DE102009059137A1 - Diagnostic method for on-board determination of wear state of fuel cell system in motor vehicle, involves using values and measuring values from operating region, which comprises reduced model accuracy, for adaptation of model parameter - Google Patents

Abstract

The method involves calculating an output value (Ym) from an input value (u) by a behavior model (4), and determining model accuracy by comparison of the values with real measurement values of a component. Continuous adaptation of a model parameter is performed. The components are used for adaptation of model parameter values and the measuring values from a predefined working region, which is divided into multiple operating regions. The values and the measuring values from one of the operating regions comprising reduced model accuracy are used for adaptation of the model parameter.

Description

The The present invention relates to an on-board diagnostic method Determining a state of wear of one of several Components of existing system, in particular a fuel cell system, in a motor vehicle.

From the DE 10 2005 020 821 A1 For example, a method for simplified real-time diagnostics using adaptive modeling is known. In this case, such a method for on-board real-time diagnosis of a mobile technical system is equipped with an adaptive technology in order to take into account the so-called mixing variables occurring during mobile use of the technical system, such as an outside temperature, to such an extent that due to the non-stationary ones obtained during mobile use Data comparison with stationary, z. B. determined in the test bench characteristics is possible, since thereby the influence of the mixing variables can be considered.

at a diagnostic method for on-board or mobile use Systems is the use of a model of each system desired, that the laws the laws of nature as closely as possible. Such the Natural laws following models achieve high accuracy, however, it also has an equally high complexity, so the suitability such exact models due to the limited computational resources z. B. in motor vehicles for mobile or on-board use is not given. Alternatively, such can be computationally intensive process based on an exact model for Determining the state of components of a system by surveying perform a stationary test bench. However, such a survey on a test bench is not only due to the necessary computer capacity expensive and expensive, but can even with very high system quantities, such as z. B. in a fleet, high costs. In contrast this may be due to mobile suitability of such a diagnostic procedure a measurement in a test stand possibly omitted so that the effort and costs regarding the Diagnosis of the wear condition of components of a System can be significantly reduced.

The The present invention addresses the problem of for a diagnostic method for onboard determination of a Wear state of a multi-component system, in particular a fuel cell system in a motor vehicle, an improved or at least another embodiment Specifically, due to the high accuracy of the Diagnostic procedure with reduced but sufficient complexity of a model underlying the diagnostic procedure.

According to the invention this problem by the subject-matter of the independent claim solved. Advantageous embodiments are the subject the dependent claims.

The The invention is based on the general idea in a diagnostic procedure for on-board determination of a state of wear of a multi-component system, in particular one Fuel cell system in a vehicle, in particular for mobile use, a parameterized one for each component a theoretically exact and an empirically approximated Partial model to use existing behavioral model, the empirical or physically interpretable model parameters, wherein for adapting these model parameters by means of a parameter estimation method values calculated by the behavioral model and real measurements a predefined work area divided into several operating areas the component can be used and where also for adaptation the model parameter preferably uses those values and measured values that come from an operating area that has a low Model accuracy has. By replacing especially the complex Subareas of the theoretically exact behavioral model by a empirically approximated partial model, can be reduce the complexity of the behavioral model, so that likewise the computer capacity necessary for the processing can be reduced. At the same time, however, a sufficiently high Accuracy among other things ensured by further measures become.

Of Furthermore, it is advantageous to have a predefined workspace consider. This can prevent that extraordinary Operating conditions, such as a system start of a fuel cell system, adversely affect the adaptation of the model parameters. Because in such unusual operating conditions mostly large fluctuations and non-reproducibility arise, appears an adaptation of the behavioral model for such unusual operating conditions not meaningful. Thus, the predefined workspace includes only those operating areas for which a list of a Behavioral model makes sense at all possible. For adaptation of the model parameters of the behavioral model are thus only Measurements and values used in this workspace are used. By this measure, both the necessary Reduce computer capacity as well as the accuracy of the Increase diagnostic procedure.

