.e., vehicle concepts) to validate the metamodel prediction. This approach also allows the causes of unusually good or poor crash performance results to be quickly visualized (approximately 1 to 2 hours, preferably less than 0.5 hours of computation time for each vehicle model 4A, 4B, and 4C).Through the subsequent mapping process to the OCP value, only the predicted structural performance under crash loads (characterized by a time-dependent acceleration profile) is used for performance evaluation, without taking into account the complexity of the interaction between restraint systems and dummy / occupant kinematics.

[0227] Finally, it should be noted that the exemplary embodiments are merely examples and are not intended to restrict the scope of protection, applications, or structure in any way. Rather, the preceding description provides the person skilled in the art with a guideline for implementing at least one exemplary embodiment, whereby various modifications, particularly with regard to the function and arrangement of the described components, can be made without departing from the scope of protection as defined by the claims and these equivalent combinations of features.

[0228] Reference number list, 1A, 1B, 1C Vehicle machine learning model

[0229] Vehicle structure A, 3B, 3C Component A, 4B, 4C Vehicle model A, 5B, 5C, 5D Volume element A, 6B, 6C Area

[0230] Vehicle configuration room

[0231] Obstacle Crash Load Case 0 Indirect Measurement System 1 First Interface 2 Computational Means 3 Second Interface 0 Machine Entry Model Training System 1 First Interface 2 Configuration Means 3 Simulation Means 4 Training Means

Claims

Patent claims 1. Computer-implemented method (100) for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed (1), using a trained machine model (2), comprising the following steps: • Acquisition (101) of geometric parameters of a large number of different vehicles (1A, 1B, 1C) measured by means of a sensor; • Calculating (102) a probability distribution for each, in particular crash-relevant, component and / or assembly within a vehicle configuration space of the vehicle to be analyzed (1), wherein the probability distribution is derived from the recorded geometry parameters; • Determining (103) boundary conditions relating to the vehicle structure (3), in particular an arrangement of the components and their stiffnesses, of the vehicle to be analyzed (1); • Determining (104) a vehicle structure (3) characterized by at least one geometric parameter of a vehicle (1) to be analyzed on the basis of the boundary conditions and the probability distribution; • Acquiring (105) at least one material parameter of individual sections (3A, 3B, 3C) of the vehicle structure (3), which characterize a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections (3A, 3B, 3C) of the vehicle structure (3), wherein the at least one geometry parameter and the at least one material parameter constitute vehicle data of the vehicle to be analyzed; • Calculating (107) output data using the machine model (2) based on input data, wherein the input data is based on the vehicle data and wherein the output data includes a value of at least one first deformation parameter for the intrusion into the vehicle (1) to be analyzed; and Output (108) of at least one value of the first deformation parameter.

2. Method (100) according to claim 1, further comprising the following step: Capturing (106) a crash load case (9) with respect to the vehicle data, wherein the input data are further based on the crash load case and wherein the value of at least one first deformation parameter characterizes the intrusion in this crash load case.

3. Computer-implemented method (200) for training a machine-needle model (2), in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed (1), comprising the following steps: • Acquisition (201) of vehicle data of a large number of different vehicles (1 A, 1 B, 1 C), wherein the vehicle data comprise at least one geometric parameter measured by means of a sensor which characterizes a vehicle structure (3), and material parameters of individual sections of the vehicle structure (3) which characterize a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections (3A, 3B, 3C) of the vehicle structure (3), wherein the vehicle data span a variation space; • Configuring (203) vehicle models (4A, 4B, 4C) based on vehicle data, wherein each configured vehicle model (4A, 4B, 4C) defines volume elements (5A, 5B, 5C, 5D) of the respective vehicle (1A, 1B, 1C), which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter of the individual sections of the vehicle structure (3), at a respective position of a volume element (5A, 5B, 5C, 5D), wherein the sections of the vehicle structure (3) of each vehicle (1A, 1B, 1C) are mapped by geometric volume regions (6A, 6B, 6C) in a vehicle configuration space (7), wherein dimensions and / or positions of the geometric volume regions (6A, 6B, 6C) in the vehicle configuration space (7) are derived from the vehicle data are derived, wherein an assignment of the volume elements (5A, 5B, 5C, 5D) to the respective sections (3A, 3B, 3C) of the vehicle structure (3) is defined by the position of the volume elements (5A, 5B, 5C, 5D) in relation to the geometric volume areas (6A, 6B, 6C); • Simulating (204) a crash load case for the configured vehicle models (4A, 4B, 4C) using a finite element method with the volume elements (5A, 5B, 5C, 5D), generating simulation data which includes at least one first deformation parameter for the intrusion into the vehicle (1A, 1B, 1C), and repeating the simulation of the crash load case for differently configured vehicle models (4A, 4B, 4C); and • Training (208) of the machine learning model (2), wherein the vehicle data and the simulation data are provided to the machine learning model (2), wherein the vehicle data of a respective vehicle (1A, 1B, 1C) are assigned to the simulation data of this vehicle (1A, 1B, 1C), wherein the trained machine learning model (2) is configured to output at least one value of the first deformation parameter as output data, based on vehicle data of a vehicle (1) to be analyzed, measured at least partially by means of a sensor, as input data.

