WO2007031138A1

WO2007031138A1 - Method and apparatus for predicting reliability

Info

Publication number: WO2007031138A1
Application number: PCT/EP2006/006906
Authority: WO
Inventors: Harald Ihle; Peter Lindenlaub
Original assignee: Daimler Ag
Priority date: 2005-09-13
Filing date: 2006-07-14
Publication date: 2007-03-22
Also published as: DE102005043461A1

Abstract

The invention relates to a method and an apparatus for predicting the reliability of a product. The product is subjected to a load in at least one trial. A relationship between the load in the trial and the load during subsequent use is determined. A measure of the load to which the product is exposed up until a first point in time during the trial is measured as an actual load. The number of faults which occurred up until the first point in time is counted. A measure of the load to which the product will be exposed between the first point in time and a subsequent second point in time during the trial is prescribed as a nominal load. The reliability is predicted using the actual load, the number of faults counted up until the first point in time, the nominal load, an expected number of faults between the first and the second point in time and the relationship.

Description

Method and device for predicting reliability

The invention relates to a method and a device for predicting the reliability of a technical product. Reliability means an average number of errors related to a load on the product during its use.

A method with the features of the preamble of claim 1 is known from WO 2005/033649 Al. A component is subjected to a trial. The test is carried out according to a predetermined test program. This experimental program determines the sequence and duration of each experiment. Preferably, individual tests are carried out as time-tapping experiments, so that the respective specified test duration corresponds to a longer service life of the component. It is calculated which reliability for the component is detectable due to the predetermined experiment.

Reliability prediction methods are also used in

K. Denkmayr: "The AVL Reliability Engineering Process for Motor and Drive Train Development", VDI Report 1713, 2002, p. 27 - 32, K. Denkmayr et al .: "The Load Matrix - The Key to the" Intelligent "Endurance Run, Motortechnische Zeitschrift (MTZ), Issue 11/2003, S. 924 - 928,

T.S. Gates & M.A. Grayson: "On the Use of Accelerated Aging Methods for Screening High Temperature Polymeric Composite Materials", Americ. Inst. Aeronautics Astronautics (AIAA) 99-1296, pp. 925-935.

DE 19713917 A1 describes a method for determining reliability characteristics of a technical installation. For this purpose, a generic, d. H. valid FMEA table generated for several possible configurations of the system. FMEA stands for "Failure Modes and Effects Analysis." Several generic tables are created from this generic table.

From US 2004/0044499 Al a method is known to continuously a technical system, for. As a motor to monitor. Stochastic methods are used to derive reliability characteristics from the results of continuous monitoring.

No. 6,502,018 B1 describes how a model is set up that describes the behavior of a technical product (there: a locomotive) with a sensor. The model is calibrated in a trial using sensor data. The calibrated model is used to monitor the technical system.

The invention has for its object to provide a method having the features of the preamble of claim 1 and an apparatus having the features of the preamble of claims 14 and 15, which allow early prediction of reliability.

The object is achieved by a method having the features of claim 1 and a data processing system having the features of claim 14 and a computer program product solved with the features of claim 15. Advantageous embodiments are specified in the subclaims.

The method predicts as the reliability an average number of errors related to a load of the product during its use. The product is subjected to stress in at least one trial. This trial extends over a given period of time. A quotient of the number of errors related to the load during the test and the number of errors related to the load during use is determined.

A first and a second time of the time period are specified, wherein the second time is after the first time. Measured as an actual load is a measure of the load that the product is exposed to during the test until the first time point. Furthermore, it is counted how many errors occurred in the experiment until the first time. As a target load, a measure is given as to which load the product will be exposed to during the test between the first time and the second time.

An expected number of errors between the first and the second time is given or predicted in dependence on the target load.

Reliability is determined using the actual load, the counted number of errors up to the first time, the target load, the expected number of errors between the first and second times, and the quotient predicted.

The method predicts the reliability using a measured actual load as well as a counted number of actually occurred errors.

The invention shows a way to systematically calculate the predicted reliability. In contrast to the method known from WO 2005/033649 A1, the prediction of the reliability includes actual values which were measured up to the first point in time, and not just plan values which are determined before the start of the experiment.

Thus, the method according to the invention makes it possible to reschedule the continuation of the experiment after the first time, if necessary, e.g. For example, by supplementing the experiment with additional individual experiments.

In contrast to known methods, the method according to the invention provides a reliability characteristic already at the first time and not only after the end of the experiment. However, this reliability characteristic does not refer to the first time but to a future second time. Thus, the first time method using the previous trial history predicts the future reliability at the second time.

The method according to the invention can be carried out at any desired first time and provides a prediction for any second time.

The procedure shows a systematic way to decide at first whether the further trial program can continue until the second time as planned or whether rescheduling is required at first. Redesigning may be required, in particular, if the method predicts a reliability that is less than a predetermined target reliability at the first time. Thus, the continuation of the present until the first time Make measurements dependent. Such a rescheduling can consist, for example, of the fact that at the first time the test program is modified so that the product is subjected to a higher load than originally planned until the second time point.