Furthermore, such a workspace can be subdivided into several operating areas. A distinction is made between operating areas in which the model accuracy is relatively high and operating areas in which the model accuracy is very low. Values and measured values are preferably collected from operating areas which have a very low model accuracy at the time of the observation. An advantage of the use of measured values from operating areas of lower model accuracy is a further reduction of the necessary computer capacity, since the diagnostic method is not concerned with the processing of data from operating areas, for which there is already a high degree of model accuracy. This can relieve the burden of the diagnostic process and focus on those operating areas that are highly uncertain of model accuracy.

Around to obtain a behavioral model that is sufficiently high Accuracy a reduced complexity and thus a low demand for computer resources, you can in the first step based on the laws of natural laws a physical model for each component derived. Such a model achieves very high accuracy, however, an equally high level of complexity, so the suitability for calculations in mobile use due to limited computer capacity is not given. By replacing complex compute-intensive model sections by empirically determined partial models arises from a theoretical one exact and an empirically approximated submodel existing Behavioral model of the respective component. In this behavioral model model complexity is reduced so much that both sufficient accuracy as well as reduced complexity is reached. Such a diagnostic method with such a behavioral model is for mobile use due to the reduced computer capacity excellent.

Of Further, to account for dynamic Effects of the component just these dynamic effects empirically be determined from measured data and as a correction element the static model upstream. This results in an entire behavioral model in the style of a Viennese model, a special model for non-linear dynamic systems with a dynamic linear Share and a static non-linear share. As a model parameter one chooses such a combined behavioral model prefers model parameters that make up physical and / or chemical, in particular wear-relevant, properties Derive the component, so that through the entire current valid set of model parameters at any time on the operating condition or state of wear of the component are inferred can. Thus, in mobile use of the system of the respective Component status can be retrieved at any time, so that For example, timely maintenance actions are initiated can, a costly measurement on the test bench omitted or the operating strategy to the respective state of wear or operating state of the component can be adjusted. Another Advantage of using such a behavioral model in one Mobile diagnostic procedure is the elimination of a component-individual parameterization within a system of multiple components, so that one Diagnostic procedures especially for high component quantities or system quantities is suitable.

to continuous adaptation of the model parameters will be out at least an input value by means of the behavioral model at least one in real system measurable output value calculated. To these values the respective real measured values of the component are determined. By Comparison of the calculated values with the measured values leaves the deviation of the values calculated by the behavioral model from the actual state of the real component Defining measured values as model accuracy. This model accuracy can in the percentage deviation of the model value from the measured value be specified.

There also from several input values in each case several output values by means of of the behavioral model are too each input value or output value also the respective measured values with respect to the component. this leads to mathematically expressed as a quadruple of n-tuples. In such a multi-dimensional system is the model accuracy by the deviation of the n-tuple of the behavioral model calculated initial values to the n-tuple of the associated given real measured values. As in the collection of real metrics always a measured value noise can occur, depending on the size the measurement noise and the accuracy requirement for example a stochastic minimum variance estimator (Kalman filter) or a bite-eliminating least-square method for noise reduction be used. It should be noted that both those belonging to the input values as well as to the Output values belonging to measured values with a measured value noise can be applied, so if necessary for all n-tuples of the measurements such noise reduction must be performed.

Furthermore, such a behavioral model already takes account of component scattering in such a diagnostic method, so that a component-individual adjustments in such a diagnostic method need not be performed.

Further important features and advantages of the invention will become apparent from the Subclaims, from the drawings and from the associated Description of the figures with reference to the drawings.

It it is understood that the above and the following yet to be explained features not only in each case specified combination, but also in other combinations or can be used in isolation, without the scope of the present To leave invention.

preferred Embodiments of the invention are in the drawings and will become more apparent in the following description explained, wherein the same reference numerals to the same or similar or functionally identical components relate.

It show, in each case schematically,

1 the diagnostic method according to the invention, including the sequences and results of an adaptation step,

2 a sequence of the diagnostic procedure including the data flows in the manner of a flowchart.

