Method and system for ascertaining physical parameters for different vehicle specifications using a transformation model

Abstract

The invention relates to a computer-implemented method (100) for ascertaining physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2CB,..., Y1D,Y2D,...) of a vehicle (1) to be analyzed for different vehicle specifications using a transformation model, wherein a vehicle specification of the vehicle (1) to be analyzed is detected, each vehicle specification is characterized by the values of the technical properties (X1A,X2A,..., X1B,X2BC,..., X1BC,X2C,..., X1D, X2D,…) of the vehicle (1) to be analyzed, the values of the physical parameters of the vehicle (1) to be analyzed are ascertained using a plurality of different sub-models (A, B, C, D) of the transformation model on the basis of the values of the technical properties (X1A,X2A,..., X1B,X2BC,..., X1BC,X2C,..., X1D, X2D,...), and the values of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) of the vehicle (1) to be analyzed are output.

Description

[0001] Method and system for determining physical parameters for different vehicle specifications using a transformation model

[0002] The invention relates to a computer-implemented method and a corresponding system for determining physical parameters of a vehicle to be analyzed for different vehicle specifications by means of a transformation model.

[0003] A large number of physical parameters of a vehicle can be measured only with difficulty or with great effort during tests, especially during crash tests.

[0004] The invention aims to make it possible to determine vehicle parameters that are difficult or expensive to measure, particularly in an early phase of vehicle development.

[0005] This problem is solved by the teaching of the independent claims. Advantageous embodiments can be found in the subclaims.

[0006] A first aspect of the invention relates to a method for training a transformation model having a plurality of submodels for determining physical parameters of a vehicle to be analyzed for different vehicle specifications, comprising the following steps:

[0007] Recording vehicle specifications of a vehicle, in particular of the vehicle to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties of the vehicle;

[0008] Simulating an operation of the vehicle, preferably in at least one defined driving maneuver and / or in at least one defined virtual scenario, by means of simulation models taking into account measurement data of at least one real vehicle, wherein the simulation models represent a virtual vehicle, wherein the simulation is carried out several times with different vehicle specifications of the vehicle and wherein operating data is generated;

[0009] Deriving values ​​of the physical parameters of the vehicle from the operating data; providing the values ​​of the physical parameters generated by the simulation models and the associated vehicle specifications, in particular at a data interface; and

[0010] Training of adjustment calculations, in particular artificial neural networks, using the values ​​of the physical parameters and the respective associated vehicle specifications, whereby each adjustment calculation forms the basis for one of the sub-models, whereby physical parameters derived from different simulation models are each read into different artificial neural networks.

[0011] Preferably, a separate submodel is trained for each physical parameter.

[0012] A second aspect of the invention relates to a computer-implemented method for determining physical parameters of a vehicle to be analyzed for different vehicle specifications by means of a transformation model, comprising the following steps:

[0013] Recording a vehicle specification of the vehicle to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties of the vehicle to be analyzed;

[0014] Determining values ​​of the physical parameters of the vehicle to be analyzed by means of a plurality of different sub-models of the transformation model on the basis of the values ​​of the technical properties, wherein the sub-models each have an assignment rule between vehicle specifications and at least one physical parameter of the vehicle to be analyzed and are each based on a compensation calculation with regard to simulation results resulting from several operational simulations by means of simulation models that represent a virtual vehicle, taking into account measurement data of at least one real vehicle with a different vehicle specification of the virtual vehicle, wherein the simulation results of different simulation models are included in different compensation calculations and wherein at least two of the compensation calculations are artificial neural networks,which were preferably trained using a method according to the first aspect of the invention; and outputting the values ​​of the physical parameters of the vehicle to be analyzed. Preferably, there is a separate submodel for each physical parameter.

[0015] A third aspect of the invention relates to a computer-implemented method for determining vehicle specifications of a vehicle to be analyzed for a set of physical parameters by means of a transformation model, comprising the following steps:

[0016] Collecting values ​​of a plurality of physical parameters of the vehicle to be analyzed;

[0017] Determining a vehicle specification, which is characterized by an expression of the values ​​of technical properties of the vehicle to be analyzed, by means of a gradient descent method starting from default values ​​for the vehicle specification, wherein within the gradient descent method, the method according to the second aspect of the invention is preferably carried out several times in order to determine the vehicle specification; and

[0018] Outputting the vehicle specification of the at least one vehicle to be analyzed.

[0019] A fourth aspect of the invention relates to a computer program or a 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 or a system, cause the computer or the system to carry out the steps of a method according to the first, second or third aspect.

[0020] A fifth aspect of the invention relates to a system for training a transformation model having a plurality of submodels for determining physical parameters of a vehicle to be analyzed for different vehicle specifications, comprising: a first interface, in particular a data interface, for detecting vehicle specifications of a vehicle, in particular of the vehicle to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties of the vehicle;Means for simulating operation of the vehicle, preferably in at least one defined driving maneuver and / or in at least one defined virtual scenario, by means of simulation models taking into account measurement data of at least one real vehicle, wherein the simulation models represent a virtual vehicle, wherein the simulation is carried out several times with different vehicle specifications of the vehicle in each case, and wherein operating data is generated;

[0021] Means for deriving values ​​of the physical parameters of the vehicle from the operating data; a second interface, in particular a data interface, for providing the values ​​of the physical parameters generated by the simulation models and the respective associated vehicle specifications for training adjustment calculations, in particular artificial neural networks, wherein each adjustment calculation forms the basis for one of the submodels, wherein physical parameters derived from different simulation models are each read into different artificial neural networks.

[0022] A sixth aspect of the invention relates to a system for determining physical parameters of a vehicle to be analyzed for different vehicle specifications by means of a transformation model, comprising the following steps: an interface, in particular a data interface, for detecting a vehicle specification of the vehicle to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties of the vehicle to be analyzed, and for outputting the values ​​of the physical parameters of the vehicle to be analyzed; and

[0023] Means for determining values ​​of the physical parameters of the vehicle to be analyzed by means of a plurality of different sub-models of the transformation model on the basis of the values ​​of the technical properties, wherein the sub-models each have an assignment rule between vehicle specifications and at least one physical parameter of the vehicle to be analyzed and are each based on a compensation calculation with regard to simulation results which result from several operating simulations by means of simulation models which depict a virtual vehicle, taking into account measurement data of at least one real vehicle, each with a different vehicle specification of the virtual vehicle, wherein the simulation results of different simulation models are included in different compensation calculations and wherein at least two of the compensation calculations are artificial neural networks.