Preferably, the method provides a prediction of reliability that will be detectable at the second time. This forecast is valid at least if the test program is carried out until the second time as planned at the first time. The method thus makes it possible to avoid "over-testing" in which the product is tested more than necessary.

The quotient between the number of errors related to the load during the test and the number of errors related to the load during use is preferably in the form of a convolution factor. The use of such a convolution factor - or another form of connection - makes it possible to interpret the individual experiments as time-consuming experiments. This saves time and makes it possible to predict reliability earlier.

Furthermore, the design with the quotient makes it possible to use a different measure of the load in the experiment than the load measure to which the reliability is related. The Raffungsfaktor then has a dimension, z. For example, 1000 kilometers per vibration cycle.

Preferably, the quotient is determined empirically in two preliminary experiments. This approach of finding the quotient saves the need to set up and validate an analytical model of the loads in the experiment and in use.

The method can also be used, conversely, to specify a value for the reliability and, at the first time, to determine whether, if so, and if so, at what second time, this predefined reliability value is predicted. The method can also be used to back off when the trial should have begun in order to predict a desired demonstrable reliability.

In the following an embodiment of the invention will be described in more detail with reference to the accompanying figures. Showing:

FIG. 1 shows an exemplary load-time function for a component; FIG.

2 shows a time course of the planned accumulated equivalent travel distance before the start of the experiment;

Fig. 3. the time course of the detectable error rate before the start of the experiment according to the planning of Fig. 2;

FIG. 4 shows the time profile of the accumulated equivalent travel distance according to a modified planning, the first time being in August 2005, as well as dashed lines showing the time course of the equivalent distance of FIG. 2; FIG.

FIG. 5 shows the time course of the detectable error rate according to the equivalent distance of FIG. 4; FIG.

FIG. 6 shows the time course of the actual and planned equivalent journey according to an alternative modified planning, the first time being in August 2005; FIG. FIG. 7 shows the time profile of the detectable error rate according to the time course of the equivalent travel distance of FIG. 6; FIG.

FIG. 8 shows the time profile of the accumulated equivalent travel distance according to the planning of FIG. 6 and, for comparison purposes, the time curve of the equivalent travel distance of FIG. 4 with the first time in November 2005;

9 shows the time course of the detectable error rate in accordance with the equivalent distance of FIG. 8, wherein an error has occurred, as well as dashed lines the temporal of the detectable error rate of FIG. 5 for comparison.

In the exemplary embodiment, the term "error" is understood as meaning a deviation of the behavior of a component from the required behavior According to DIN EN ISO 8402, a viewing unit has an error if one of the quality features of the viewing unit assumes an undesired value.

In the following, a "degradation" of a component is understood to mean a deviation of a quality value from a desired value or desired range., In particular, the degradation may be a physical and / or chemical change of the component. If the degradation of a component has progressed so far that the component does not fulfill its required function at all or only to a limited extent and / or provides cause for a complaint, then there is an error For example, the use of a component can cause its surface to corrode, which is a degradation. If the corrosion layer is thicker than a predefined threshold value and / or the corroded area is larger than a predetermined barrier, then the component has an error.

In the exemplary embodiment, the method is used in each case once for different components of a truck. At least one fault type can occur on each of these components. It is possible that different types of errors occur on the same component. Before applying the method, relevant component-fault type combinations are selected. Preferably, an FMEA is used to determine in advance which component-fault-type combinations are relevant and therefore to be taken into account. Such FMEA is z. B. from WO 2005/033649 Al and DE 19713917 Al known. The process is then executed once for each selected component-fault type combination. It is possible to generate different predictions for different component-error type combinations.

In order to apply the method to a selected component-defect-type combination, several components are tested in the actual experiment. For example, several identical components are tested. The experiment is carried out over a longer period of time. The experiment includes n individual experiments in which each one of the same components is tested. In some of these n individual tests, one copy of the component is installed in each truck. These trucks with the built-in components are driven over longer journeys in the test and / or subjected to extensive stress on test benches. In other individual tests copies of the components in test benches are subjected to the loads which are similar to the loads during the journey. This z. B. components installed in a trial. The surface of the silencer body component can corrode, which is a degradation of the component, which can lead to the failure of the silencer body visibly corroded.

Internal parts of the muffler may wear out due to the load to which the truck is subjected. Abrasion or enlargement of a bore are examples of degradation. If the wear is so great that the muffler when driving audible noise, eg. B. rattling, caused, so there is a mistake. In particular, vehicle-side shrinkages cause material fatigue, which can lead to wear. Fig. 1 shows an example of a stress-time function for a muffler. The time is plotted on the x-axis, and on the y-axis a time-varying vibration acceleration in [g].

For example, within a year, the following individual experiments are performed to investigate the error "muffler causes noise":

- On lanes with high vibration excitation, nl components are tested several thousand kilometers within a few months.

- With n2 components, endurance tests are carried out on test benches in which over a distance of more than 100,000 km per component is covered within six months.

