WO2018177526A1

WO2018177526A1 - Robustness analysis for vehicles

Info

Publication number: WO2018177526A1
Application number: PCT/EP2017/057555
Authority: WO
Inventors: Martin PIFFL; Martin Schüssler
Original assignee: Avl List Gmbh
Priority date: 2017-03-30
Filing date: 2017-03-30
Publication date: 2018-10-04

Abstract

The invention relates to a method and a system for analysing the robustness of a type of vehicle that has a plurality of components, the method comprising the following operating steps: simulating the operating behaviour of a plurality of vehicles of the vehicle type with different respective configurations by means of a vehicle model of the vehicle type, each configuration characterising at least one property of a component of a vehicle of the vehicle type; carrying out a regression on the basis of the simulated operating behaviour of the plurality of vehicles; and applying a transformation model on the basis of the regression, the transformation model comprising a correlation rule between the configuration and at least one target variable and being designed to emit a value of a target variable relating to the robustness on the basis of configurations of the vehicle type.

Description

Robustness analysis in vehicles

Description

The invention relates to a method for analyzing a robustness of at least one vehicle of a vehicle class, which has a multiplicity of components, and to a method for analyzing a robustness of a vehicle class whose vehicles have a large number of components. Moreover, the invention relates to corresponding robustness analysis systems.

Quality is one of the most important factors in deciding between competitor products. Therefore, understanding the impact on quality and influencing quality in the right way is critical to the business success of a product and therefore a business.

In addition to the resulting quality requirements on the manufacturer side, quality standards are also specified by the legislator. For example, vehicle manufacturers must be able to guarantee that all vehicles of a vehicle type which are delivered by the vehicle manufacturer comply with emission values.

Another example of quality is the probability of failure of individual components or components of the vehicle. This probability of default has a direct impact on the total cost of ownership of a vehicle, which is very important to the end user. Also in this regard, the entire vehicle fleet of a vehicle type should not exceed a predetermined value.

However, maintaining a consistent quality is a challenge, since each component of the vehicle is subject to certain tolerances and each vehicle is used differently by the respective holder and accordingly ages differently.

Against this background, the robustness of vehicles as an overall system against fluctuations is becoming increasingly important. Such fluctuations can be caused on the one hand by production tolerances, but also by the user behavior and the aging of the components. These variations must be taken into account in the development, calibration and validation of the vehicles, and in particular their powertrains, in order to achieve the desired robustness of the overall system. The classic development process of a vehicle or its drivetrain is insufficient to meet the challenges posed by vehicle fleet fluctuations. To take account of the fluctuations, most of the engineering know-how or experiments relating to the worst case scenario are used.

For example, the calibration of a vehicle concept is performed on one or a few vehicles or powertrains to verify the desired performance and emission characteristics. Variations in the vehicle fleet are taken into account by experience-based development goals or by testing at minimum or maximum conditions. In general, some characteristics of components are individually set to their lower and upper limits with respect to the vehicle fleet. For example, the best possible turbocharger or the worst possible turbocharger can be assumed here and then the system can be validated. This minimal / maximal approach however, can not account for two or more factorial interactions or the impact on the overall vehicle fleet.

Another example is the validation of emission values of a vehicle type. However, the fulfillment of the emission limit values has so far been validated by means of individual vehicles. Development is therefore attempted to take account of experience-based safety margins, such as the so-called "Development Goal." Sometimes, the worst case scenario is supplemented, but orientation towards a general margin of safety is unsatisfactory as the one selected The statutory emission limits will in future be set in relation to the entire vehicle fleet of a vehicle type, which will presumably no longer be met by the introduction of new safety margins.

Basically, the variations within a vehicle fleet can be modeled mathematically by a large number of correlated or uncorrelated random variables. These random variables are subsequently grouped together into a random vector whose multidimensional distribution must be taken into account in the development of a vehicle or powertrain. Each realization or sample from this multi-dimensional distribution then corresponds to a single vehicle or a single drive train with certain production tolerances, certain measurement tolerances and a specific aging history, which is operated under certain conditions.

A set of realizations or samples of the random vector is then called representative of the underlying distribution, if the empirical distribution properties, in particular the empirical quantiles, with the theoretical distribution properties, in particular the theoretical distribution Quantile, agree sufficiently precisely. Unless the theoretical distribution is known, a set of implementations is considered representative if any number of additional implementations do not significantly alter the properties of the empirical distribution.

In order to ensure robustness, in particular the significance, for the entire vehicle fleet, therefore, when developing a vehicle, all technical decisions must be made with regard to a representative quantity

Realizations are made. That is, a vehicle calibration is only robust if this is true for the entire representative quantity

satisfactory fails. Depending on the number of random variables, a representative quantity must contain several million implementations.

Therefore, full consideration of variations in a vehicle fleet in the development process so far causes an experimental effort that can not be accomplished in a reasonable amount of time.

The use of mathematically calculated robustness analyzes is known from various technical fields. For example, EP 3 038 026 A1 relates to a system having a communication device configured to provide a scheduled flight schedule that includes flight details relating to each flight of the scheduled flight schedule, and a current flight plan that includes flight details relating to each flight of the current flight schedule. a robustness analysis module including a simulation-based model to receive the scheduled flight schedule and the current flight schedule and check its robustness, a memory to store program instructions, and at least one processor coupled to the memory and to the The robustness analysis module communicates, wherein the at least one processor is configured to execute program instructions. It is an object of the invention to enable an analysis of the robustness of at least one vehicle of a vehicle genus or of the entire vehicle genus. In particular, it is an object of the invention to accelerate a simulation-based robustness analysis for a vehicle having a plurality of components.

This object is achieved by the method according to the invention and the systems according to the invention according to the independent claims. Advantageous embodiments are claimed in the subclaims. The teaching of the claims is expressly made part of the disclosure.