[0024] A seventh aspect of the invention relates to a computer-implemented machine learning model, in particular an artificial neural network, for determining physical parameters of a vehicle to be analyzed for different vehicle specifications, wherein the machine learning model has been configured by performing the working steps of a method according to the first aspect of the invention for each training input.

[0025] A vehicle specification within the meaning of the invention is preferably a manifestation of a plurality of technical characteristics of a vehicle. Further preferably, it is a detailed description of the technical characteristics of a vehicle, containing technical information and features of the vehicle.

[0026] Technical properties within the meaning of the invention are preferably features of a device which are suitable for characterising it.

[0027] A genus of units within the meaning of the invention is preferably a set of units whose essential technical properties lie within a defined range of values ​​or preferably correspond. A specific unit is therefore preferably a realization of the genus of units.

[0028] A transformation model within the meaning of the invention preferably establishes a relationship between observed variables and target variables or a group of target variables. In particular, a transformation model establishes a relationship between at least one physical parameter and at least one technical property of at least one vehicle. Preferably, the transformation model is not time-resolved. However, the transformation model itself is more preferably based on a time-resolved simulation of the operation of a vehicle with a plurality of different vehicle specifications. Simulating the operation of a unit within the meaning of the invention is preferably a time-resolved, distance-resolved, frequency-resolved, or static simulation of the operating behavior of a unit, in particular a vehicle, using a simulation model.The operation of a unit is characterized in particular by observed variables such as speed, torque, mass flows, pressures, temperatures, energy consumption, consumption of operating materials, emissions, OBD values, speed, gear selection, etc.

[0029] A regression calculation within the meaning of the invention is preferably a regression method or a pattern recognition method, in particular based on an artificial neural network, or an artificial neural network, and / or a combination of both, in particular in the sense of statistical theory. In particular, a regression calculation can be a polynomial model, a linear regression, or a decision tree. Artificial neural networks are particularly advantageous for complex, particularly non-linear, relationships. Preferably, different submodels can be based on different model types.

[0030] Acquiring measurement data within the meaning of the invention preferably involves reading in measurement data via a data interface. Alternatively or additionally, acquiring includes determining a measurement signal using a sensor and / or post-processing a measurement signal to generate the measurement data.

[0031] Outputting, within the meaning of the invention, preferably means providing. Providing is preferably done via a data interface. However, it is also possible to output information via a user interface, for example, a screen.

[0032] An operating cycle in the sense of the invention is preferably a time-resolved and / or distance-resolved sequence of operating points of an operation of a unit.

[0033] A scenario within the meaning of the invention is preferably formed from a time-resolved and / or distance-resolved sequence of operating points that characterize a specific load case. A vehicle within the meaning of the invention is any type of land, water, air, or space vehicle, in particular a motor vehicle.

[0034] A means within the meaning of the present invention can preferably be designed in hardware and / or software, in particular a processing unit, in particular a microprocessor unit (CPU), which is preferably connected to a memory and / or bus system in terms of data or signals, and / or one or more programs or program modules. The microprocessor unit can be designed to process instructions implemented as a program stored in a memory system, to detect input signals from a data bus and / or to output signals to a data bus. A memory system can have one or more, in particular different, storage media, in particular optical, magnetic, solid-state and / or other non-volatile media. The program can be designed in such a way that it embodies the methods described here oris capable of carrying out such a method, so that the microprocessor unit can carry out the steps of such methods and thus in particular can analyse at least one device or train a transformation model.

[0035] The invention is based on the approach of being able to determine physical parameters of a vehicle, which are difficult or very costly to determine, on the basis of a limited number of measurements on at least one real vehicle.

[0036] This is achieved by generating a calculation model or prediction model, called transformation model, based on the limited number of measurements, with which predictions of the difficult-to-determine physical parameters can be determined in a comparatively large range of validity of the calculation model.

[0037] To create such a calculation model or transformation model, parameterizable virtual prototypes are first created. These comprise a whole range of simulation models, for example, 3D CAD models, crash models, driving dynamics simulation models, etc. Each of the virtual prototypes has specific technical properties. These technical properties are variation parameters and span a parameter space. Using these virtual prototypes, virtual tests, in particular virtual test drives, are carried out, generating operating data. This operating data initially consists of signals that are discretized with respect to other parameters, i.e., dependent on them. Usually, time-discrete signals are generated, but other signal types are also conceivable, for example, dependencies on a distance traveled, an excitation frequency, or even a bending stress.The virtual tests can also cover very specific aspects of vehicle operation, for example a driving dynamics simulation or a crash simulation.

[0038] Key figures representing physical parameters, preferably scalars, are derived from the parameter-discretized operating data. These key figures are compared with measured data of the physical parameters of real vehicles, preferably from real test drives. The measured data can be measured directly with sensors or derived from measurements. Based on this comparison, the virtual prototype, i.e., its simulation models, are modified or calibrated until a sufficient match is established between the measured data and the physical parameters determined using the virtual prototypes, in particular until a deviation in the physical parameters falls below a defined limit.

[0039] Based on these calibrated virtual prototypes, parameter variations are then performed, i.e., the specification and / or the values ​​of the virtual prototype are changed. A statistical experimental design is preferably used for this parameter variation to achieve the best possible coverage of the parameter space with a comparatively small number of variations.

[0040] Defined virtual scenarios are then processed, which are the same scenarios as in the virtual tests performed at the beginning. This in turn generates operating data and, after evaluation or post-processing of this operating data, values ​​of physical parameters.

[0041] The values ​​of the physical parameters generated using or derived from the virtual prototypes and the corresponding vehicle specifications are provided at a data interface.

[0042] This data is input into a fitting calculation, which forms the basis for each sub-model. Sub-models are trained in this way. These fitting calculations are preferably artificial neural networks. In this case, the values ​​of the physical parameters are provided at the output layer, and the corresponding vehicle specifications are provided at the input layer, allowing the artificial neural network to develop appropriate weights and biases to represent the corresponding relationships.

[0043] Using the prediction model or transformation model, the physical parameters on which the prediction or transformation model was trained can be determined by feeding vehicle specifications into the prediction or transformation model as input. The physical parameters themselves do not need to be measured, nor are further measurements of other parameters necessary to derive the physical parameters.