In trucks, n3 components are installed. The trucks are driven under the same conditions as in later use, with several 100,000 km being traveled over two years.

Test runs with n4 components are carried out in a test set-up. In this case, several thousand load changes are carried out for two months each.

To test how strongly the degradation "corrosion of the silencer housing" depends on the loads and how often the error "silencer housing visibly corroded" occurs, for example, test stands and test drives are carried out in a very humid environment, eg a climatic change chamber.

In further individual tests and validation steps loads are simulated. For this purpose, a simulation model is used, which simulates the loaded component on a data processing system. The simulation simulates which degradations trigger these loads. Errors are counted by evaluating simulation results.

The stresses to which the components are subjected in the tests, including the simulations, are, above all, irregular superpositions of different vibrations.

The values measured during the test or computed in the simulations are used to predict values for the reliability of a selected component-fault type combination during productive use. Each prediction thus refers to a component and a type of error.

The reliability is related to a measure of the load to which the component is exposed. As a measure of the load on the component, in this example, both during the tests and during its use, the route in km traveled by a truck with the component is used. Instead of the route can also be z. B. use the hours of operation or the number of load changes as a measure of the load.

The experiment is preferably carried out as a time-trial experiment. In the test, the trucks are driven over rough terrain with many uneven bumps causing vibrations. During later use, the trucks are predominantly on flat stretches, z. B. removed roads, driven. Therefore, in at least two preliminary experiments, a gathering factor r is determined. This refraction factor r is then used in the actual experiment. The gathering factor is the ratio of the degradation of the component by the load during the test and the degradation by the load during later use. If Dl is the degradation in the experiment and D2 is the damage in use, then r = Dl / D2. This refractive factor r generally varies from component-fault type combination to component-fault type combination. The rapping factor r refers to a type of test and may vary from one type of trial to another.

If the distance traveled is used as a measure of the load, then the rapping factor r has the following meaning: A degradation of the component due to a load of x km in the test corresponds to a degradation by a distance of r * x km during later use , To put it simply: x km of the test load as much as r * x km during later use. The product r * x km is referred to below as the equivalent route.

In the first preliminary test, several trucks with the components are exposed to the same loads as in the actual test. For example, these trucks are driven on rough routes in the first preliminary attempt. In the second preliminary test several trucks are driven with the components under the same conditions as in the later use. For example, these trucks are driven in the second pre-trial mainly on developed roads.

In both preliminary tests, at least one parameter for the load to which the component is exposed in the actual test and during use is specified or determined. This load characteristic may be the above-introduced measure of the load to which the reliability relates. For example, the Load characteristic as well as the load measure the distance traveled.

Preferably, however, the load characteristic is a parameter which is more correlated with the degradation and thus the error frequency and is harder to measure than the load measurement. For example, the load characteristic is the number of operations that trigger the degradation. Or the load characteristic characterizes the driveway, z. B. a straight ahead on a developed road, an uphill on a developed road or drive on a dirt road. Two other examples of the load characteristic are the engine speed of the truck and the relative humidity around the muffler. For example, the component is part of the engine or, for other reasons, is exposed to the speed-dependent vibrations of the engine and the air humidity. It is possible to specify several load characteristics, z. B. engine speed and distance.

The range of values of the at least one load characteristic is subdivided into k subregions Tb_l,..., Tb_k. For example, the relative humidity is divided into the k = six ranges 0-20%, 20% -40%, 40% -60%, 60% -80%, 80% -90% and over 90%. Furthermore, a reference is given, z. B. the residence time in the climate change chamber, the route or the travel time.

As a rule, the at least one parameter for the load varies. It is measured to what proportion n_p (1) the load parameter in the first pretrial attempt assumes a value in the partial range Tb_p (p = 1,..., K). The shares are related to the benchmark. A proportion of n_5 (1) of 23% means that the load parameter in the first preliminary test was over 23% of the total travel distance or travel time in the partial area Tb_5. For example, the relative humidity over 23% of the journey in the Tb_5 range was between 80% and 90%. Thus becomes a stress collective determined, to which the components are subjected in the first pre-trial.

For two load characteristics, a proportion n_p, q (l) is measured corresponding to which the first load parameter Kg_a has a value in the partial range Tb_p (a) and the second load parameter Kg_b has a value in the partial range Tb_q (b).

Accordingly, it is measured to what proportion n_p (2) (p = 1,..., K) the load parameter assumes a value in the partial area Tb_p in the second preliminary test.

Furthermore, a parameter for the degradation of the component is specified or determined. The wear that occurs on the component due to the load, or the voltage occurring on the component are examples of this characteristic.

Furthermore, an average value d_p (p = 1,..., K) for the degradation of the component is measured or calculated on the basis of a degradation model. This value d_p is the degradation which occurs on average when the load parameter has a value in the range Tb_p (p = 1,..., K). For example, d_5 is the average corrosion of a given component at a relative humidity of between 80% and 90%. These values d_p (p = 1, ..., k) are the same during the actual experiment and during use, so they apply to each individual experiment as well as to its use. This degradation can z. B. be a measure of the relative degradation.