As in 1 illustrated includes a diagnostic method 1 essentially a sub-procedure 2 to take account of measured value noise, an adaptation method 3 for adjusting the model parameters, a behavioral model 4 for calculating at least one output value y _m from an input value u, a diagnostic check / error detection 5 and a measuring method 6 to measure the measured values u, y _k . Furthermore, the diagnostic method with a deviation determination 7 equipped, which determines the deviation e between the measured value y _k and the calculated output value y _m . In this case, deviation e is the subprocedure 2 the measured value noise takes into account the measured value noise.

The adaptation method 3 essentially has a step of model error detection 8th and a step of model matching 9 on. After model adaptation 9 is usually a valid parameter set 10 determines from which a correspondingly adapted behavioral model 4 results. Furthermore, due to the parameter set, a characteristic curve 11 be determined. From the parameter set itself or from the characteristic curve 11 Because the model parameters reflect the actual state of the particular component, it can become a wear state 12 and a remaining life 13 possibly with the help of historical data. Through the adaptation process 3 the model parameters are continuously adjusted so that the deviation between the real measured value y _{k of} the component and the calculated output value y _{m is} minimized. This will achieve that behavioral model 4 At all times the real behavior of the component is based and the difference between model and component is minimized as much as possible. Because the behavioral model 4 and whose model parameters are based on the physical and / or chemical properties of the vehicle component, are the determined model parameters for both the behavioral model 4 as well as for the component. This makes it possible to generate a statement about the state of wear or the operating state of the component due to the valid parameter set.

The deviation e is both included in the adaptation process 3 as well as in the diagnostic control / error detection 5 , which due to the deviation e also control signals to the adaptation method 3 passes. Due to the adaptation process 3 finds an adaptation of the behavioral model as described above 4 instead of. The adaptation method 3 is chosen such that it has a diminishing memory for tracking a creeping model parameter drift. It is also due to the sub-procedure 2 robust against noisy input signals. Furthermore, a confidence interval is provided for significance evaluation of the model parameters, and it is fit for purpose for mobile use by recursive estimation of the model parameters.

According to the 2 will be in a first step 14 of the diagnostic procedure 1 read a new measurement data record from the vehicle bus. In a first filter 15 a check is made as to whether the system is within the valid working range, with the aid of the predetermined limits 16 the valid work area. If the system is not in the valid workspace, a new measurement data record is read. If it is in the valid workspace, it will be in a prediction step 17 calculated by means of the component model at least one output value y _m . Using the threshold value calculation 18 becomes a deviation score 19 carried out. If necessary, an error treatment is performed if necessary 20 initiated. In a second filter 21 is due to a measure 22 the current model accuracy judges whether the data set contributes to increasing model accuracy. If the dataset does not increase the model accuracy, then the current dataset is discarded. If the data set increases the model accuracy, then in a subsequent correction step 23 an evaluation of the model error lers and a correction of the model parameters.

In a subsequent update step 24 is both the measure 22 the current model accuracy, the threshold value calculation 18 and the limits 16 Updated the valid workspace. In another verification step 25 Checks whether the model parameters have changed to the last cycle. If this is the case, essentially three actions are carried out, while in the case of unchanged model parameters, a new data record is read from the vehicle bus. A first action 26 results in a saving of the model parameters, whereby a tracking of the model parameter drift is also performed in this action. In this case, a wear of the component and the associated change in the component-internal parameters directly from the consideration of the gradients of the model parameters can be seen. To track component wear, the model parameters are checked for a change after each adjustment. If a deviation to past parameters is detected, then the newly determined model parameters are stored and from these a new wear index by means of the second action 27 calculated. In this case, the wear index consist of a single parameter or derived, such. B. the internal resistance of a fuel cell or it may represent a combination of several parameters or a derivation of a combination of several parameters.