A first aspect of the invention relates to a method for analyzing a robustness of at least one vehicle of a vehicle type, which has a multiplicity of components, having the following working steps:

Detecting configurations of the at least one vehicle that characterize at least one property of at least one component of the vehicle;

Determining a value of a target relative to the robustness of the at least one vehicle by means of a transformation model having an assignment rule to the configurations of the at least one vehicle and the target, the transformation model being based on a compensation calculation for simulation results based on a performance simulation; in particular, a time-resolved cycle simulation, a plurality of vehicles of the vehicle genus, each with different configurations by means of a vehicle model of the vehicle genus result; and

Output the value. A second aspect of the invention relates to a method for analyzing a robustness of a vehicle type whose vehicles have a multiplicity of components, having the following working steps:

Simulating performance of a plurality of vehicles of the vehicle genre, each having a different configuration, by means of a vehicle model of the vehicle genre, each configuration characterizing at least one characteristic of a component of a vehicle of the vehicle genre;

Performing a compensation calculation based on the simulated performance of the plurality of vehicles; and

Applying a transformation model based on a balancing calculation, the transformation model having an allocation rule between the configuration and at least one target size, the transformation model configured to output a value of a target size related to the robustness of the vehicle genre based on configurations of the vehicle genre.

A compensation calculation in the sense of the invention is a regression process or a pattern recognition process or / and a combination of both in the sense of statistical teaching.

A transformation model according to the invention establishes a relationship between the configurations of the individual vehicle, the group of vehicles or the vehicle type and a target variable or a group of target variables. Preferably, the transformation model is not time resolved. More preferably, the transformation model itself is based, however, preferably on a time-resolved simulation of the operation of a plurality of vehicles. A performance simulation in the sense of the invention is a time-resolved simulation of the operating behavior of a vehicle, drive train or engine. Preferably, it is a cycle simulation. In this case, a specific driving cycle or a selection of driving cycles of the vehicle of the vehicle type representative of a cycle distribution is preferably used. The operating behavior preferably comprises at least one of the variables rotational speed, torque or accelerator pedal position. Additionally or alternatively, the variables speed of the vehicle and / or gear selection can be used.

A target in the sense of the invention is a variable to be analyzed, which preferably depends on the configuration and the operating behavior.

Robustness in the sense of the invention is the ability of a vehicle and / or its components to function reliably taking into account variations of various factors. Examples of such factors are production tolerances, measurement tolerances, different usage, aging, environmental conditions, etc.

A type of vehicle within the meaning of the invention is a multiplicity of vehicles which have at least substantially components of the same type. Preferably, at least those components of the vehicle genus of the same type, for which a robustness analysis is to be performed. A type of vehicle may in particular be the entire vehicle fleet of a vehicle type of a manufacturer.

An attempt in the sense of the invention is a test run of a real or virtual technical device with a specific configuration and a specific operating behavior over a fixed time. The experiment is preferably carried out by means of a performance simulation. A component in the sense of the invention is a component or an assembly of a vehicle.

A vehicle according to the invention is any type of land, water, air or spacecraft, in particular a motor vehicle.

Detecting in the sense of the invention is a reading in of information for further processing. Preferably, the information is provided thereby, in particular via a data interface, the method according to the invention or the system according to the invention.

Output according to the invention is a provision. The provision preferably takes place via a data interface. In particular, however, it is also possible to output information via a user interface, for example a screen.

A configuration in the sense of the invention corresponds to a point in a test room, which is defined by those properties which determine the robustness of the vehicle type.

The invention is based on the finding that a functional relationship can be found between the variation of fluctuation ranges of properties of vehicles, in particular their components, and target variables with regard to the robustness. The invention is based on the approach of evaluating failure statistics of a vehicle fleet in the field, to replace attempts on a (continuous) test bench or time-resolved experiments on a virtual test stand for robustness by an empirical model.

The use of the invention in the robustness analysis allows the robustness of a vehicle, a group of vehicles or a whole To analyze vehicle genus without having the components of the vehicle or the vehicle itself as a physical DUT available. The entire robustness analysis can be carried out mathematically by means of the transformation model.

Compared with existing simulation methods, the number of time-resolved simulations of individual implementations or samples of the operating behavior can be significantly reduced.

For the complete robustness analysis of a powertrain common at the time of application, only a few hundred time-resolved simulations have to be made by using the invention instead of a representative selection of configurations from the test room on the order of 10 ⁶ . The results of the remaining configurations of the representative selection, on the other hand, can be predicted by means of the transformation model. If the transformation model for a given experimental space is known, then even all configurations of distributions within the original range of variation can be predicted by the transformation model without having to perform additional time-resolved simulations.

According to the invention, therefore, a significant acceleration of the computer-aided simulation analysis is achieved. Simulation times of several months can be reduced in this way to a few hours or minutes. So far, it often has been necessary to dispense with a computer-aided robustness analysis in an early developmental stage of a vehicle or a vehicle component on which there was no prototype yet, the robustness analysis with the method according to the invention can already be carried out in the presence of a vehicle model for a vehicle genre. Furthermore, the invention enables a root cause analysis with respect to the robustness of the vehicle genus. Thus, for example, the fluctuation ranges of the individual properties of the analyzed configurations and thus the respective distributions can be continuously changed and the corresponding effects on the robustness can be directly observed. In this way, areas of the test room, which are of high relevance for robustness, can be easily identified and examined.

In an advantageous embodiment, a method according to the invention also has the following working step:

Generating a representative selection of a group of vehicles of the vehicle genus or a representative selection of vehicles of the entire vehicle genus whose configurations are detected.

By considering a representative selection from the experimental space, it can be ensured that the resulting values of the target size in terms of robustness are representative of the total distribution considered.

In a further advantageous embodiment of a method according to the invention, the basic configurations are characterized by using the at least one component and / or the at least one vehicle, in particular by characteristic driving cycles and / or by at least one ambient condition under which use takes place, as further properties.

On the one hand, this embodiment allows the entire vehicle genre to be examined for the influence of external environmental conditions on the components; on the other hand, a single vehicle can also be used determined deterministic tolerances on the influence of external conditions.

In a further advantageous embodiment of a method according to the invention, the use is characterized by load characteristics and dynamic characteristics.

Load characteristics and dynamic characteristics have proven to be particularly suitable for the differentiation of different characteristic driving cycles.