The corresponding measurements and / or calculations for the same degradation parameter, the same at least one stress parameter and the same reference are performed in a second pretrial experiment. In the second preliminary test, the component is subjected to stress during use. In the actual test and also in the first preliminary test, for example, the truck is driven over a much longer travel distance or travel time in high engine speeds than in use and also in the second preliminary test. Try or stay exposed to high humidity for a significantly longer time.

The degradation D1 in the first preliminary test is preferably calculated according to the following calculation rule:

Dl = Σn_p (l) * d_p p = l

Accordingly, the degradation D2 in the second preliminary test is preferably calculated according to the following calculation rule:

D2 = Σn_p (2) * d_j> p = l

The gathering factor r is then calculated according to the calculation rule

Σn_p (l) * d_p _{r =} Hl _{= -} Eϊ!

⁰² Σnjp (2) * djp p = l.

The gathering factor r depends on the component-error type combination, ie it can vary from combination to combination. It can also vary from single trial to single trial. If n is the number of individual tests, let m be the number of selected component-fault type combinations. Then, with r [i, j], the convolution factor for the i-th single trial (i = 1, ..., n) and the j-th selected combination (j = 1, ..., m).

Preferably, the reliability of a component relative to a type of error is indicated by means of the error rate. This error rate is limited to a given observation period, eg. A defect rate of 100 ppm per year means that, on average, 100 out of every 1 million identical components that are used at the same time are used, for example one year, and in faulty components per one million parts per million (ppm) are within one year, an error of the type of error occurs. The error rate is related to the distance traveled until an error occurs on average as follows: dist a

Error rate =

MTTF

Where dist_a is the average distance traveled in [km] traveled by a lorry over a period of 1 year and MTTF (mean time to failures) is the mean distance traveled in [km] until the occurrence of an error, MTTF depends on the component Error type combination, therefore, the designation MTTF [J] is used for the MTTF value of the jth combination.

As a measure of reliability can also directly use the average distance MTTF.

Preferably, a first quantity indicative of the load of the product during its use and a second size indicative of the load of the product during the trial are used. In this embodiment, the first size used is the distance traveled during use in [km]. This route is in the use of the component, the route that covers the truck with the component.

The second size used in this example in most of the individual tests is the route, which, depending on the design of the individual test, covers the truck or which is simulated on the test bench or in the simulation. In some individual tests, the number of load changes is used as second size.

The gathering factor in this embodiment is the quotient of the load during the test, characterized by the second size, and the load during use, characterized by the first size. For example, the following factors are calculated for the component-error type combination "muffler causes noise":

- One bad-km, which is one kilometer of driving on a high-vibration track, equals 50 equivalent km. The rapping factor r was r = 50.

One km in endurance equals one km during later use, so r = 1.

For the test runs carried out under the same conditions as the later use, r = 1 also applies.

For the test stand run with the number of load changes as the second quantity, a gathering factor r of 5 equivalent km per load change is determined.

In the i-th individual test, the travel distance dist [i, j] (τ) is covered by the component of the j-th component-fault type combination up to a point in time τ (i = 1,..., N). The load in the test over the distance dist [i, j] (τ) corresponds to an equivalent load for the jth component-error type combination dist_a [i, j] (τ) = dist [i, j] (τ) * r [i, j] during use. The total load in the experiment therefore corresponds to an equivalent distance dist_ges a [j] (τ) until the time τ during the use of the component of the jth combination. This equivalent travel distance is calculated according to the calculation rule n dist_ges_a [j] (r) = ^ dist [i, j] (r) * r [i, j].

1 = 1

As the first variable, for example, the number of events that occurred during use of the component, which cause a degradation of the component and which can therefore cause a fault of the type of fault, can be used instead of the travel distance. Such events are z. B. vibration processes, inputs and Abstellvorgänge, Load changes, number of climatic changes, the occurrence of temperature peaks or the contact of the component with corrosive, corrosive or extremely hot or cold liquids. The second size can be used according to the number of events in the actual experiment.

During the test, it is measured or otherwise determined how many faults of which types of faults occur in the tested components. In this case, the determined numbers of errors of a type of error are added. With f_ges [j] (τ) the total number of all errors of the type of error of the j-th combination is called, which occurred at the time τ in total on all copies of the component of the j-th combination, that is summed over all n individual tests.

The method is repeatedly used in the exemplary embodiment for each component-error type combination. Each application takes place at a first time τl. This time τl acts as the first time of the claims. As stated above, the tested components were exposed to a total load equivalent to an equivalent travel distance of dist_ges_a [j] (τl) until time τl. The first time τl is, for example, in the first half of the period over which the experiment extends.

By f_ges [j] (τ2; τl) is meant the total number of all errors of the type of error of the jth combination expected according to a prediction in the period from τl to τ2.

In one exemplary embodiment, it is assumed that no error of the type of error occurs between the times τ1 and τ2, ie that f_ges [j] (τ2; τ1) = 0. This assumption is made to allow a prediction of reliability in the event that no errors occur.