[0044] Because the submodels created by the trained adjustment calculations are each trained using data from different simulation models, they are independent of each other. This allows the individual submodels to be validated and optimized separately. As a result, the transformation model also exhibits a high degree of accuracy compared to a global transformation model that does not have submodels. In a global transformation model, the values ​​of the physical parameters generated by the simulation models would be imported into a single adjustment calculation. In particular, in the transformation model with submodels, cross-influences between different physical parameters can be made visible to the user. However, such a global transformation model would be much more difficult to validate.Individual submodels can be validated or verified before being linked to the transformation model. The relevant technical properties that must be modified for validation can be determined, preferably using a sensitivity analysis. Furthermore, the variation space of a global transformation model is much more difficult to define. In particular, individual variation spaces can be defined for each submodel, even for identical technical properties. Training efficiency can also be increased, as each submodel has fewer input parameters than the transformation model. This can significantly reduce the training time for the entire transformation model due to the lower complexity.

[0045] The individual submodels are easily validated for their respective variation space. Thus, the use of submodels can prevent or at least reduce so-called hallucinations of trained fitting calculations—incorrect predictions, which occur particularly in trained artificial neural networks.

[0046] The invention offers a significant advantage over directly training the prediction or transformation models based on measurement data from real vehicles. For example, if crash tests are to be simulated, a large number of vehicles (in particular, more than 1,000 vehicles with different masses, battery capacities, and wheelbases) would have to be sacrificed in crash tests to determine a relationship between the specifications of these vehicles and the maximum acceleration on a crash test dummy. This would ultimately also yield the necessary training data for the prediction or transformation models, but would obviously be significantly more complex and only feasible if the necessary number of vehicles are available for destruction. In an advantageous embodiment, the method further comprises at least one of the following steps:

[0047] Simulating an operation of a virtual vehicle by means of, preferably in at least one defined driving maneuver and / or in at least one defined virtual scenario, wherein the simulation models depict the virtual vehicle, and wherein operating data is generated;

[0048] Deriving values ​​of physical parameters of the virtual vehicle from the operating data;

[0049] Collecting the measurement data of the physical parameters of at least one real vehicle;

[0050] Comparing the derived values ​​with the measured data of the physical parameters; and calibrating the simulation models if a deviation measure with respect to the physical parameters exceeds a defined limit.

[0051] By taking the measurement data into account when calibrating the simulation models, particularly realistic simulation models can be created.

[0052] In a further advantageous embodiment, the method comprises the following work step:

[0053] Selection of vehicle specifications used for operational simulation using different simulation models for each simulation model of a statistical design of experiments.

[0054] By applying a statistical design of experiments, the number of operational simulations can be reduced.

[0055] In a further advantageous embodiment of the method, when deriving from the operating data which depend on a parameter, in particular are time-dependent, scalars are generated as values ​​of the physical parameters by means of data post-processing.

[0056] By using scalars, information compression can be achieved.

[0057] This allows the transformation models to be significantly simplified. In a further advantageous embodiment of the method, event-based physical parameters are generated from time-based data when derived from the operating data through data post-processing.

[0058] By using event-based physical parameters, information compression can be achieved. Therefore, transformation models can be significantly simplified.

[0059] In a further advantageous embodiment of the method, the parameter space is limited when simulating operation in such a way that vehicle specifications or operating conditions that differ greatly from vehicles of the respective vehicle type do not occur.

[0060] This allows the transformation model derived from the operating data to be kept relatively simple. Operating areas that are no longer technically feasible can be excluded.

[0061] In a further advantageous embodiment of the method, the simulated operation of the vehicle covers a variety of driving maneuvers and / or defined virtual scenarios.

[0062] This provides particularly good coverage of the parameter space under investigation, increasing the quality of the submodels of the transformation model.

[0063] In a further advantageous embodiment of the method, at least one physical characteristic value determined by means of a first sub-model is used as an input parameter in a second sub-model.

[0064] By nesting the submodels in this way, a particularly efficient transformation model can be realized.

[0065] In a further advantageous embodiment of the method, the operating data of different simulation models are generated simultaneously or at different times. This allows for optimal utilization of existing computing capacity, depending on availability.

[0066] In a further advantageous embodiment of the method, the technical properties used for operational simulation by means of different simulation models are identical, different or overlapping.

[0067] In a further advantageous embodiment of the method, at least one technical property is an input parameter for several submodels, in particular submodels of different transformation models.

[0068] This makes it possible to determine the effects of different values ​​of technical properties as input parameters on different physical parameters of the vehicle as output parameters using one or more transformation models. In particular, the optimal value of a technical property for different, sometimes contradictory, target values ​​of the physical parameters can be determined during back-calculation. In particular, this allows for a comprehensive optimization for conflicting optimization goals.

[0069] In a further advantageous embodiment of the method, different submodels each have different physical parameters as output parameters.

[0070] The features and advantages described with respect to the first aspect of the invention apply accordingly to the second, third, fourth, fifth, sixth and seventh aspects of the invention and vice versa.

[0071] Further features and advantages will become apparent from the following description with reference to the figures, which are at least partially schematic: Figure 1: is a representation of an embodiment of a data flow when determining physical parameters of a vehicle to be analyzed for various vehicle specifications;

[0072] Figure 2: is a flowchart of an embodiment of a method for training a transformation model for determining physical parameters of a vehicle to be analyzed for different vehicle specifications;

[0073] Figure 3: is a flowchart of an embodiment of a computer-aided method for determining physical parameters of a vehicle to be analyzed for different vehicle specifications by means of a

[0074] T ransformation model;

[0075] Figure 4: is a computer-aided method for determining

[0076] Vehicle specifications of a vehicle to be analyzed for various physical parameters using a transformation model;

[0077] Figure 5: is an illustration of an embodiment of a system for training a transformation model for determining physical parameters of a vehicle to be analyzed for different vehicle specifications;

[0078] Figure 6: is an illustration of an embodiment of a system for determining physical parameters of a vehicle to be analyzed for different vehicle specifications using a transformation model.

[0079] Figure 1 is an embodiment of a data flow when determining physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C, ..., Y1D,Y2D,... of a vehicle 1 to be analyzed for different vehicle specifications.

[0080] In a first phase, a virtual vehicle 2, formed by a plurality of simulation models a, b, c, d, simulates vehicle operation according to operating cycles in virtual scenarios. The virtual vehicle is preferably of the same type as the vehicle 1 to be analyzed, whose physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D, ... are to be determined later.