In a further advantageous embodiment of a method according to the invention, this furthermore has the following working steps:

Generating a representative selection of possible configurations of the further properties, these configurations being detected.

Preferably, the representative selection of the possible configurations of the further characteristics may be combined with the representative selection of vehicles of the vehicle genus, such that a representative selection with respect to the distribution of the characteristics of the components of the vehicle and the further characteristics arises.

In a further advantageous embodiment of a method according to the invention, the representative selection is based on a random experiment or several random experiments, in particular a Monte Carlo simulation.

In a further advantageous embodiment of a method according to the invention, a plurality of properties of a multiplicity of components is detected for a configuration, and the transformation model is suitable for determining one or a few target value values with respect to the at least one vehicle from these configurations. According to the invention, therefore, a multidimensional distribution can be mapped to a one-dimensional target size.

In a further advantageous embodiment of a method according to the invention, the simulation or the simulation is based on a selection of configurations, which is selected by means of statistical experimental design within limits of a first distribution of all possible configurations.

In this way it can be ensured that the transformation model depicts the entire test room in which the first distribution is spanned.

In a further advantageous embodiment of a method according to the invention, the representative selection is selected within limits of a second distribution. In particular, the transformation model is suitable for imaging any distribution in which the fluctuation ranges of the properties lie within the fluctuation ranges which were predetermined in the first distribution.

In a further advantageous embodiment of a method according to the invention, the dimension of the first and / or the second distribution is equal to the number of properties which is taken into account in the configurations.

In a further advantageous refinement of a method according to the invention, the target variable is at least one of the following group of target variables: emissions of the vehicle, in particular pollutant emissions in the field, an error message and / or an error entry in the control unit, a repair message, a warning light, a safety message to a vehicle component, in particular a chassis component, a powertrain component, a driver assistance system or a servo drive. Furthermore, a target variable can also be a damage event and / or a Load collective, in particular one or more potential harmful input variables for a component and / or their subcomponents, be. Also, an adaptation of measurement and / or production tolerances in a learning phase, or the final loading of a catalyst at the end of a preconditioning phase may be an objective.

In a further advantageous embodiment of a method according to the invention, the at least one property of at least one component of the vehicle is selected from the following group: a constructive property of the at least one component, calibration of control of the at least one component, measurement tolerances of the component, production tolerances of the component Aging of the at least one component and / or the at least one vehicle.

In a further advantageous embodiment of a method according to the invention, the at least one property is indicated by a parametric / non-parametric probability distribution. In particular, the measurement and production tolerances are given as a parametric distribution and the usage is given as a non-parametric probability distribution.

The features and advantages described above in relation to a method according to the invention also apply correspondingly to a system according to the invention, a computer program according to the invention and a computer-readable medium according to the invention correspondingly and vice versa.

A third aspect of the invention relates to a system for analyzing a robustness of at least one vehicle of a vehicle type, which has a multiplicity of components, having the following working steps: Means for detecting configurations of the at least one vehicle characterized by at least one characteristic of at least one component of the vehicle;

Means for determining a value of a target variable with regard to the robustness of the at least one vehicle by means of a transformation model which has an assignment rule between configurations of the at least one vehicle and the one target variable, wherein the transformation model is based on a regression with respect to simulation results which result from a performance analysis. Simulation of a plurality of vehicles of the vehicle genus, each with different configurations by means of a vehicle model of the vehicle genus result; and

an interface for outputting the value.

A fourth aspect of the invention relates to a system for analyzing a robustness of a vehicle of a vehicle type, which has a multiplicity of components, comprising the following working steps:

Means for detecting configurations of a plurality of vehicles of the vehicle genre, each characterized by at least one characteristic of at least one component of the vehicle;

Means for simulating performance of the plurality of vehicles each having a different configuration by means of a vehicle model (FM) of the vehicle genre, each configuration characterizing at least one characteristic of a component of at least one vehicle of the vehicle genre; Means for performing a regression based on the simulated performance of the plurality of vehicles; and

Means for applying a transformation model based on the regression, the transformation model comprising an ordering rule between configurations and at least one target size, the transformation model configured to output a value of a target size related to the robustness of the vehicle genre based on configurations of the vehicle genre. A means in the sense of the present invention may be designed as a hardware and / or software, in particular a data or signal-connected, preferably digital, processing, in particular microprocessor unit (CPU) and / or bus system. or one or more programs or program modules. The CPU may be configured to execute instructions implemented as a program stored in a memory system, to capture input signals from a data bus, and / or to provide output signals to a data bus. A storage system may comprise one or more, in particular different, storage media, in particular optical, magnetic, solid state and / or other non-volatile media. The program may be arranged to embody or perform the methods described herein that the CPU may perform the steps of such methods and, in particular, may determine a value of a target relative to the robustness of at least one vehicle of a vehicle genre.

Further features and advantages will become apparent from the following description of the embodiments with reference to the figures. Show it:

1 shows a flow chart of a method according to the invention according to the first aspect of the invention;

FIG. 2 is a flowchart relating to creation of various ones. FIG

Sets of samples or realizations from one

Total distribution;

Fig. 3 is an illustration for illustrating a derivation of a

Edge distribution from nonlinear measuring tolerances of an NO _x sensor; FIG. 4 is an illustration of an edge distribution with respect to a use of a vehicle genre; FIG.

Fig. 5 is an illustration of an edge distribution with respect to

Environmental conditions when using vehicles;

Fig. 6 is a flowchart for illustrating an inventive

Method according to the second aspect of the invention; 7 is a graphic representation of a possible result of the methods according to the invention; a further graphical representation of a possible result of the inventive method; and a further graphical representation of a possible result of the method according to the invention.

In the following, the methods according to the invention are shown in the overall context of a determination of robustness. The determination of the robustness of a vehicle genre is subdivided to illustrate this in seven steps. However, the person skilled in the art knows that the described procedure could also be structured in another way.