In another embodiment, depending on the planned load, the increase in the degradation of the component in the period between .tau.l and .tau.2 is first predicted. This planned load in this example depends on the equivalent distance traveled or another load factor. Parameter. Depending on this degradation increase, it is predicted how many errors will occur in the period between τl and τ2. For both predictions, models are preferably used. The first model describes the connection between the load and the degradation, the second model the relationship between the occurrence of errors and the degradation.

In each case, a prediction value for the reliability which is achieved when the test is continued up to several second times is calculated. Let τ2 be such a time. Given and used for the prediction is for each of the n individual tests, the target load that will be exposed to the respective component from the time τl to the time τ2, τl is preferably the current time, τ2 is a variable time in the future. In this embodiment, this is a planned equivalent travel distance of dist_plan [i, j] (τ2, τl) for the ith single trial (i = 1, ..., n). If, therefore, in all individual tests, the actual distance traveled until the time τ2 corresponds to the planning at the time τ1, then for i = 1, ..., n and j = 1,..., M: dist [i, j] ( τ2) = dist_plan [i, j] (τ2; τl) + dist [i, j] (τl)

Accordingly, for the equivalent distance dist_d: dist_a [i, j] (τ2) = dist_plan_a [i, j] (τ2; τl) + dist_a [i, j] (τl)

In one embodiment, an increase in the distance traveled is assumed proportional to the elapsed time. Accordingly, dist_plan_a [i, j] (τ2; τl) is calculated according to the calculation rule dist_plan_a [i, j] (τ2; τl) = c * (τ2-τl). In the following, Tl [i, j] (τl) denote the actual duration of the i-th individual experiment until the time τl and T [i, j] the planned total duration of the i-th individual experiment (i = l, ... , n) for the jth combination (j = l, ..., m). The equivalent route, which is planned from the beginning to the completion of the ith A single test should be completed in order to test the final combination is called dist_plan_ä [i, j]. The proportionality factor c is calculated according to the following calculation rule: _{c =} dist_d [i, j] (rl) _^ falls dist_ä [i, jj (τ \) _> ΔT [i, j] (τ 1) 'dist_plan _ prefix and dist plan ä [i, j]. ,. , c = - ^~ else. The barrier Δ is now for example 0.1. The case distinction ensures a stable prognosis even in the case that only a few percent of the i-th individual experiment was completed.

With dist_plan [i, j] (τ2; τl), the actual travel distance in the ith attempt is called the jth combination. With dist_plan_ges_ä [j] (τ2; τl) is meant the equivalent travel distance which, according to the planning, is covered in the period from τ1 to τ2 in all individual tests for the jth combination (j = 1, ..., m). This total equivalent distance is calculated according to the calculation rule nn dist_plan_ges_a [j] (Y2; rl) = Σdist_plan_a [i, j] (r2; rl) = Σdist_plan [i, j] (r2; rl) * r [i, j] i =] i = 1 calculated.

The total equivalent distance dist_ges_ä [j] (τ2) that has been covered and that will be covered according to the planning until time τ2 is therefore dist_ges_ä [j] (τ2) = dist_ges_ä [j] (τl) + dist_plan_ges_ä [j] (τ2; tL)

The error behavior of a component is preferably described statistically by a parametric error occurrence model, for example by a Weibull distribution. The Weibull distribution was first presented in W. Weibull: "A Statistical Distribution of Wide Applicability," J. Appl. Mechanics, Vol. 18 (1951), pp. 293-297 is described in detail in RA Abernethy: The New Weibull Handbook, 4th ed., 2000.

It has a distribution function F (r) = F [λ, β] (τ) = 1-exp [(-λ * r) ^β ] which depends on the two parameters λ and β. Here, the parameter λ defines the error rate and the parameter β ("slope") the form of the distribution function The probability that an error of a certain type of error has occurred at a component up to the time τ is just F (τ) related to a year and a type of error is thus F (I year) The MTTF one

Weibull distribution is MTTF = -T (l + -). in this connection

F (x) denotes the value of the gamma function at the point x, n and for every natural number n: F (n + 1) = n! = TTi.

If β = 1, the Weibull distribution changes to the exponential distribution. This has the distribution function F (τ) = F [λ] (τ) = 1 - exp [- (λ * τ)] with constant time error rate λ> 0.