[0081] The operating data generated by the virtual prototype or vehicle 2 are evaluated so that physical parameters Y1A,Y2A, ..., Y1B,Y2B,..., Y1C,Y2C, ..., Y1D,Y2D, ... can be derived from them. The simulation models include, for example, 3D CAD models, crash models, vehicle dynamics simulation models, etc. Crash models are preferably derived from predefined finite element models for class-typical vehicles, i.e., vehicles of the same type as a vehicle 1 to be analyzed.

[0082] For example, if the electric range of a vehicle is to be examined, the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C, ..., Y1D,Y2D, ... are a range in kilometers and an efficiency in kilowatt-hours per 100 km. These physical parameters Y1A,Y2A,..., Y1B,Y2B, ..., Y1C,Y2C,..., Y1D,Y2D, ... can be determined, for example, in a "WLTP shortened" scenario. Furthermore, these physical parameters can also be determined in other scenarios, such as the "MCT (EPA 2-Cycle)" scenario. If crash safety is to be investigated, these physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D, ... can be, for example, the mean impulse in meters per second squared or the deformation of the passenger cabin in meters. The "US NCAP" scenario with "frontal 56 km / h" and the "Offset-deformable barrier-ODB 40%-64 km / h" scenario are preferred.The values ​​of these physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... determined using the virtual prototype 2 are compared with measurement data of these physical parameters Y1A,Y2A,..., Y1B,Y2B, ... , Y1c,Y2c,..., Y1D,Y2D,... on real vehicles 3. Therefore, the measurement data, indicated by the dashed line in Figure 1 , are included in the simulation models a, b, c, d of the virtual prototype 2. Based on a comparison of the values, these simulation models a, b, c, d are calibrated or adjusted.

[0083] The measurement data, in particular from road measurements, are generated by means of one or more real vehicles, preferably in real test drives, more preferably on the test bench or on a real road, in particular in real road traffic.

[0084] Using the virtual prototype 2 generated in this way using the measurement data of real vehicles 3, operational simulations are then carried out. In these operational simulations, the operation of the vehicle 1 to be analyzed is simulated by providing the virtual prototype 2, i.e., its simulation models a, b, c, d, with a vehicle specification of the vehicle 1 to be analyzed. This is indicated in Figure 1 by dashed connecting lines. Each vehicle specification is characterized by an expression or realization of the values ​​of technical properties X1A, X2A,..., X1B, X2BC,..., X1BC, X2C,..., X1D, X2D,... of the vehicle. The operational simulations are carried out using the virtual prototype calibrated using real measurement data and the values ​​for technical properties X1A,X2A, ... , X1 B,X2BC,..., X1 BC,X2C, ... , X1 D, X2D, ... resulting from the vehicle specification.of the vehicle 1, which are transferred to the prototype 2. The operational simulations are carried out several times, each with a different vehicle specification, whereby operational data is again generated by the simulation models a, b, c, d.

[0085] In turn, values ​​of physical parameters Y1A,Y2A,..., Y1 B,Y2B,..., Y1 C,Y2C,..., Y1 D,Y2D,... are derived from these operating data through data post-processing.

[0086] As shown in Figure 1, these values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C, ..., Y1D,Y2D,... as well as the respective associated vehicle specification, ie the vehicle specification by means of which the respective values ​​were generated, are provided to the submodels A, B, C, D of the transformation model T. In this way, the submodels A, B, C, D, which are preferably based on an artificial neural network or regression, are trained.

[0087] As also shown in Figure 1, in a second phase, vehicle specifications of a vehicle 1 to be analyzed can now be made directly available to this trained transformation model, as also indicated by dashed lines in Figure 1. These vehicle specifications do not have to be identical to those provided to the virtual prototype 2 in the first phase. However, they should be within the validity range of the transformation model T.

[0088] This validity range is preferably the range of the transformation model in which the submodels were calibrated with measurement data from real vehicles.

[0089] If the submodels A, B, C, D of the transformation model T are provided with a specification of a vehicle to be analyzed as input, they output the values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... as indirect measured values.

[0090] As shown in Figure 1, a technical property X2BC, X1BC can be an input parameter for several submodels, in this case B, C. However, different submodels A, B, C, D of the transformation model T each have different physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... as output parameters.

[0091] Examples of transformation models T are a crash transformation model T, which outputs physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... related to crash safety, a handling / driveability transformation model T, which outputs physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... related to driveability, a braking transformation model T, which outputs physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... related to the braking behavior of the vehicle, a driveability transformation model T, which outputs physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... relating to the vehicle's driving behavior, a performance transformation model T which outputs physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... relating to the driving performance, and an energy management transformation model T which outputs physical parameters Y1A,Y2A,..., Y1B,Y2B,...., Y1C,Y2C,..., Y1D,Y2D,... in relation to the vehicle's energy consumption. An exemplary list of technical properties X1A,X2A,..., X1B,X2BC,... , X1BC,X2C, ... , X1D, X2D,.., which can be selected as input parameters for the sub-models A, B, C, D of the crash transformation model T, is given in Table 1, which is reproduced below. The technical properties X1A,X2A,..., X1B,X2BC,..., X1BC,X2C, ..., X1D, X2D,.. applicable to the transformation model T are marked with "x". In addition, Table 1 provides default values ​​for the technical properties X1A, X2A, ..., X1B, X2BC, ..., X1BC, X2C, ..., X1D, X2D, ..., as well as minimum and maximum values ​​where applicable. The default values ​​are preferably used for the initial operational simulation of the virtual vehicle 2 using the simulation models a, b, c, d. The respective unit of the technical property X1 is also specified. A ,X2 A ,..., X1 B,X2 BC ,..., X1 BC ,X2C,..., X1 D , X2 D ,.. is given in Table 1. Dimensionless properties are indicated by a dash.

[0092] Table 1:

[0093] The technical properties X1A, X2A,..., X1 B, X2BC,..., X1 BC, X2C,..., X1 D, X2D,.. for the respective transformation models T for handling, braking, drivability, power and energy management can be selected from the list in Table 2 below. The

[0094] The applicable technical properties X1A, X2A, ... , X1B, X2BC, ... , X1BC, X2C, ..., X1D, X2D, ... are marked with an "x". The default, minimum, and maximum values, as well as the unit, are also indicated here.

[0095] Table 2:

[0096]

[0097]

[0098]

[0099] The crosses in Tables 1 and 2 indicate which technical property can be used as an input variable for which transformation model T. From the following list in Table 3, the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... can be selected, the values ​​of which are to be output by the respective transformation model T. This table also specifies the units of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.... In addition, it is specified which property of a

[0100] Vehicle, the respective physical parameter Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... as well as the scenario or driving maneuver in which the physical parameter Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... can be determined during the operational simulation of the vehicle using the simulation models a, b, c, d, ie can be derived from the operating data.