In a first step, income is made on work assumptions. These preferably include:

• Definition of robustness analysis objectives (eg CoP, ISC, OBD, durability studies preferably combined with parameter optimization)

• Defining the simulation platform for a performance simulation (eg HiL, SiL, ETB, CTB, ...) Defining the target values (for example pollutants, defect entries, ...), the test cycles (eg WHTC-, WHSC, RDE, ...) and the variable parameter

Define engine and / or EAS preconditioning measures (e.g., warm-up, NH3 loading, ...)

Set the ECU initialization (for example, learning phases, adjustment activity, etc.)

Modeling of production tolerances by:

o Application of a normal distribution N (μ, σ) modeling hardware (o = toi I 4), sensor and actuator tolerances (o = toi I 2 or o = tol l 3)

o Application of a beta distribution with skew tolerance behavior Modeling of a usage variability by:

o Application of a non-parametric Gaussian modeling kernel approach to model correlated environmental data (temperature, pressure and humidity)

Usage modeling by one or more given cycles or by an existing usage distribution depending on load and dynamics. An example of such a usage distribution is shown in FIG. In this case, preferably a so-called "depth-outlyingness-quantile-rank paradigm" is used to select realizations or samples from this distribution)

Determine how aging of the components should be included in the modeling.

o Alternative 1: Aging is treated as an independent input of the analysis. In this case, the aging is treated like a manufacturing tolerance.

Alternative 2: Aging correlates with production tolerances and / or variability in use. Then this must be represented by aging models. In this case, too, the input variables of the aging model to be monitored have to be monitored be determined. To take account of accelerated aging, a respective observation period is preferably divided into reference points. For each interpolation point, the engine model is corrected according to the aging models

In a second step, the modeling structure is defined. These preferably include:

Generation and / or allocation of so-called access points to which factor variability, in particular the production and measurement tolerances, can be transferred to the model. An example of the factor variability of a NOx sensor is shown in FIG. 3a

• Access points can therefore be distinguished as follows:

o offset errors (e.g., hardware dimensions, sensor errors), i.e., a given parameter or signal in the model is preempted with a constant value

o factor error (e.g., actuator functionalities, sensor errors), i. a given signal in the model is multiplied by a value close to 1

o nonlinearity errors (e.g., dependent functionalities, sensor errors), i. a defined signal deviates as a function of one or more signals; a non-linear error refers to a function between the specification limits USL, LSL for a particular motor (a performance simulation) for which it is kept constant. An example of this is shown in FIG. 3b

• Test the access point implementation and data flow throughout the simulation environment

• Check the correctness of the start conditions for test runs and correct the data export

• Check the access points for the correct dependencies, preferably by a One-Factor-a-Time (OFT) simulation, where each factor is related to its lower (LSL) and upper (USL) Specification limit is changed. This happens especially at demanding stationary operating points or in a transient cycle.

From the above-mentioned variability for use, aging, environmental conditions and production and measurement tolerances, edge distributions F ¹ ,..., F ^k result respectively. These are preferably combined in an overall distribution F ^{1 * '* k} . This overall distribution spans a trial space X ^k .

In a third step, a screening process is performed on the total distribution F ^{1 * '* k} . The screening procedure serves to validate the validity of the simulation environment at the boundaries of the experimental space X ^k . High variability areas can be identified for efficient space-filling design. In this case, the screening data will be used to create a room-filling design. In addition, a first overview of which factor is significant for which system response or target size can be obtained.

On the other hand, the screening method can be used to save long-lived preconditioning and learning phases of components of the vehicle and thus reduce the tolerance ranges of the properties of the components. Because based on the data obtained in the screening process, second-order regression models can be created for the dependencies of target variables, based on which a change in the tolerances during the learning phase of a control of a component or a loading of a component as target variables Catalyst during a conditioning phase can be predicted.

Preferably, the screening design should be able to examine all factors at the three levels "nominal,""LSL," and "USL." For example, the screening method of Jones and Nachtsheim, Journal of Quality Technology, Vol , pp. 1 -15, 201 1 are used. By means of the screening method, the significance of individual factors can be examined by a t-test test procedure. This study preferably incorporates linear dependencies, quadratic dependencies, and interaction effects between two factors, and more preferably, examines the power of each test (ie, the probability of false statements).

In a fourth step, a space-filling experimental design for the total distribution F ^{1 * * k} in the test room X ^{k is} performed. This is necessary because data obtained by the screening method is generally insufficient to adequately represent the dependencies between the characteristics or configurations of a vehicle genre and the targets in a transformation model TM for the entire experimental space X ^k .

Samples or realizations are required for the performance simulation especially in those regions of the experimental space X ^k , in which the dependencies to be examined have a high variability.

Preferably, the performance of the performance simulation 203 is divided into several simulation sections for this purpose:

In simulation section 1, the screening results are used so as to create a space-filling design for the experimental space X ^k , which sets more samples in regions with high variability of the target sizes. The performance simulation is carried out for the samples or implementations defined in this way.

In simulation section 2, the results of the in simulation section 1 are used to determine which regions of the experimental space X ^{k are} not are satisfactorily investigated, ie the data of the target size show serious gaps, so that more data is required.

In these regions of the test room X ^k , samples or realizations are again selected. Such a refined study of individual regions is shown in FIG. The corresponding regions have a higher sample density and are framed in FIG. 6 by means of dashed lines.

The simulation section 2 can be repeated as often as required until all dependencies have been satisfactorily investigated.

In a fifth step, one or more transformation models TM are generated 204, preferably for each observed target variable. The respective transformation model TM is valid only within the examined tolerance limits.

The transformation models TM are preferably based on regression and pattern recognition approaches, in particular multi-layer perceptrons, i. neural networks, created.

For example, training algorithms that can be used for these neural networks are included as part of the "Neural Networks Toolbox" offered by Mathworks ©, and the performance of the transformation models TM is assessed by dedicated performance behavior simulation data, which is preferably considered sufficient if the coefficient of determination R ^{2 is} greater than 0.6 with respect to all data.

In a sixth step, a Monte Carlo simulation is carried out to generate random values, in particular with regard to configurations. The biggest challenge here is the "complete" integration a given uncertainty in the robustness analysis. "Complete" means that the included variability is examined representatively.