In one embodiment, a time constant error rate λ [j] is assumed for the jth combination and the exponential distribution is used as the error occurrence model. Preferably, a time-dependent estimated value MTTF '[j] (τ2; τl) for the parameter MTTF [J] is then calculated according to the calculation rule dist _ ges _ ä [j] {τ \) + dist _ plan _ ges _ ä [j] ( τ2; τl)

MTTF '[j] (τ2; τ \) = f__ges [j] (τl) + f_ges [j] (τ2; τl) ₊ \. As mentioned above, f_ges [j] (τl) is the total number of all the errors of the j-th component-error-type combination that have occurred in all trials up to the time τl. By f_ges [j] (τ2; τl), the total number of all errors of the j-th combination that will occur according to the prediction in the period from τl to τ2 is designated. The denominator can never be zero. For the time-constant error rate λ [j], an estimate λ '[j] (τ2; τl) which depends on the two times τ1 and τ2 is calculated, in accordance with the calculation rule

1 λ '[j] (τ2; τ \) =

MTTF '[j] (τ2; tL)

As stated above, the reliability is predicted using the distribution function F (t). As a measure of the reliability with respect to the jth component-error type combination, a failure probability U [j] = U [j] (τ2; dist_a, τl) is predicted according to the calculation rule

U [j] (τ2; dist_a, τl) = F [λ '[j] (τ2; τl)] (dist_a) = 1-exp [-λ' [j] (τ2; τl) * dist_a].

At time τ 1, it is therefore predicted that an error occurrence probability with respect to an annual travel distance from dist_a to time τ 2 will be achievable, which is U [j] (τ 2; dist_a, τ 1).

Preferably, this method is performed once for each of the m component-error type combinations. At the time τ 1, m predictions U [j] (τ 2, dist_a, τ 1) (j = 1,..., M) are thereby generated. These m predictions can be compared. As a result, it can be found out at the time τ1 which component-error-type combination at the time τ2 will have a low predicted reliability. Already at the time τl it is possible to change the test plan and to test this combination in more detail. It is also possible, contrary to the original planning, not to carry out individual tests if the required reliability will be demonstrated even without these individual tests. This saves the process attempts and thus time.

In a first alternative embodiment, for an estimation, a maximum error probability α, which lies between 0 and 1 and therefore between 0% and 100%, specified. Common values for α are 0.1 or 0.05 or 0.01, ie 10%, 5% or 1%. The confidence level 1-α denotes the statistical reliability with which the reliability is predicted.

If no errors of the type of error occur up to the time τ 1 and it is predicted that no errors will also occur by the time τ 2, then in the first alternative embodiment a

Exponential distribution is used as the error occurrence model. Preferably, furthermore, β = 1 is assumed. A lower bound MTTF '[j] (τ2; α, τl) for the MTTF value MTTF [J], which depends on the confidence level 1-α, is calculated according to the calculation precedent

MTTF '/ / (f 2- a τl) = ^diSt - ^geS - ^d ^ ^' K ^τl ) ^{+ dist} - ^plan - ^ges - ^d ^ ^' K ^{r2> rl} )

-ln (l-α) is calculated.

If errors have occurred up to the time τ1 f_ges [j] (τl) or if it is predicted that errors will occur in the period from τ1 to τ2 f_ges [j] (τ2; τl), a non-α dependent estimate MTTF ' [j] (τ2; τl) according to the calculation rule dist _ ges _ ä [j] (rl) + dist _ plan _ ges _ ä [j] (τ2; τl)

MTTF '[j] (τ2; τ \) = f_ges [j] (τl) + f_ges [j] (τ2; τ \). As in the above embodiment, an estimated value λ '[j] (τ2; τl) for the time constant error rate λ [j] is calculated therefrom. Using the estimate of the error rate, an estimate of the achievable reliability is determined. In a second alternative embodiment, a Weibull distribution is used instead of the exponential distribution. The shape parameter β is assumed to be known. The following calculation rules use the individual equivalent routes of the n individual tests. With dist_ä [i, j] (τ) becomes the equivalent distance for the jth Combined, which was completed in the i-th individual experiment until the time τ (i = l, ..., n; j = l, ..., m).

If no errors occur by the time τ 1, in the second alternative embodiment, a lower limit MTTF '[j] (α, τ 2, τ 1) for the MTTF value MTTF [J] is calculated according to the calculation rule

Σ [dist _ ä [i,

_±

MTTF '[j] (a, τ2; τl) = {^ -}' T (I ₊ -)

Calculated in (I-a) β A one-tailed confidence interval for a Weibull distribution is described in R.A. Abernethy, supra, Chapter E-2.

If errors have occurred up to the time τ 1 f_ges [j] (τ 1) and f_ges [j] (τ 2; τ 1) errors are predicted in the period from τ 1 to τ 2, an estimate α MTTF '[j] (τ 2 ; τl) for the MTTF value MTTF [J] according to the calculation rule

Σ [dist _ ä [i, /] (τl) + dist_plan _ ä [i, j] (τ2; τ \)] ^ß _λ

MTTF '[j] (τ2; τ \) = {^} ^ß * T (I + -) f _ges [j] (τ \) + f_ges [j] (τ2; τ \) ß'. An estimate of the Weibull distribution is also described in RA Abernethy, loc. Cit., Chapter E-2.

In the following calculations, a required reliability must be demonstrated for each of the m component-fault-type combinations. This reliability must be proven in the form of a required maximum error rate. The required maximum error rate in this example is 150,000 ppm per year for the combination "muffler housing visibly corroded." For example, with 1 million muffler housings per year, a maximum of 150,000 housings per year may be visibly corroded per year. The required reliability values as well as the reliability values predicted in FIGS. 2 to 9 are purely hypothetical values, which serve exclusively to illustrate the exemplary embodiment. Actual ppm setpoints and proven ppm actual values are orders of magnitude below these hypothetical values.