[0101] Table 3

[0102]

[0103]

[0104] Figure 2 shows an embodiment of a method 100 for training a transformation model T. The method 100 according to the invention is described by way of example for training a crash transformation model T and an energy management transformation model T. However, it will be apparent to those skilled in the art that other transformation models T, such as for braking, drivability / driving behavior, handling / handling, and performance, are also applicable. In a first work step 101, operation of a virtual vehicle 2 is preferably simulated using simulation models a, b, c, d. This simulation preferably takes place in a defined driving maneuver and / or in at least one defined virtual scenario.A defined virtual scenario is, for example, a crash scenario, such as the scenario "US NCAP full with frontal 56 km / h" or "Offset deformal-barrier ODB-40%-64 km / h", each for training a crash transformation model T.

[0105] For an energy management transformation model T, for example, the virtual scenario "WLTP shortened" or the virtual scenario "MCT EBA 2-Cycle" can be used. Examples of such simulation models a, b, c, and d include a 3D CAD model, crash models, and vehicle dynamics simulation models. Vehicle simulation using simulation models is well-known in the art, and there are numerous providers of such models.

[0106] Technical properties X1 A, X2A, ... , X1 B, X2BC, ... , X1 BC, X2C, ... , X1 D, X2D, .. of vehicle specifications that can be used to train a crash transformation model T include, for example, the vehicle mass, vehicle height, vehicle width, ground clearance, or battery size. Please also refer to Table 1 for further information.

[0107] Technical properties X1 A, X2A, ... , X1 B, X2BC, ... , X1 BC, X2C, ... , X1 D, X2D, ... of vehicle specifications used to train an energy management transformation model T include, for example, the vehicle mass, the driving resistance, and the usable battery capacity. The driving resistance of a motor vehicle can be determined, for example, by the drag coefficient cW, the frontal area, and the rolling resistance. In this regard, reference is also made to Table 2.

[0108] During the initial simulation in step 101, default values ​​for the vehicle configuration are preferably used. In a second step 102, values ​​of physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. of the vehicle 1 to be analyzed are derived from the operating data. With respect to a crash transformation model T, such physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. can be, for example, the mean value of the impulse or cabin intrusion during a crash. For an energy management transformation model T, these can be the range and energy efficiency. Furthermore, with regard to the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. for the respective transformation models T, reference is made to Table 3.

[0109] In a third work step 103, measurement data of physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,... of at least one real vehicle 3 are recorded. The measurement data preferably originate from real operation of the at least one real vehicle 3, preferably on a test bench or on the road.

[0110] In a fourth work step 104, the derived values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. are preferably compared with the measured data of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,..

[0111] In a fifth step 105, the simulation models a, b, c, d are calibrated. This means that the values ​​of the technical properties X1A, X2A,..., X1B, X2BC,..., X1BC, X2C,..., X1D, X2D,... and / or other parameters and coefficients of the simulation models a, b, c, d are changed to approximate the actual measured data.

[0112] The first to fifth work steps 101 to 105 are preferably repeated until a deviation of the derived values ​​is less than a defined limit value. The limit value preferably depends on the respective physical parameter Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... under consideration. Furthermore, typical limit values ​​are preferably 0.5% to 10%, preferably up to 5%, above and / or below the values ​​of the measured data. The derived physical parameters Y1A, Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... are preferably scalar key figures that are derived from the operating data in data post-processing.

[0113] Using the first to fifth work steps 101 to 105, the significance of the individual technical properties X1A,X2A,..., X1B,X2BC,..., X1BC,X2C,..., X1D, X2D,.. of a vehicle, which are usually summarized in a vehicle specification, can also be examined for various physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,..

[0114] In a sixth step 106, the vehicle specifications that will subsequently be used for further operational simulations are selected using a statistical experimental design. The statistical experimental design can also take into account the previously determined significance of the individual technical properties X1A, X2A,..., X1B, X2BC,..., X1BC, X2C,..., X1D, X2D,... for the respective physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,...

[0115] In a seventh step 107, the selected vehicle specifications of the vehicle 1 to be analyzed are recorded.

[0116] In an eighth work step 108, operation of the vehicle 1 to be analyzed is simulated using simulation models a, b, c, d. The simulation models a, b, c, d were determined, as described above, on the basis of measurement data from a real vehicle 3. These simulations preferably also take place during defined driving maneuvers and / or in at least one defined scenario. The simulation is preferably carried out for all recorded vehicle specifications and also for all virtual scenarios that are relevant with regard to the respective physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,... to be determined. The simulation models a, b, c, d represent a virtual vehicle 2 that corresponds as closely as possible to the vehicle 1 to be analyzed.

[0117] In a ninth work step 109, values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... are derived from the operating data. For this purpose, the operating data preferably undergoes data post-processing. In particular, scalars are generated from the parameter-discrete signals, in particular time-discrete signals, of the operating data as values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... Event-based physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... can also be generated from time-based operating data.

[0118] In a tenth work step 110, the values ​​of the physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D, ... and the respective associated vehicle specifications are provided, in particular at a data interface. The associated vehicle specifications refer to those vehicle specifications that were used to simulate the vehicle operation from which the physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D, ... were then derived.

[0119] In an eleventh step 111, adjustment calculations, in particular artificial neural networks, are then trained, each of which forms one of the submodels A, B, C, D of the respective transformation model T. Preferably, physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. derived from different simulation models a, b, c, d are each converted into different

[0120] Adjustment calculations are read in. The adjustment calculations can be artificial neural networks or regressions.

[0121] For training, the technical properties X1A,X2A,..., X1B,X2BC,..., X1BC,X2C, ..., X1D, X2D, ... of the vehicle specification are entered on the input side of the adjustment calculation, for example an input layer of an artificial neural network, and the corresponding values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B, ... , Y1C,Y2C,..., Y1D,Y2D, .. are entered into the adjustment calculation on the output side of the adjustment calculation, in particular an output layer of an artificial neural network, which were determined by means of the operational simulations. Preferably, in a twelfth work step 112, values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... determined by means of the trained submodels A, B, C, D are compared with data of physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D simulated by means of the simulation models a, b, c, d.This allows the validity of the trained submodels A, B, C, D of the transformation model T to be validated against the simulation models a, b, c, d for the entire parameter space or for a variation space of the statistical design of experiments (DoE domain).