According to the "law of large numbers", a series of random samples represents the underlying distribution if the deviation between empirical and theoretical quantiles is sufficiently small.

Such a series of random samples or realizations can be generated from the original distribution F ^{1 * * k} by means of the Monte Carlo simulation 101. These samples or implementations are entered 102 into the transformation model (s).

For these representative samples or implementations, the distribution of the respectively desired target variable can be calculated by means of the transformation model or models TM, 103, 205.

By using the transformation models TM, the samples can be processed immediately and therefore the distribution of the target size can be calculated immediately without long simulation times.

Using the results of the calculations, a sensitivity analysis can be performed. In such a sensitivity analysis, the transformation models TM according to the invention have the advantage that other total distributions F ' ^{1 * "* k} , F" ^{1 * "* k} within the test room X ^{k can} be analyzed immediately without having to perform further time-resolved performance simulations ,

For example, the significance of the variation of individual properties can be examined by restricting their variability to sub-intervals. The resulting new total distribution F ' ^{1 * "* k} , F" ^{1 * "* k} can be determined by means of the at least one transformation model TM, which was derived for the original overall distribution F ^{1 *""* k} , and a renewed Monte Carlo - Simulation for the new overall distribution will be examined. In this way, the target quantities can again be determined for the new total distribution F ' ^{1 * "* k} , F" ^{1 * "* k} and then the difference obtained in comparison to the original total distribution F ^{1 *""* k} can be measured.

This difference is expressed without limiting the generality of the results for the vehicle or the vehicle type in a change in the probability for a given event.

In a seventh step, the results are interpreted.

The probabilities or probability distributions obtained for a given event based on the above-described operations must be interpreted and / or processed with respect to the objectives of the robustness study

For example, the objective of the robustness analysis could be to assess the product compliance (CoP) failure probability for a high-performance on-road application ("on-road heavy-duty engine").

The Product Compliance Testing Procedure is a random sampling test that successively tests engines in the production line for emission compliance. Starting with three engines, the test method evaluates the test statistics based on the legally regulated pollutants. Depending on the result of the test statistics, the product conformity check passed, failed or an additional motor is taken from the production line.

Probability distributions obtained may be based on pass, fail or expected number of engines for which the product compliance test must be tested.

Figures 8 and 9 show the probability distributions of product conformity testing for high NOx sensors (Figure 9) and lower Quality (FIG. 8), which can be determined by means of the methods 100, 200 according to the invention.

Another example of the use of the teaching of the invention is the question of how to change a tolerance with respect to a property in order to obtain a satisfactory probability distribution of the target size. Another example is the assessment of the aging of the components with respect to a given tolerance. In this case, the probability distribution of the target variable could, for example, relate to temperature peaks which considerably influence the aging process of a component. For example, this probability distribution could be interpreted in terms of expected repair and warranty costs.

The methods 100, 200 according to the invention preferably relate to partial aspects of the determination of the robustness of a vehicle genre described above. Hereinafter, preferred embodiments of the inventive method 100, 200 are explained in detail, in particular with reference to Figures 1, 2 and 6.

1 shows a flow chart of an embodiment of a method 100 for analyzing a robustness of at least one vehicle of a vehicle genus.

Starting point for this embodiment are the total distributions pr ^{... * k} _^ p.-r ^{... *} ^ ρ ..- r ^.. - ^k These total distributions arise from dependent or preferably independent random variables Xi, ..., X _k Edge distributions F ¹ , ..., F ^k , which each relate to an aspect of possible variations in relation to a vehicle of a vehicle genus.

Such edge distributions F ¹ ,..., F ^k with respect to individual aspects of the variations of a vehicle are shown in FIGS. 3 to 5. FIG. 3a here represents a fluctuation with respect to production tolerances or measurement tolerances of a component of a vehicle which is generally subject to a parametric distribution, in particular a normal distribution.

In general, the production tolerances are characterized by an upper and lower specification limit (LSL or L, Upper Specification Limit - USL or U), which is generally +/- 4 o, ie four times the standard deviation. This means that with 100,000 random samples you can expect 6 pieces to be outside these limits. Sensors generally speak of measurement tolerances, which are caused for example by an offset and / or a gain and / or a non-linearity error. This definition corresponds to ISO standard 1 1462.

The measurement inaccuracy caused by this is generally stated by the manufacturers by specifying the limit of +/- 2 o, ie twice the standard deviation, as the lower specification limit or upper specification limit. This means that 5 pieces out of 100 random samples lie on average outside these limits. This definition corresponds to DIN standard 1319.

Some manufacturers of measuring sensors, such as Bosch, on the other hand, specify the limits at +/- 3 o, ie three times the standard deviation. Outside its limits, there are only 3 pieces out of every 1,000 random samples on average.

Such an edge distribution with specification of the lower specification limit LSL and the upper specification limit USL is shown in FIG. 3a. A typical specification of the measurement tolerances of a NO _x sensor is as follows: - At a concentration of NO _x <100 ppm, the tolerance range is +/- 10 ppm at +/- 3 o;

- At a NO _x concentration of> 100 ppm, the tolerance range is +/- 10% at +/- 3 o.

From these specifications of measurement tolerances and / or production tolerances, an error of the output signal can be derived, as shown in Fig. 3b. The respective exemplary random samples from FIG. 1 a cause, as indicated by the two curved arrows, the respective error curves as a function of the NO _x concentrations in the exhaust gas.

Similar distributions can also be used to map aging processes of individual components. In this case, aging is an independent input to the robustness analysis.

Alternatively, aging may be correlated with production tolerances and usage variability. For this purpose, separate aging models are preferably created. This includes the setting of input variables of the aging model, eg. b. a mileage. In the performance simulation, the individual models of the components must then be corrected based on the aging model (s).

FIG. 4 illustrates a distribution in terms of usage variations of vehicles of a vehicle genre. FIG.