4. Method (200) according to claim 3, wherein the configuration of the vehicle models (4A, 4B, 4C) comprises the conversion of vehicle data defining a vehicle structure (3) with different components (3A, 3B, 3C) into geometric volume regions (6A, 6B, 6C) made up of connected volume elements (5A, 5B, 5C, 5D).

5. Method according to claim 3 or 4, wherein the vehicle data in the variation space define the subsequent dimensions of the volume areas (6A, 6B, 6C) in the vehicle configuration space (7).

6. Method according to any one of claims 3 to 5, wherein the vehicle data further specify predefined volume ranges (6A, 6B, 6C) which in particular correspond to an assembly of components, for example an engine assembly.

7. Method (200) according to one of claims 3 to 6, wherein an assignment of the volume elements (5A, 5B, 5C, 5D) to the respective sections (3A, 3B, 3C) of the vehicle structure (3) is defined by the position of the volume elements (5A, 5B, 5C, 5D) in relation to the geometric volume regions (6A, 6B, 6C).

8. Method (200) according to claim 6 or 7, wherein, when configuring the vehicle models, the geometric volume regions (6A, 6B, 6C) and / or intermediate volume elements (5D) which lie between the geometric volume regions (6A, 6B, 6C) are each represented with a substitute material which is determined on the basis of the material properties of the respective sections (3A, 3B, 3C) of the vehicle structure (3), in particular the component or components of the respective section (3A, 3B, 3C), and / or the respective space.

9. Method (200) according to claim 8, wherein the intermediate volume elements (5D) additionally compensate for a missing contribution to the stiffness of the vehicle components (1A, 1B, 1C) not considered with their own volume elements.

10. Method (200) according to any one of claims 3 to 9, further comprising the following steps: • Simulating (205A) the crash load case based on vehicle data from a multitude of different vehicles (1A, 1B, 1C) using a finite element method with surface elements, wherein FE simulation data are generated for each vehicle (1A, 1B, 1C) which include at least one deformation parameter for the intrusion into the vehicle, and / or • Reading (205B) measurement data from a real crash test, which includes at least one deformation parameter for intrusion into the vehicle; and • Validating (206) the configured vehicle models (4A, 4B, 4C) by comparing the simulation data with the FE simulation data and / or the measurement data.

11. The method of claim 10 in conjunction with one of claims 8 or 9, further comprising the following step: • Adjusting (207) properties of the substitute material, wherein the simulation of crash load cases for the defined number of configured vehicle models (4A, 4B, 4C) using a finite element method with volume elements (3A, 3B, 3C) is repeated until a deviation between the simulation data on the one hand and the FE simulation data and / or the measurement data on the other hand falls below a predefined threshold.

12. Method (200) according to any one of claims 3 to 11, further comprising the following step: • Selections (202) of the vehicle data used to generate the vehicle models (4A, 4B, 4C) from the variation space by means of a statistical design of experiments.

13. Method (200) according to claim 12, wherein the statistical experimental design comprises an automated adjustment of the vehicle data, in particular the geometry parameters and material parameters, by means of predefined mathematical relationships in order to fill the variation space with geometrically meaningful related vehicle structures (3) that simultaneously meet predefined installation space and distance requirements.

14. Method (200) according to one of claims 3 to 13, wherein the simulated crash load case is additionally provided to the machine learning model during training, wherein the simulated load case is assigned to the simulation data and the vehicle data.

15. Method (200) according to any one of claims 3 to 14, wherein the simulation is repeated for different load cases.

16. Method (100) according to one of claims 1 or 2, wherein the trained machine learning model (2) is trained by means of a method according to one of claims 3 to 15.

17. System (10) for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed (1), using a trained machine model (2), comprising: • a first interface (11) for acquiring geometric parameters measured by a sensor of a plurality of different vehicles (1A, 1B, 1C) and of at least one material parameter of individual sections (3A, 3B, 3C) of the vehicle structure (3), which characterize a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections (3A, 3B, 3C) of the vehicle structure (3), wherein the at least one geometric parameter and the at least one material parameter constitute vehicle data of the vehicle to be analyzed; • Means (12) for calculating a probability distribution for each component and / or assembly, in particular a crash-relevant one, within a vehicle configuration space of the vehicle (1) to be analyzed, wherein the probability distribution is derived from the acquired geometric parameters, for determining boundary conditions with respect to the vehicle structure (3), in particular an arrangement of the components and their stiffnesses, of the vehicle (1) to be analyzed, for determining a vehicle structure (3) characterized by at least one geometric parameter of a vehicle (1) to be analyzed on the basis of the boundary conditions and the probability distribution, and for calculating output data using the machine model (2) on the basis of input data,wherein the input data are based on the vehicle data and wherein the output data include a value of at least one first deformation parameter for the intrusion into the vehicle to be analyzed (2); and, • A second interface (13) for outputting (109) at least one value of the first deformation parameter.