[0122] Preferably, in a thirteenth work step 113, values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... determined using the trained submodels A, B, C, D are compared with measurement data of physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D of at least one real vehicle, in particular the measurement data from the third work step 103. These measurement data also preferably originate from real operation of the at least one real vehicle, preferably on a test bench or on the road. In particular, real vehicles are used here which are different from the vehicle 3 used to calibrate the simulation models a, b, c, d.

[0123] The sixth to eleventh work steps 106 to 111 are then preferably repeated until the deviation of the values ​​determined using submodels A, B, C, and D from the measured data is less than a defined limit value. The limit value preferably depends on the respective physical parameter Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,... under consideration. Furthermore, typical limit values ​​are preferably 0.5% to 10%, preferably up to 5%, above and / or below the measured data values.

[0124] Preferably, at each repetition, the number of vehicle specifications selected by means of the statistical test plan, for which 108 simulations are carried out in the eighth step, is increased until the submodels A, B, C, D trained by means of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D, which are derived from the operating data generated by means of the simulations, are sufficiently accurate, ie until the above-mentioned deviation lies within the limit value.

[0125] Alternatively or additionally, the parameter space or variation space of the technical properties X1 A ,X2 A ,..., X1 B ,X2 BC ,..., X1 BC ,X2 C ,..., X1 D , X2 DI ... the

[0126] Vehicle specification limited so that submodels A, B, C, D are valid in the remaining variation space.

[0127] The method 100 for training a transformation model T is preferably carried out in a computer-implemented manner. In particular, the seventh to eleventh steps 107 to 111 are carried out in a computer-implemented manner, even if at least one of the steps is not carried out in a computer-implemented manner. The sequence of steps specified above is only exemplary and can be modified for different applications.

[0128] Figure 3 shows an embodiment of a computer-implemented method 200 for determining physical parameters Y1 A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1c,Y2c,..., Y1 D,Y2D,.. of a vehicle 1 to be analyzed using a transformation model T.

[0129] Again, this method 200 is explained for a crash or energy management transformation model T. However, like method 100, this is also applicable to other transformation models T.

[0130] In a first work step 201, a vehicle specification of the vehicle 1 to be analyzed is recorded.

[0131] In a second step 202, values ​​of the physical parameters Y1 A ,Y2 A ,..., Y1 B ,Y2 B,..., Y1 C,Y2C, ..., Y1 D,Y2D,.. of the vehicle 1 to be analyzed is determined using the trained sub-models A, B, C, D of the transformation model T on the basis of the vehicle specification of the vehicle 1 to be analyzed. The analyzed vehicle 1 is not identical to the vehicle to be analyzed in method 100, but rather is a vehicle of the same type as the vehicle 1 to be analyzed in method 100. This means that the vehicle specifications of the vehicle 1 to be analyzed in method 200 do not have to be identical to the vehicle specification of the vehicle 1 to be analyzed from method 100. A type of vehicle is, for example, a sedan, an SUV, a van, a compact car, etc. By using vehicles 1 of the same type during training and analysis, the validity of the transformation model T can be guaranteed.

[0132] In a third work step 203, the determined values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. of the vehicle 1 to be analyzed are output, in particular at a data interface or a user interface.

[0133] Preferably, the method 200 is carried out using a single transformation model T. Thus, for example, in a first pass, physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D, .. are determined with respect to a crash transformation model T, and in a second pass, physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. are determined with respect to an energy management transformation model T.

[0134] Particularly preferably, the various transformation models T, which each output different physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D, .., can be linked by identifying identical or dependent technical properties X1A, X2A, ..., X1B, X2BC,..., X1BC, X2C,..., X1D, X2D, ..., which are input parameters to the transformation models T. With regard to a crash test and energy management, this is, for example, the vehicle mass.

[0135] This allows the respective physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1c,Y2c,..., Y1D,Y2D,... to be determined simultaneously by a crash transformation model T and an energy management transformation model T on the basis of an adapted vehicle specification. Other transformation models T can also be considered additionally or alternatively. This is particularly advantageous if a comprehensive optimization with respect to the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... of a vehicle 1 to be analyzed is to be performed.

[0136] With regard to the example of considering a crash transformation model T and an energy management transformation model T, a simultaneous prediction of range and crash safety can be made in real time based on the defined physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,... for any specifications (within the scope of the transformation models T). This also preferably allows an assessment of target achievement, for example, maximum range versus maximum safety, to be made at the overall vehicle level via averaging or extreme value weighting. This also allows a determination of the sensitivity, ie the dependence of the physical parameters Y1A, Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. on influencing factors, and the significance, ie the relative impact of all influencing factors on a physical parameter Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. possible to prepare an optimization.

[0137] Finally, a simultaneous optimization of the physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,... of the energy management and crash tests can be performed using the transformation models T by changing the influencing factors, particularly the vehicle specifications. This is referred to as "constraint multiobjective optimization."

[0138] Figure 4 shows an embodiment for determining vehicle specifications of a vehicle 1 to be analyzed based on physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,.. by means of a transformation model T.

[0139] In a first step 301, values ​​of a plurality of physical parameters Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,... of the vehicle 1 to be analyzed are recorded. Again, the vehicle 1 to be analyzed is not necessarily identical to the vehicle used to train the transformation model T. Rather, as explained above, they should simply be of the same type.

[0140] In a second step 302, a vehicle specification is determined using a gradient descent method based on default values ​​for the vehicle specification. Within this gradient descent method, method 100 is executed multiple times to determine the vehicle specification.

[0141] In a third work step 103, the vehicle specifications of the at least one vehicle 1 to be analyzed are output, in particular at a data interface or a user interface.

[0142] Figure 5 is an embodiment of a system 10 for training a transformation model T.

[0143] Preferably, the transformation model T has a first interface n , in particular a data interface, for detecting vehicle specifications of a vehicle 1 , in particular of the vehicle to be analyzed.

[0144] Furthermore, the system 10 preferably has means for simulating 12 an operation of the vehicle 1, preferably in at least one defined driving maneuver and / or at least in one defined scenario, by means of simulation models a, b, c, d taking into account measurement data of at least one real vehicle 3, wherein the simulation models a, b, c, d represent a virtual vehicle 2, wherein the simulation is carried out several times with different vehicle specifications of the vehicle 1 and wherein the operating data are generated.

[0145] Furthermore, the system 10 preferably comprises means for deriving 13 values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D, ... of the vehicle 1 from the operating data.