However, as indicated in the figure, such a distribution is made on the basis of raw data collected on existing vehicles in the field. The raw data can hereby be recorded on vehicles which do not belong to the same vehicle type as those with respect to which a robustness analysis is subsequently carried out. Preferably, for a collection of raw data, several hundred to several thousand Evaluated rides. In this case, as measuring channels preferably the vehicle speed, a covered mileage, the engine speed, a torque, an accelerator pedal position, a brake signal, a clutch signal, a clutch position, a height, an ambient temperature and an ambient humidity are taken into account.

Preferably, a distribution is generated from these data, which relates to the distribution density of individual driven cycles with respect to the load and the driving dynamics, as shown in Fig. 4.

In FIG. 4, in each case regions are bounded by the three lines, in which 99.99% or 99.73% or 95% of the driven cycles are contained. From the density distribution of the cycles in terms of load and dynamics, a 17 hour reference cycle is determined. Significant areas are identified in the density distribution, which are designated by S in FIG. 4, in which malfunctions of the components of the vehicles, in particular of the engine, occur more frequently. In the present embodiment, the reference cycle is preferably shortened to these significant ranges, leaving a shortened reference cycle R of about 30 minutes. This shortened cycle is assumed to represent the operating conditions under which all vehicles of the vehicle type are operated. Alternatively, however, it may also be provided that the vehicles of the vehicle type are also differentiated with regard to different driving cycles.

Fig. 5 illustrates an edge distribution for the variability of environmental conditions, in the illustrated case air density, temperature and absolute humidity, under which the vehicles of the vehicle type are operated. These data can also be derived from the raw data obtained in the field.

Other aspects that influence the robustness of a vehicle can be taken into account by further edge distributions F ^x . The individual marginal distributions in F ^1, F ^k, if any, ^{* k} summarized above correlation structures to a cumulative distribution F ^{1 * '",} which is clamped in a k-dimensional test room X ^k. Such a test room X ^k in Fig. 2 shown in three dimensions.

In an exemplary robustness analysis, the risk of a faulty activation of 24 error codes or error messages of a vehicle is examined. Due to the various components of the vehicle involved in the determination of such error codes, in particular the sensors, the production and measuring tolerances 38 result in random variables, ie 38 dimensions.

Considering the aging effects of the components adds two more dimensions or random variables.

The use of vehicles of the vehicle category is taken into account by a representative driving cycle. If larger fluctuations occur during use, a random variable can also be provided here. Finally, in the example, the environmental conditions are also taken into account by a random variable, in particular a two-dimensional random vector.

In the example, this results in a test room with 42 dimensions.

A graphical representation of an overall distribution, reduced to three dimensions for better representability, is shown in FIG. 2 as a distribution curve F ^{1 * * k} , which is spanned in the experimental space X ^k .

Each realization or sample from the total distribution F ^{1 * * K} corresponds to a vehicle of the vehicle type with a specific configuration. Such a configuration is determined, for example, by the properties of the components of the respective vehicle, in particular their specific production and / or measurement tolerances, use of the components of the vehicle or the vehicle itself, in particular in the form of a specific characteristic driving cycle, certain environmental conditions under which Vehicle was operated, as well as a certain aging structure of the components of the vehicle or the vehicle itself.

Depending on which properties of the vehicle genre are given as edge distribution or which properties a vehicle genus has, and depending on which properties are determined deterministically, different overall distributions F ^{1 * * k} , F ' ^{1 * "* k} , F" result. ^{1 * "* k} .

As shown in FIG. 1 and FIG. 2, from the respective total distribution F ^{1 * k} , F ' ^{1 * K} , F " ^{1 * * k,} a group of vehicles is selected in operation 101 a, 101 b, which preferably is a representative one Selection of samples from the respective total distribution F ^{1 * * k} , F ' ^{1 * "* K} , F" ^{1 * * k} represents.

As an alternative to an entire vehicle category or a group of vehicles from the vehicle category, individual vehicles can also be analyzed with the invention. For this, preferably all available variables of properties with respect to the components of the vehicle are determined deterministically. In this case, a single vehicle is defined with very specific production tolerances and / or measurement tolerances, which is operated, for example, taking into account a distribution of ambient conditions or is also operated taking into account a distribution of driving cycles. In this case, therefore, a vehicle with quite specific characteristics of the components but variable characteristics of use of this vehicle is considered. In this way, by means of the invention, a detailed root cause analysis with regard to the robustness can be carried out. In particular, one variable at a time can be varied with respect to one property, while all other variables are captured and the effects on the target in terms of robustness are analyzed. In addition, design validation plans can be supported by identifying critical areas of application. Finally, by changing the configurations while observing the evolution of the targets, an optimization can be made.

As described above, a representative selection from the respective overall distribution p ^{1 * * k} , F ' ^{1 * "* k} , F" ^{1 * "* k can be created} by a random experiment, in particular a Monte Carlo simulation preferably so many random samples are required until the properties of the theoretical distribution, in particular the theoretical quantiles, coincide with the properties of the empirical distribution resulting in the random experiment, in particular the empirical quantiles.

The respective representative selection is, as shown in FIG. 1, supplied in a further working step 102 to a transformation model TM.

The transformation model TM has an assignment rule, in particular a function or a table, which establishes a connection between the configurations of the individual vehicle, the group of vehicles or the vehicle type and a target variable or a group of target variables. This target variable is preferably suitable for characterizing or expressing the robustness of the vehicle, the group of vehicles or the entire vehicle type. In particular, the results indicate a probability with which a certain event or events occur. This can be, for example, the undesired occurrence of an error code or an error message, or with what probability an emission limit is exceeded or complied with. In principle, however, any type of target variable with respect to the vehicle can be analyzed with respect to its robustness. In particular, combinations of occurrence probabilities with respect to two or more target variables can also be analyzed. Preferably, the value of the target size determined in operation 103 is therefore a probability value.

The value of the target size is finally output in step 104 as shown in FIG. An output can be made both via a display and by providing it at a data interface. In FIG. 1, for example, a probability of default and a non-default probability are output.

