EP3959571A1

EP3959571A1 - Method for determining remaining useful life cycles, remaining useful life cycle determination circuit, and remaining useful life cycle determination apparatus

Info

Publication number: EP3959571A1
Application number: EP20712294.6A
Authority: EP
Inventors: Jonathan MANRIQUE GARAY; Carsten Trimborn
Original assignee: Volkswagen AG
Current assignee: Volkswagen AG
Priority date: 2019-04-23
Filing date: 2020-03-16
Publication date: 2022-03-02
Also published as: CN114008549A

Abstract

1. The invention relates to a method (1; 10; 40) for determining remaining useful life cycles of a production apparatus (20; 73) based on wear, comprising the steps of: determining (2; 11; 41) a first state value of a maintenance variable of a part of the production apparatus (20; 73) based on sensor data of a sensor for the production apparatus; and determining (3; 16; 49) a future use cycle of the part of the production apparatus (20; 73) for which the maintenance variable has a previously specified second state value, wherein the future use cycle is determined based on the first state value and based on a discrete stochastic degradation model.

Description

description

Procedure for determining remaining usage cycles,

Remaining usage cycle determining circuit, remaining usage cycle determining device

The invention relates to a method for determining remaining usage cycles, a

Remaining usage cycle determining circuit, and a

Remaining usage cycle determining device.

In general, methods are known for determining a Remaining Useful Lifetime (RUL) of a device.

Such methods are typically based on stochastic methods or models and can be used to predict a threshold value at which a

Device fails due to wear and tear, such as in the context of

Turbo machinery (e.g. in the aerospace industry) and / or in the context of

Weapons (systems).

In contrast, methods for determining a RUL for a production device are often based on online adaptation, a value that is used for the determination being continuously updated in such methods, so that the determination is continuously updated.

For example, US Pat. No. 8,725,456 B1 discloses a prognostic tool for determining a RUL of a component or a sub-system on the basis of two different regression models. The RUL is determined here based on an artificial intelligence training.

Furthermore, on the laid-open specification CN 107480440 A, a

Remaining service life prediction method known, which is based on a random

Degradation modeling determines a remaining service life.

In addition, the laid-open specification CN 107194478 A describes a method for predicting a remaining service life in which a drift parameter and a diffusion parameter are used to describe a degradation process. With such methods, however, no future usage cycle is determined, but a duration.

The object of the present invention is to provide a method for determining

Remaining use cycles, a remaining use cycle determining circuit, and a

To provide residual usage cycle determining device which at least partially overcome the above-mentioned disadvantages.

This object is achieved by the method according to the invention for determining

Residual usage cycles according to claim 1, by the inventive

Remaining cycle of use determination according to claim 14, and by the invention

Remaining use cycle determining device according to claim 15 solved.

According to a first aspect, a method according to the invention for determining remaining usage cycles of a production device due to wear includes:

Determining a first state value of a maintenance variable of a part of the

Production device based on sensor data from a sensor for the

Production device; and

Determination of a future usage cycle of the part of the production device for which the maintenance variable has a predetermined second status value, the future usage cycle being determined based on the first status value and based on a discrete stochastic degradation model.

In a second aspect, one is according to the invention

Remaining usage cycle determination circuit set up to carry out a method according to the invention according to the first aspect.

According to a third aspect, includes an inventive

Remaining usage cycle determining device comprises a remaining usage determining circuit according to the second aspect.

Further advantageous embodiments of the invention emerge from the subclaims and the following description of preferred exemplary embodiments of the present invention.

As already discussed, known methods determine a remaining useful life (RUL). However, it was recognized that such procedures do not involve breaks in production (e.g. weekends) (Can) take into account that necessary maintenance may be carried out too early, which can be cost-inefficient.

It was therefore recognized that it can be advantageous and simpler if residual usage cycles are determined, since in this case no production break has to be taken into account. For example, maintenance can be planned on the basis of parts that have already been produced, which reduces unwanted uncertainty due to a long break (e.g. due to man-made schedules).

Furthermore, it was recognized that it can be desirable if production costs can be reduced and product quality can be increased or at least kept constant at the same time.

This can be achieved through the early detection or anticipation of a possible malfunction, a possible malfunction or other malfunctions in a production line.

With known methods, such an early detection can take place imprecisely.

For example, in model-based methods that can describe a physical representation of the behavior of a device, a prediction can be imprecise because simplifications have to be made, which can result in the complexity of the device being lost, and interactions of the device with the environment not

necessarily be considered.