[0146] Further preferably, the system 10 has a second interface 14, in particular a data interface, for providing the values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C, ..., Y1D,Y2D,.., which were generated by means of the simulation models a, b, c, d, and the respectively associated vehicle specifications for training adjustment calculations, in particular artificial neural networks, wherein each adjustment calculation forms the basis for one of the sub-models A, B, C, D, wherein physical parameters Y1A,Y2A,..., Y1B,Y2B, ... , Y1C,Y2C,... , Y1D,Y2D, ... derived from different simulation models a, b, c, d are each read into different sub-models A, B, C, D.

[0147] Figure 6 shows an embodiment of a system 20 for determining physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D, ... of a vehicle 1 to be analyzed for different vehicle specifications by means of a transformation model T.

[0148] Preferably, the system 20 has an interface 21, in particular a data interface, for detecting a vehicle specification of the vehicle 1 to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties X1A,X2A,..., X1B,X2BC,..., X1BC,X2C, ..., X1D,X2D,... of the vehicle 1 to be analyzed, and for outputting the values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D, ... of the vehicle 1 to be analyzed.

[0149] Furthermore, the system 20 preferably comprises means for determining values ​​of the physical parameters Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D, ... of the vehicle 1 to be analyzed by means of a plurality of different sub-models A, B, C, D of the transformation model T on the basis of the values ​​of the technical properties X1A,X2A, ..., X1B,X2BC,..., X1BC,X2C,... , X1D, X2D,...

[0150] It should be noted that the embodiments are merely examples and are not intended to limit the scope of protection, application, or structure in any way. Rather, the preceding description provides the skilled person with a guide for implementing at least one embodiment. Various modifications, particularly with regard to the function and arrangement of the described components, may be made without departing from the scope of protection as defined by the claims and equivalent combinations of features.

[0151] List of reference symbols

[0152] 1 vehicle

[0153] 2 virtual vehicle 3 real vehicle

[0154] A, B, C, D submodel

[0155] T transformation model a, b, c, d simulation model

[0156] X1, X2... technical property Y1, Y2... physical parameters

[0157] 10 System for training a transformation model

[0158] 11 first interface

[0159] 12 Means of Simulation

[0160] 13 Means of derivation 14 second interface

[0161] 20 System for determining physical parameters

[0162] 21 Interface

[0163] 22 means of investigation

Claims

Patent claims 1. Method (100) for training a transformation model (T) which has a plurality of submodels (A, B, C, D) for determining physical parameters (Y1 A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1 C ,Y2 C ,..., Y1 D ,Y2 D ,...) of a vehicle to be analyzed for different vehicle specifications, comprising the following work steps: Recording (107) vehicle specifications of a vehicle (1), in particular of the vehicle to be analyzed, wherein each vehicle specification is represented by an expression of the values ​​of technical properties (X1 A ,X2 A ,..., X1 B ,X2 B c,..., X1 B c,X2c,..., X1D, X2D,...) of the vehicle (1); Simulating (108) an operation of the vehicle (1), preferably in at least one defined driving maneuver and / or in at least one defined virtual scenario, by means of simulation models (a, b, c, d) taking into account measurement data of at least one real vehicle (3), wherein the simulation models (a, b, c, d) represent a virtual vehicle (2), wherein the simulation is carried out several times with a different vehicle specification of the vehicle (1) and wherein operating data is generated; Deriving (109) values ​​of the physical parameters (Y1 A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1c,Y2c,..., Y1 D,Y2D,...) of the vehicle (1) from the operating data; Providing (110) the values ​​of the physical parameters (Y1 A ,Y2 A ,..., Y1 B ,Y2 B,..., Y1C,Y2C,..., Y1D,Y2D,...), which were generated by means of the simulation models (a, b, c, d), and the respective associated vehicle specifications, in particular at a data interface; and Training (111) of adjustment calculations, in particular artificial neural networks, using the values ​​of the physical parameters (Y1 A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1c,Y2c,..., Y1 D,Y2D,...) and the respective associated vehicle specifications, whereby each adjustment calculation forms the basis for one of the sub-models (A, B, C, D), whereby physical parameters (Y1 A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1c,Y2c,..., Y1 D,Y2D, ...) are each read into different adjustment calculations.

2. The method (100) of claim 1, further comprising the following steps: Simulating (101) an operation of a virtual vehicle (2) by means of the simulation models (a, b, c, d), preferably in at least one defined driving maneuver and / or in at least one defined virtual scenario, wherein the simulation models (a, b, c, d) depict the virtual vehicle (2), and wherein operating data are generated; Deriving (102) values ​​of physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) of the virtual vehicle (2) from the operating data; Acquiring (103) the measurement data of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) of at least one real vehicle (3); Compare (104) the derived values ​​with the measured data of the physical parameters (Y1 A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1 C ,Y2 C ,..., Y1 D ,Y2 D ,...); and Calibrating (105) the simulation models (a, b, c, d) if a deviation measure with respect to the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) is above a defined limit value.

3. The method according to claim 1 or 2, further comprising the step of: selecting (106) the vehicle specifications used for operational simulation by means of different simulation models (a, b, c, d) for each simulation model (a, b, c, d) by means of a statistical design of experiments.

4. Method according to one of the preceding claims, wherein, when deriving from the operating data which depend on a parameter, in particular are time-dependent, scalars are generated as values ​​of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) by means of data post-processing.

5. Method according to one of the preceding claims, wherein event-based physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) are generated from time-based data when deriving from the operating data by data post-processing.

6. Method according to one of the preceding claims, wherein the parameter space is limited when simulating the operation in such a way that operating states which have a predefined deviation from the prior art with respect to the respective vehicle type do not occur.

7. Method according to one of the preceding claims, wherein the simulated operation of the vehicle (2) covers a plurality of driving maneuvers and / or defined virtual scenarios.