In a third action, the determined model parameters immediately become a characteristic curve 11 the component is generated. It is about the parameterized behavioral model 4 with the aid of previously selected interpolation points, the respective associated output value y _{m is} calculated. As a result, by a suitable selection of nodes of the input value u on the corresponding characteristic 11 can be reconstructed in arbitrary ranges and moreover its dependency on other input values u 'can be represented. This information is similar to the result of a static measurement of the IU characteristic on a test bench, which can be replaced by this procedure. Furthermore, the characteristic can be incorporated directly into the control and operating strategy of the component.

Furthermore, the adaptation of the behavioral model 4 be used to track the component wear, since the adaptation with a certain inertia follows the changes of the real component. Thus, rapid behavioral changes that are outside the expected dynamics of the component's wear-related behavioral changes can be evaluated by evaluating the deviation e over the score 19 be recognized. This is done with every first step 14 , in which a new measured data record is read in by the vehicle bus, the deviation e is evaluated and with one from the threshold value calculation 18 resulting threshold value compared. If the threshold value is reached or exceeded, the component behavior is outside the nominal behavior and a corresponding error treatment can be initiated. By tracking the model accuracy over the accuracy measure, this threshold can be adapted to the model accuracy to provide both high robustness and high sensitivity of the diagnostic procedure 1 to reach.

Such a diagnostic procedure 1 For example, it is applicable to a fuel cell system having a plurality of fuel cells as components. Furthermore, it can also be used, for example, in a vehicle fleet, wherein the vehicle fleet represents the system with several vehicles as components. In the case of a fuel cell system or a vehicle fleet having multiple vehicles having a fuel cell system, the current of a fuel cell or a fuel cell system as the temperature may be used as at least one input value, wherein the output of the voltage of the fuel cell system or the fuel cell can be determined. Accordingly, as a characteristic 11 a current-voltage characteristic through the diagnostic process 1 determined. In this case, such a fuel cell suitable diagnostic method 1 have a model parameter, for example, where appropriate, together with other parameters, the internal resistance of the fuel cell or the fuel cell system reproduces. Starting from the physical and / or chemically exact description of the fuel cell and the fuel cell system, an inventive behavioral model could be based on empirically determined approximations and connection of model parameters with wear-relevant properties of the fuel cell or the fuel cell system 4 be derived.

1

diagnostic procedures

2

indexing method Measurement noise

3

adaptation method

4

model behavior

5

Diagnostic control / fault detection

6

measurement methods

7

deviation determination

8th

Model error detection

9

Model adaptation

10

valid

11

curve

12

wear state

13

Remaining life

14

first step

15

first filter

16

border

17

Prädikationsschritt

18

thresholding

19

deviation review

20

error handling

21

second filter

22

measure

23

correction step

24

update step

25

verification step

26

first action

27

second action

28

third action

QUOTES INCLUDE IN THE DESCRIPTION

This list The documents listed by the applicant have been automated generated and is solely for better information recorded by the reader. The list is not part of the German Patent or utility model application. The DPMA takes over no liability for any errors or omissions.

Cited patent literature

• DE 102005020821 A1 [0002]

Claims (3)

Diagnostic procedure ( 1 ) for the on-board determination of a state of wear of a multi-component system, in particular of a fuel cell system in a motor vehicle, wherein each component has a parameterized behavioral model consisting of a theoretically exact and an empirically approximated partial model ( 4 ), which has at least one model parameter, from which at least one chemical and / or physical, in particular wear-relevant, property value (y _k ) of the associated component can be derived, - wherein at least one input value (u) with the behavioral model ( 4 ) at least one output value (y _m ) is calculated and by comparing these values with real measured values (y _m ) of the component, a model accuracy is determined, - in order to improve the model accuracy, a continuous adaptation of the model parameters is performed, - wherein for adapting the model parameters values and measured values from a predefined operating range of the component subdivided into several operating ranges are used, wherein values and measured values from at least one operating range having a low model accuracy are preferably used to adapt the model parameters. Diagnostic method according to claim 1, characterized that such an operating range with low model accuracy has a modeling accuracy of less than 90%. Diagnostic method according to one of the preceding claims, characterized in that more than 10% of the used for the adaptation of the model parameters Values and measured values from at least one operating range with low Model accuracy come.