18. System (20) for training a machine learning model (2), in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed (1), comprising: • a first interface (21 ) for acquiring vehicle data from a multitude of different vehicles (1A, 1B, 1C), wherein the vehicle data comprise at least one geometric parameter measured by means of a sensor, which characterizes a vehicle structure (3), and material parameters of individual sections of the vehicle structure (3), which characterize a mass as well as plastic and elastic properties, in particular E-moduli and flow curves, of the individual sections (3A, 3B, 3C) of the vehicle structure (3), wherein the vehicle data span a variation space; • Means (22) for configuring vehicle models (4A, 4B, 4C) on the The basis of the vehicle data, wherein each configured vehicle model (4A, 4B, 4C) defines volume elements (5A, 5B, 5C, 5D) of the respective vehicle (1A, 1B, 1C), which relationships and to which a stiffness and a material property are assigned depending on the geometry parameter and the material parameter of the individual sections of the vehicle structure (3) at a respective position of a volume element (5A, 5B, 5C, 5D), wherein the sections of the vehicle structure (3) of each vehicle (1A, 1B, 1C) are mapped by geometric volume regions (6A, 6B, 6C) in a vehicle configuration space (7), wherein dimensions and / or positions of the geometric volume regions (6A, 6B, 6C) in the Vehicle configuration space (7) are derived from the vehicle data, wherein an assignment of the volume elements (5A, 5B, 5C, 5D) to the respective sections (3A, 3B, 3C) of the vehicle structure (3) is defined by the position of the volume elements (5A, 5B, 5C, 5D) in relation to the geometric volume areas (6A, 6B, 6C); • Means (23) for simulating a crash load case for the configured vehicle models (4A, 4B, 4C) using a finite element method with the volume elements (5A, 5B, 5C, 5D), generating simulation data which includes at least one first deformation parameter for the intrusion into the vehicle (1A, 1B, 1C), wherein the simulation for differently configured vehicle models (4A, 4B, 4C) are repeated; and • Means (24) for training the machine learning model (2), wherein the vehicle data and the simulation data are provided to the machine learning model (2), wherein the vehicle data of a respective vehicle (1A, 1B, 1C) are assigned to the simulation data of that vehicle, wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle (1) to be analyzed, measured at least partially by means of a sensor, as input data.

19. System (10) according to claim 17, wherein the trained machine learning model (2) is trained by means of a system (20) according to claim 18.

20. Computer-implemented classifier (2), in particular an artificial neural network, for the indirect measurement of deformation parameters for a crash test of a vehicle to be analyzed, wherein the classifier is generated by training a classification algorithm, wherein the classification algorithm was configured by the following steps, which are performed for each training input of a plurality of training inputs: • Acquisition of vehicle data from a large number of different vehicles, wherein the vehicle data includes at least one geometric parameter characterizing a vehicle structure and material parameters of individual sections of the vehicle structure, characterizing a mass as well as plastic and elastic properties, in particular E-modules and flow curves, of the individual sections of the vehicle structure, wherein the vehicle data spans a variation space; • Configuring vehicle models based on the Vehicle data, including each configured vehicle model Volume elements of the respective vehicle are defined, which relationships and each of which has a stiffness and a material property depending on the geometry parameter and the material parameter at a respective position of a volume element is assigned, wherein the sections of the vehicle structure (3) of each vehicle (1A, 1B, 1C) are mapped by geometric volume areas (6A, 6B, 6C) in a vehicle configuration space (7), wherein dimensions and / or positions of the geometric volume areas (6A, 6B, 6C) in the vehicle configuration space (7) are derived from the vehicle data, wherein an assignment of the volume elements (5A, 5B, 5C, 5D) to the respective sections (3A, 3B, 3C) of the vehicle structure (3) is defined by the position of the volume elements (5A, 5B, 5C, 5D) in relation to the geometric volume areas (6A, 6B, 6C); • Simulating a single crash load case for the configured vehicle models using a finite element method with volume elements, generating simulation data that includes at least one first deformation parameter for the intrusion into the vehicle, and repeating the crash load case simulation for differently configured vehicle models; and • Training a machine learning model to generate the classifier, wherein the vehicle data and the simulation data are provided to the machine learning model, wherein the vehicle data of a given vehicle are associated with the simulation data of that vehicle, and wherein the trained machine learning model is configured to output at least one value of the first deformation parameter as output data based on vehicle data of a vehicle to be analyzed as input data.

Images