8. Computer-implemented method (200) for determining physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) of a vehicle (1) to be analyzed for different vehicle specifications by means of a transformation model (T), comprising the following steps: Acquiring (201) a vehicle specification of the vehicle (1) to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties (X1A,X2A,..., X1B,X2BC,..., X1BC,X2C,..., X1D, X2D,...) of the vehicle (1) to be analyzed; Determining (202) values ​​of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1c,Y2c,..., Y1D,Y2D,...) of the vehicle (1) to be analyzed by means of a plurality of different sub-models (A, B, C, D) of the transformation model (T) on the basis of the values ​​of the technical properties (X1A,X2A,..., X1B,X2BC,..., X1BC,X2C,..., X1D, X2D,...), wherein the sub-models (A, B, C, D) each contain an assignment rule between vehicle specifications and at least one physical parameter (Y1A,Y2 A ,..., Y1B,Y2B,..., Y1 C ,Y2 C ,..., Y1 D ,Y2 D ,...) (Y1 A ,Y2 A ,..., Y1 B ,Y2 B,..., Y1c,Y2c,..., Y1D,Y2D,...) of the vehicle to be analyzed and are each based on a compensation calculation with regard to simulation results, which result from several operational simulations by means of simulation models (a, b, c, d), which depict a virtual vehicle (2), taking into account measurement data of at least one real vehicle (3) with a different vehicle specification of the virtual vehicle (2), wherein the simulation results of different simulation models (a, b, c, d) and wherein at least one of the adjustment calculations is an artificial neural network, which has preferably been trained by means of a method (100) according to one of claims 1 to 7; and Outputting (203) the values ​​of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1c,Y2c,..., Y1D,Y2D,...) of the vehicle to be analyzed (1).

9. Computer-implemented method (300) for determining vehicle specifications of a vehicle (1) to be analyzed for a set of physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) by means of a transformation model (T), comprising the following work steps: detecting (301) values ​​of a plurality of physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) of the vehicle (1) to be analyzed; Determining (302) a vehicle specification, which is characterized by an expression of the values ​​of technical properties (X1A,X2A,..., X1B,X2B,..., X1C,X2C,..., X1D,X2D,...) of the vehicle (1) to be analyzed, by means of a gradient descent method starting from default values ​​for the vehicle specification, wherein within the gradient descent method, the method (100) according to claim 8 is executed several times in order to determine the vehicle specification; and Outputting (303) the vehicle specification of the at least one vehicle to be analyzed.

10. The method (200; 300) according to claim 8 or 9, wherein at least one physical parameter (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) determined by means of a first sub-model (A, B, C, D) is input into a second sub-model (A, B, C, D) as an input parameter.

11. Method according to one of the preceding claims, wherein a technical property (X1 A ,X2 A ,..., X1 B ,X2 BC ,..., X1 BC ,X2C,..., X1 D , X2 D ,...) is used as an input parameter for several submodels (A, B, C, D).

12. System (10) for training a transformation model (T), which has a plurality of submodels (A, B, C, D), for determining physical parameters (Y1A, Y2A,..., Y1B, Y2B,..., Y1C, Y2C,..., Y1D, Y2D,...) of a vehicle to be analyzed for different vehicle specifications, comprising: a first interface (11), in particular a data interface, for recording vehicle specifications of a vehicle (1), in particular of the vehicle to be analyzed, wherein each vehicle specification is characterized by an expression of the values ​​of technical properties (X1A, X2A,..., X1B, X2BC,..., X1BC, X2C,..., X1D, X2D,...) of the vehicle (1); Means for simulating (12) an operation of the vehicle (1), preferably in at least one defined driving maneuver and / or in at least one defined virtual scenario, by means of simulation models (a, b, c, d) taking into account measurement data of at least one real vehicle (3), wherein the simulation models (a, b, c, d) depict a virtual vehicle (2), wherein the simulation is carried out several times with a different vehicle specification of the vehicle (1) in each case and wherein operating data is generated; Means for deriving (13) values ​​of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) of the vehicle (1) from the operating data; a second interface (14), in particular a data interface, for providing the values ​​of the physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) which were generated by means of the simulation models (a, b, c, d), and the respective associated vehicle specifications for training adjustment calculations, in particular artificial neural networks, wherein each adjustment calculation forms the basis for one of the sub-models (A, B, C, D), wherein physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1C,Y2C,..., Y1D,Y2D,...) derived from different simulation models (a, b, c, d) are each read into different adjustment calculations.

13. System (20) for determining physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1c,Y2c,..., Y1D,Y2D,...) of a vehicle (1) to be analyzed for different vehicle specifications by means of a transformation model (T), comprising the following work steps: an interface (21), in particular a data interface, for detecting a vehicle specification of the vehicle (1) to be analyzed, each vehicle specification being represented by an expression of the values ​​of technical properties (X1 A ,X2 A ,..., X1 B ,X2 BC ,..., X1 BC ,X2 C ,..., X1 D , X2 D ,...) of the vehicle (1) to be analyzed and for outputting (203) the values ​​of the physical parameters (Y1 A ,Y2 A ,..., Y1 B,Y2 B ,..., Y1C,Y2C,..., Y1D,Y2D, ...) of the vehicle to be analyzed (1); and Means for determining (22) values ​​of the physical parameters (Y1A ,Y2 A ,..., Y1 B ,Y2 B ,..., Y1C,Y2C,..., Y1D,Y2D,...) of the vehicle to be analyzed (1) by means of a plurality of different sub-models (A, B, C, D) of the transformation model (T) on the basis of the values ​​of the technical characteristics (X1 A ,X2 A ,..., X1 B ,X2 B c,..., X1 B c,X2c,..., X1 D, X2D,...), wherein the submodels (A, B, C, D) each contain an assignment rule between vehicle specifications and at least one physical parameter (Y1 A ,Y2 A ,.. „ Y1 B ,Y2B,... , Y1 C,Y2 C , ... , Y1 D ,Y2 D ,...) (Y1 A ,Y2 A ,.. „ Y1 B ,Y2 B,.. „ Y1c,Y2c,..., Y1 D,Y2D,...) of the vehicle to be analyzed (1) and are each based on a compensation calculation with regard to simulation results which result from several operating simulations by means of simulation models (a, b, c, d) which depict a virtual vehicle (2), taking into account measurement data of at least one real vehicle (3) each with a different vehicle specification of the virtual vehicle (2), wherein the simulation results of different simulation models (a, b, c, d) are included in different compensation calculations and wherein at least one of the compensation calculations is an artificial neural network.

14. 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 or a system (10; 20) according to claim 12 or 13, cause the computer or the system (10; 20) to carry out the steps of a method (100, 200) according to one of claims 1 to 11.

15. Computer-implemented machine learning model, in particular an artificial neural network, for determining physical parameters (Y1A,Y2A,..., Y1B,Y2B,..., Y1c,Y2c,..., Y1D,Y2D,...) of a vehicle (1) to be analyzed for different vehicle specifications, wherein the machine learning model has been configured by carrying out the work steps of a method (100) according to one of claims 1 to 7 for training inputs.

Images