On the x-axis of the diagrams of FIG. 2 to FIG. 9, the second times τ2 are respectively entered. The respective first time τl is indicated by an arrow. The trial will begin in February 2005. The plan is to complete the trial in December 2006. On the y-axis of the graphs of Figs. 2, 4, 6 and 8, the total accumulated equivalent travel distance dist_ges_a [j] (τ2) is plotted in [km] indicating the total trucks in the trial until the respective second time τ2 have completed or will cover according to the plan. As stated above, this total equivalent travel distance is calculated according to the calculation law value dist_ges_a [j] (τ2) = dist_ges_a [j] (τl) + dist_plan_ges_a [j] (τ2; τl).

The respective predicted error rate U = U [j] (τ2; dist_a, τl) is entered on the y-axis of the diagrams of FIGS. 3, 5, 7 and 9.

In Fig. 2 and Fig. 3, a plan before the start of the experiment is illustrated. The first time τl is so z. The diagram of FIG. 2 shows the respectively planned equivalent travel distance dist_ges_a [j] (τ2) up to the respective instant τ2. The diagram of Fig. 3 shows the time course of the predicted error rate U = U [j] (τ2; dist_a, τl) for the jth combination at the respective time. As can be seen from these hypothetical values, at the end of the experiment only an error rate of 330,000 ppm per year is predicted instead of the required error rate of 150,000 ppm per year. Therefore, the experimental program is changed so that the equivalent distance will increase. This is achieved, for example, by planning to increase the actual travel distance during the experiment.

FIGS. 4 and 5 illustrate the changed planning and a result of this changed planning. The first time τl is now after the beginning of the experiments, z. For example, in August 2005. Up to this time, no errors were measured in the trial. 4 shows the time course of the accumulated equivalent travel route. Until the first time in August 2005, this is the actual equivalent distance (product of the actual distance to the first time and the factor of refraction), then the planned equivalent distance. For comparison, the originally planned time profile of the accumulated equivalent travel distance, which has already been shown in FIG. 2, is shown in dashed lines in FIG. 4. 5 shows the time profile of the predicted error rate. For this purpose, the measurement result is used that no error of the type of fault has occurred up to the first time. For comparison, the time curve of the error rate predicted in accordance with the original planning, which was already shown in FIG. 3, is shown in dashed lines in FIG.

FIGS. 6 and 7 show an alternative continuation of the experiment. Also in this example, the first time after the beginning of the experiments, z. For example, in August 2005. Up to this time, no error type errors were measured in the trial. However, a smaller overall journey was actually covered by the first time than originally planned. This also reduces the equivalent distance. FIG. 6 shows the time profile of the equivalent travel route according to the alternative course. Until the first time in August 2005, this is the actual equivalent distance (product of the actual distance to the first time and the factor of refraction), then the planned equivalent distance. For comparison 4, the originally planned time course of the equivalent travel route, which has already been shown in FIG. 2, is shown in dashed lines in FIG. Fig. 7 shows the time course of the predicted error rate. For this purpose, the measurement result is used that no error of the type of fault has occurred up to the first time. For comparison, the time curve of the error rate predicted in accordance with the original planning, which has already been shown in FIG. 3, is shown in dashed lines in FIG.

8 shows the time course of the equivalent distance according to the alternative planning of FIG. 6. In the example of FIG. 8, the first time is at the end of November 2005. Until the first time in November 2005, the equivalent distance is shown in FIG the actual equivalent distance (product of the actual distance to the first time and the factor of refraction), then the planned equivalent distance. For comparison, in Fig. 8, the equivalent route of Fig. 4 was shown in dashed lines.

FIG. 9 shows the detectable error rate predicted in the experiment of FIG. 8. In the example of FIG. 9, two errors were counted before the first time in November 2005. As a result, the predicted error rate deteriorates significantly. For comparison, the course of the detectable error rate of Fig. 5 is shown in dashed lines in Fig. 9.

For October 2005, before the two errors, an error rate of 500,000 ppm / year was predicted. For November 2005, after the occurrence of the two errors, an increased error rate of approximately 660,000 ppm / year was predicted. The predicted error rate also lies clearly above that of FIG. 5 in the following months. is the reference numerals and symbols used

Claims

claims

1. A method for predicting the reliability of a technical product, wherein

when the reliability is predicted an average number of faults related to a load of the product during its use, the reliability during use of the product is calculated using a parametric fault occurrence model, the fault occurrence model counting the number of faults as a function of a measure of the load during describes the use in which the product is subjected to stress in at least one test and the test extends over a predetermined period of time and a quotient (r) of the number of errors related to the load during the test and the number of errors related to the load during use is determined, characterized in that a first and a second time of the time period are predetermined, wherein the second time is after the first time, is counted how many errors in the experiment occurred up to the first time, as an actual load a measure is measured for the load to which the product is subjected during the test until the first time, as a target load is a measure of what the load will be exposed to the product during the test between the first time and the second time, an expected number of errors between the first and the second time is predefined or predicted in dependence on the desired load, and at least one parameter of the error occurrence model is automatically calculated using the actual load, the counted number of errors up to the first time, the desired value -Belast, the expected number of errors between the first and de m second time and the quotient is predicted.