Furthermore, with data-driven models (e.g. artificial intelligence, machine learning), a high data quality and a certain number of failures are required so that such models can learn a prediction, which can be cost-inefficient

Furthermore, labeled data may have to be used to create a

to train a data-driven model, whereby such labeled data is not necessarily available, for example because a necessary infrastructure does not exist or because the costs for such an infrastructure can be disproportionately high.

Furthermore, such methods are not necessarily transferable between two

Devices, even if the devices are similar or if they vary slightly with regard to a task or in their construction. For this reason, machine learning would have to be carried out again for a transfer, whereby, as mentioned above, the amount of data for training may not be sufficient. In addition, such methods cannot be applied to new devices.

In this respect, it was recognized that a combination of statistical models and stochastic processes to create a probability density function for risk analysis and for deciding whether and when maintenance has to be carried out can overcome such disadvantages.

Furthermore, it was recognized that it can be desirable if a remaining useful life does not have to be carried out at the beginning, shortly after commissioning of a production device, or shortly after maintenance, since it can cause unnecessary costs and necessary (labeled) data are not necessarily available as discussed herein.

It was also recognized that determining the remaining life among such

Conditions can lead to an inaccurate forecast. A Monte Carlo simulation based on a linear Wiener process model can be carried out, which can lead to large variations or inaccuracies if there are small variations in a drift of the Wiener process, so that initial values are falsified by condition monitoring For example, a positive value can become negative or close to zero. In addition, a determined confidence interval in a forecast can be too large, so that a Monte Carlo simulation does not converge to a defined failure threshold.

Therefore it was recognized that a drift control can be implemented in order to exclude a drift around a zero value or a negative drift, so that a Monte Carlo simulation converges. In this way, computing power can advantageously also be reduced.

In some exemplary embodiments, however, a negative drift can also indicate a statement about an improvement in a state of a production device.

In other words, it was recognized that a determination of a remaining service life began

Commissioning of a production device does not necessarily have to be carried out and can be confusing if a need for maintenance occurs too early, since this increases the susceptibility to errors in determining the remaining service life. Therefore, some embodiments relate to a method for determining

Residual usage cycles of a production device due to wear and tear, comprising:

Determining a first state value of a maintenance variable of a part of the

Production device based on sensor sensor data of a sensor for the

Production device; and

The device can comprise any device that must or can be serviced due to wear, such as a production device in FIG

Automotive industry and the like.

In this context, a cycle can comprise a predetermined operating duration, a predetermined operating time, a predetermined period, and the like, in which the device executes a predetermined or device-typical process or method.

In each operating cycle, part or all of the production device

Production device are subject to wear, for example by mechanical

Stress, through erosion or other circumstances which lead to the part or the entire production device having to or can be serviced.

In general, it is desirable to know a point in time when maintenance is to be carried out in order, for example, to schedule it (ie to set a date for maintenance), since, for example, if maintenance is carried out too late, there will be an operational downtime and thus higher Costs, while maintenance performed too early can lead to increased costs due to excessive maintenance frequency.

In some exemplary embodiments, a first status value of a maintenance variable of a part of the production device is determined.

The maintenance variable can be a physical, mechanical and / or electronic variable, and the like, which can indicate wear. For example, the maintenance quantity can include an elongation of the part, a temperature, a thickness, a pressure, a humidity, a filling amount, and the like.

The first status value can comprise a measured value of the maintenance variable, for example ten bar if the maintenance variable comprises a pressure.

The sensor can accordingly be set up to generate sensor data which are indicative of the state value.

Therefore, the sensor can, for example, be a pressure sensor, a distance sensor, a

Temperature sensor, a humidity sensor, a color sensor, and the like.

The sensor data can also be indicative for several (e.g.

consecutive) status values, which flow into the determination of the future usage cycle individually, in groups or as a (weighted) average value.

The sensor data can originate, for example, from condition monitoring (CM), which is typically provided in known production devices. So must

advantageously no additional sensors are provided, which reduces costs. CM data may include or be indicative of statistical parameters of each work cycle of the production device and stored in the production device and retrieved after (or during) each production cycle or after each part produced by the production device.

Furthermore, a buffer can thereby advantageously be defined, which should be present so that if the production device fails, enough parts are present so that a production line does not have to be shut down.

In addition, such a buffer can advantageously be dispensed with in some exemplary embodiments, since as a result no space or storage capacity (which is typically limited) has to be released and no financial means have to be expended for the buffer.

In other words: In some exemplary embodiments, a cycle can be for a produced component (or product) (or for a specific number of produced components or

Products), so that cycles of a production device advantageously with the Component buffers can be compared so that a method according to the invention makes a prognosis about a remaining useful life more accurate.