The inventive method 100 for analyzing a robustness of at least one vehicle of a vehicle genus is preferably carried out by a corresponding system 10. This system 10 preferably has means or modules, which software can be implemented technically or hardware-technically. Preferably, such a system 10 comprises means 1 1 for detecting configurations of the at least one vehicle, means 12 for determining a value of a target relative to the robustness of the at least one vehicle by means of the transformation model TM, and an interface 13 for outputting the determined one Value on. Also, making a representative selection of configurations that are captured may preferably be performed by other means of the system 10.

FIG. 7 shows a further result which can be output by method 100 according to the invention in operation 104. These are a multitude of possible error codes or error messages whose probability of occurrence was determined by means of the method according to the invention. For this purpose, a transformation model TM according to the invention has a multiplicity of assignment specifications which correspond to the configurations, which assign to a vehicle, a group of vehicles of a vehicle genus or the entire vehicle genus a target size in terms of robustness, in the present case a probability.

The transformation model TM, which is used in the method 100, 200 according to the invention, is preferably based on a performance simulation of a large number of vehicles of the vehicle type.

The operating behavior simulation is preferably carried out by means of a vehicle model FM, which simulates the operating behavior of each vehicle of the vehicle type.

The vehicle model FM, which maps the powertrain of the vehicles of the vehicle genus or the entire vehicle or another assembly, is preferably operated on a virtual test bench, which performs a so-called hardware in the loop (HiL) simulation.

Further possibilities for generating measurement data or simulated measurement data from a performance simulation, on the basis of which the transformation model TM can be formed, are software in the loop (SiL), model in the loop (MiL), test on a component test bench, on an engine test bench, on a vehicle roll, on a real vehicle, on a powertrain test stand, or an evaluation of combined possibilities for performance simulation, in particular big data.

The virtual test bench is shown in FIG. 6 by a computer 20. On this a method for analyzing a robustness of a vehicle genus 200 may preferably be performed.

In a step 202 of the method 200, a performance of vehicles of the vehicle type is preferably simulated. The aim of this performance simulation is to produce dependencies between different configurations of the vehicle genus and the target variables, which characterize a robustness of a component of the vehicle genus or vehicles of the vehicle genus, thus empirically developing the ZuOrdnungsvorschriften between the configurations and the target variables. From this, the transformation model TM can be derived in a next step.

The ZuOrdnungsvorschriften to be identified preferably cover the entire test room X ^k .

In order not to have to measure the entire test space X ^k , that is to say each individual operating point of the test room X ^k or at least perform a raster measurement of the test room X ^k , it is preferably provided within the limits of the possible variations of the properties of the total distribution F ^{1 * * k} apply a Design of Experiments (DOE) in step 201. By means of this working step 201, which is shown in FIG. 2, the entire test space X ^k can be examined for relationships with one or more target variables with regard to the robustness of the vehicles with a manageable simulation effort.

In this case, known methods can be used for statistical experimental design. For example, as described above, a statistical design by Jones and Nachtsheim, Journal of Quality Technology, Vol. 1 -15, 201 1 are used. In particular, as also described above, the performance simulation can be divided into several simulation sections.

Preferably, in the statistical experimental design used according to the invention, a sensitivity screening is carried out in a simulation section 1, each property being tested at three points of the test room X ^k is examined to get a first impression of the behavior of the individual properties on the target sizes.

In a simulation section 2, a space-filling design of samples is created, which makes it possible to investigate the sensitivity between the properties and the target values in detail on the basis of measurements.

For those areas which were not well covered by space-filling design or in which in the simulation section 3 strong changes are determined in a third step, a further space-filling design created and the resulting random samples in the performance behavior simulation approached to a possible to be able to determine exact regulations.

In the present embodiment, preferably in a test space X ^k with 42 dimensions in this way 600 configurations are identified in operation 201, whose performance simulation by means of the vehicle model FM in operation 203 provides sufficient information to order regulations f (X ^k ) between the configurations and determine the target variables that characterize the robustness of the components or the vehicles or the vehicle genus.

The ZuOrdnungsvorschriften f (X ^k ) can ultimately be determined by a regression in step 204. Preferably, statistical regression methods are used for the regression.

The empirically determined ZuOrdnungsvorschriften can finally be assembled into an empirical transformation model TM, which, if this is applied in step 205, is suitable for outputting a target value with respect to the robustness of the vehicle fleet in step 206a. Alternatively or additionally, the transformation model itself can also be output in step 206b. The transformation model TM determined on the basis of the total distribution F ^{1 * "* k} is suitable for examining other, second, total distributions F ' ^{1 *" * k} , F ^{1 * * k} with respect to the same type of vehicle. Preferably, the fluctuation ranges of the properties of these other total distributions F ' ^{1 * "* k} , F ^{1 *""* k are} within the ranges of fluctuation given in relation to the first distribution (F ^{1 *} - ^* ).

Accordingly, the computer or other system 20 includes means or modules that are hardware or software implemented and configured to execute the method 200. Thus, the computer or system 20 includes means for authoring configurations of a plurality of vehicles of the vehicle genre. This or these means can be set up in particular as a data interface or as a user interface, so that the configurations can be read in. Furthermore, the computer or system 20 comprises means for simulating performance of the plurality of vehicles each having different configuration, means 21 of a vehicle model FM of the vehicle genre, means 22 for performing regression based on the simulated performance of the plurality of vehicles, and means 23 for applying a transformation model TM based on the regression, on. Finally, the computer or system 20 has another interface adapted to output the value of the target size relative to the robustness and / or the transformation model TM itself. Also, making a selection of configurations by statistic design of the experiment which are detected may preferably be performed by further means of the system 10.

The embodiments described above are merely examples that do not in any way extend the scope, application, and structure of the methods and systems of the invention to restrict. Rather, the skilled person is given by the preceding description, a guide for the implementation of at least one embodiment, with various changes, in particular with regard to the function and arrangement of the components described, can be made without departing from the scope, as he from the Claims and their equivalent feature combinations.