2. The method according to claim 1,

characterized in that the number of faults depends on a degradation that occurs on the product due to the load during the test, depending on the target load predicted an expected degradation of the product between the first and the second time and the expected number of errors between the first and the second time depending on the expected degradation is predicted.

3. The method according to claim 1,

characterized in that it is specified as the expected number of errors that no errors occur between the first and the second time.

4. The method according to any one of claims 1 to 3,

characterized in that as reliability the number of errors related to a time-dependent first quantity characterizing the load of the product during its use is calculated as the actual load of the product during the test a second quantity representing the load of the product during the test Test characterizes, is measured and as the target load, a value for the second size is given.

5. The method according to claim 4,

characterized in that the first size is the duration of the load on the product during its use and the second size is the duration of the load on the product during the test.

6. The method according to claim 4,

characterized in that both the loading of the product during use and the loading of the product during the trial include the occurrence of distressing events, the first magnitude being the number of stressful events during use and the second magnitude being the number of adverse events during the exercise Trial is.

7. The method according to any one of claims 1 to 6,

characterized in that the product is a vehicle or a component of a vehicle, the first quantity is the distance traveled in the use of the vehicle and the second quantity is the distance traveled by the vehicle during the test.

8. The method according to any one of claims 1 to I _1st

characterized in that in a first preliminary test the product is exposed to stress during the experiment, in the first pretrial test, measuring a first parameter characterizing the load of the product during the test, in a second pretrial test subjecting the product to the load during use, in the second pretrial test, a second parameter representing the load of the product during use is measured, both in the first and in the second preliminary test, a characteristic which characterizes the degradation caused by the respective load on the component is measured, and the quotient is determined using the measured values of the first and second second characteristic of the load characteristic, the measured value of the degradation characterizing characteristic is determined.

9. The method according to any one of claims 1 to 8,

CHARACTERIZED IN THAT a reliability set point is predicted, a number of errors to be expected after the first time is predicted and using the actual load, the counted number of errors up to the first time, the target load, the predicted number of errors after the first time and the quotient is predicted, at which second time a value is predicted for the reliability which is greater than or equal to the predetermined desired value.

10. The method according to any one of claims 1 to 9,

characterized in that the method steps are formulated as program code and the program code is part of a computer program that runs on a data processing system.

11. Computer program product that can be loaded into the internal memory of a computer and

Software sections with which a method according to any one of claims 1 to 9 can be performed when the product is running on a computer.

A computer program product stored on a computer readable medium and having computer readable program means for causing the computer to carry out a method according to any one of claims 1 to 9.

13. Digital storage medium with electronically readable control signals, which can cooperate with a programmable data processing system so that a method according to one of claims 1 to 9 can be executed.

14. Data processing system for the prediction of

Reliability of a technical product, whereby

the data processing system is configured for predicting an average number of errors related to a load of the product during its use as the reliability, the data processing system has read access to a data memory in which a computer analyzable protocol of at least one test in which the product has been subjected to a load and the extending over a predetermined period of time, and storing a parametric error-occurrence model describing the number of errors as a function of a measure of the load during use, and the data processing system for automatically applying a quotient of the number of errors related to the load during of the test and the number of errors related to the load during use, characterized in that the log shows the number of errors that have occurred in the test up to the first time and unchecked d as an actual load, a measure of the load that the product is exposed to during the test until the first time point comprises, in the data memory, a computer-accessible description of a desired load as a measure of what load the product has undergone during the test between the test stored for the first time and a subsequent second time of the period is stored, the data processing system is designed to predict an expected number of errors between the first and second time depending on the target load and the data processing system for predicting the reliability of Use of the actual load, the counted number of errors to the first time, the target load, the expected number of errors between the first and the second time and the quotient designed.

15. Computer program product for predicting the

Reliability of a technical product, whereby

the computer program product is designed to predict an average number of errors related to a load of the product during its use as the reliability, the computer program product has read access to a data store storing a computer evaluable log of at least one trial in which the product has been subjected to a load and which extends over a predetermined period of time, and the computer program product for automatically applying a quotient from the Number of errors related to the load during the test and the number of errors related to the load during use, characterized in that the log records the number of errors that occurred in the test until the first time as an actual Stress, a measure of the load experienced by the product during the trial until the first time, includes in the data memory a computer-accessible description of a target load as a measure of what strain the product has undergone during the trial between the first time and one subsequent second Time of the period will be exposed, and the computer program product for predicting an expected number of errors between the first and the second time depending on the target load is configured and the computer program product for predicting the reliability using the actual load, the counted number of errors up to the first time, the target load,

- The expected number of errors between the first and the second time and the quotient is configured.