LIST OF REFERENCE NUMBERS

TM transformation model

FM vehicle model

F ¹ edge distribution

F ^k k th edge distribution

X ^k k-dimensional distribution space

/ (X ^k ) Code of practice

10 system for analyzing a robustness of at least one vehicle vehicle genus

1 1 means for detecting

12 means for determining

13 interface

20 System for analyzing a robustness of a vehicle genus

21 means of simulating

22 means for performing

23 means to apply

24 interface

Claims

claims

Method (100) for analyzing a robustness of at least one vehicle of a vehicle genus comprising a multiplicity of components, comprising the following working steps:

Detecting (102) configurations of the at least one vehicle characterized by at least one characteristic of at least one component of the vehicle;

Determining (103) a value of a target variable with respect to the robustness of the at least one vehicle by means of a transformation model (TM) having an assignment rule (f (X ^k )) between configurations of the at least one vehicle and the one target variable, wherein the transformation model ( TM) is based on a regression with respect to simulation results resulting from a performance simulation of a plurality of vehicles of the vehicle genre, each having different configurations, by means of a vehicle model (FM) of the vehicle genre; and

Output (104) the value.

Method (200) for analyzing a robustness of a vehicle type whose vehicles have a plurality of components, comprising the following working steps:

Detecting (202) configurations of a plurality of vehicles of the vehicle genre, each characterized by at least one characteristic of at least one component of the vehicle;

Simulating (203) performance of the plurality of vehicles each having a different configuration by means of a vehicle model (FM) the vehicle genre, each configuration characterizing at least one property of a component of at least one vehicle of the vehicle genre;

Performing (204) a regression based on the simulated performance of the plurality of vehicles; and

Applying (205) a transformation model (TM) based on the regression, wherein the transformation model (TM) comprises an allocation rule (f (X ^k )) between configurations and at least one target size, the transformation model (TM) being established based of configurations of

Vehicle genus to issue a value of a target size in relation to the robustness of the vehicle genre (206).

Method according to Claim 1 or 2, wherein the configurations are characterized by at least one further property, in particular use of the at least one component and / or the at least one vehicle, in particular by means of characteristic driving cycles, and / or at least one ambient condition under which use takes place become.

Method according to one of the preceding claims, further comprising the following operation:

Generating (101a) a representative selection of vehicles of a group of vehicles of the vehicle genus or a representative selection of vehicles of the entire vehicle genus whose configurations are detected.

The method of claim 4, wherein the usage is characterized by load characteristics and dynamics characteristics.

6. The method according to any one of claims 3 to 5, wherein the representative selection on a plurality of random experiments, in particular a

Monte Carlo simulation, based.

Method according to one of the preceding claims, wherein for a configuration a plurality of properties of a plurality of components are detected and the transformation model (TM) is adapted to determine from this configuration one or a few target value values with respect to the at least one vehicle.

Method according to one of the preceding claims, wherein the simulation or simulation is based on a selection of configurations which is selected (201) by means of statistical design (DoE) within limits of a first distribution (F - * ^k ) of all possible configurations.

Method according to one of claims 3 to 8, wherein the representative selection is selected within limits of a second distribution (F ^fl * - * ^k ), in which the fluctuation ranges of the properties are preferably within the fluctuation ranges which in the first distribution (F - * ^k ) are given. Method according to claim 8 or 9, wherein the number of dimensions (k) of the first and / or second distribution (F ¹ * - * ^fe ₎ f " ⁱ * - * ^f c) is equal to the number of properties which in the configurations is taken into account.

Method according to one of the preceding claims, wherein the target variable is at least one of the following group of target variables: emission of the vehicle, in particular emission in the field, an error message, in particular a repair message or a safety message relating to a vehicle component, in particular a chassis component

Powertrain component, a driver assistance system, a servo drive, an adaptation of measurement and / or production tolerances in a learning phase, or the final loading of a catalyst at the end of a preconditioning phase.

Method according to one of the preceding claims, wherein the at least one property of at least one component of the vehicle is selected from the following group: a constructive property of the at least one component, a calibration of a control of the at least one component, a measurement and / or production tolerance of at least a component, an aging of the at least one component and / or the at least one vehicle.

Method according to one of the preceding claims, wherein the at least one property by parametric or non-parametric probability distributions (F ¹ , F ² , F ³ , ..., F ^k ) and / or their correlation variables is specified. Method according to one of the preceding claims, wherein the at least one property and / or further property is changed until the value of a target variable or an objective function formed from target variables is at an optimum, in particular absolute optimum, wherein preferably the other properties are fixed.

A computer program comprising instructions which, when executed by a computer, cause it to perform the steps of a method according to any one of claims 1 to 14.

A computer-readable medium having stored thereon a computer program according to claim 15.

System (10) for analyzing a robustness of at least one vehicle of a vehicle genus comprising a plurality of components, comprising the following working steps:

Means (1 1) for detecting configurations of the at least one vehicle characterized by at least one characteristic of at least one component of the vehicle;

Means (12) for determining a value of a target relative to the robustness of the at least one vehicle by means of a transformation model (TM) having an assignment rule ((/ (X ^fe )) between configurations of the at least one vehicle and the one target the transformation model (TM) a regression based on simulation results, which result from a performance simulation of a plurality of vehicles of the vehicle genus, each with different configurations by means of a vehicle model (FM) of the vehicle genus; and an interface (13) for outputting the value.

18. A system (20) for analyzing a robustness of a vehicle of a vehicle genus comprising a plurality of components, comprising the following steps: means (21) for detecting configurations of a plurality of

Vehicles of the vehicle class, which are each characterized by at least one property of at least one component of the vehicle;

Means (21) for simulating performance of the plurality of vehicles each having a different configuration by means of a

Vehicle Model (FM) of the vehicle genus, wherein each configuration characterizes at least one property of a component of at least one vehicle of the vehicle genus;

Means (22) for performing a regression based on the simulated performance of the plurality of vehicles; and

Means (23) for applying a transformation model (TM) based on the regression, wherein the transformation model (TM) comprises an assignment rule (f (X ^k )) between configurations and at least one target size, the transformation model (TM) being established the basis of configurations of

Vehicle genus to issue a value of a target size in relation to the robustness of the vehicle genus.