The future usage cycle can comprise a usage cycle which is determined on the basis of a current and / or past usage cycle or from the first status value in the current and / or past usage cycle, for which the

Maintenance variable assumes a predetermined second state value.

For example, the second status value (for example five bar) can be a threshold value at which maintenance must be carried out or at which maintenance must be scheduled. The second status value can also be a value from which another method can be used or from which subsequent values can be known.

The second state value can be determined based on the first state value and on a discrete stochastic degradation model. For example, the first

State value can be used as the initial value for the discrete stochastic degradation model.

The discrete stochastic degradation model can generally include a model, an algorithm, and the like, which, based on a stochastic and / or statistical analysis and based on the first state value, allows a prediction of the maintenance variable in future usage cycles.

The stochastic degradation model is discrete in that a variable des

The degradation model is discrete. In particular, in this invention, the usage cycle is a discrete variable as opposed to a continuous variable such as time.

A problem with the continuous variable time can, for example, be that in such a model it cannot typically be taken into account that the

Production device or the part of the production device (unexpected or expected) can be switched off. In other words, in such a model it is typically assumed that each usage cycle is evenly distributed in the time dimension. Furthermore, it can be problematic that it can be difficult to compare remaining parts in the buffer (as described above) with the remaining time, since there is not necessarily a correspondence.

However, in an industrial production device, it may

for example on public holidays, on weekends, during company holidays, and the like

is switched off, so that such an even distribution is not given. In the case of continuous stochastic models, this can lead to an incorrect prediction of a future time at which a maintenance variable assumes a fixed value.

The present invention thus has the advantage that not one of these

Uniform distribution must be assumed, which is initially advantageous

Simplified degradation model and, on the other hand, advantageously a more precise one

Allows prediction of the previously determined second state value.

In some exemplary embodiments, the method further comprises: determining a number of remaining usage cycles of the production device based on the future one

Usage cycle.

The remaining usage cycles can result from or correspond to the future usage cycle, for example the remaining usage cycles can be the number of

Use cycles include until the future use cycle is reached, or another use cycle, which is determined on the basis of the future use cycle, is reached, and the like.

In this way, a correlation between the parts of the buffer (as described above) with the remaining usage cycles can advantageously be established, so that maintenance can advantageously be better planned.

In some exemplary embodiments, the method further comprises: generating a

Probability density function for determining the remaining usage cycles.

As is well known, the probability density function can be indicative of a probability. Typically, a probability can be specified as a function of a variable, for example as a function of time. However, as discussed herein, using the continuous variable of time can be disadvantageous, so that, in some

Embodiments, the variable is a usage cycle, so that the future

Use cycle can advantageously be determined.

Therefore, in some exemplary embodiments, the probability density function is indicative of a probability for which future usage cycle the maintenance variable assumes the second state value.

In some exemplary embodiments, the degradation model includes a Wiener process model.

For example, degradation (or wear and tear), for example in an electrochemical device, can occur in a random manner.

A stochastic model, such as a Wiener process, can therefore be used to describe the degradation.

Here, in some exemplary embodiments, it is taken into account that the degradation can be described partly deterministically and partly stochastically, with the deterministic part being able to be the same for all tested devices (that is to say for one

Total population), and the stochastic part can represent an uncertainty which is generated by a difference in the devices within the total population.

Therefore, in some exemplary embodiments, a Wiener process is considered which does not describe a pure random walk, but a random walk with a drift. As is well known, such a Wiener process can be described with the following formula (1):

In formula (1), X (t) corresponds to the degradation, x _{i to} the first state value, l (t) to a drift coefficient (thus describes the deterministic part), s to a dispersion coefficient, and B (t) to a Brownian movement (thus describes the stochastic part). In this description of the Wiener process, the continuous variable becomes time

described. In order to find a description of the degradation X (t), a set of known values (i.e. first state values based on a sensor measurement, which is indicative of condition monitoring (CM)) X _{1: i} = {x1, x2,. .., x _i } should be considered. Based on the quantity Xi ,, a remaining service life (RUL (Remaining Useful Lifetime)) can be defined, as shown in formula (2):

In formula (2), T describes the remaining service life, inf stands for infimum (as is generally known), and w stands for a failure threshold (FT) to which the

Device is typically no longer usable.

This can result in a probability density function of the remaining service life:

In formula (3) stands for the probability density function, p for the circle number, and exp for an exponential function to the base e (Euler number), without restricting the present invention thereto, since every possible exponential function can be used.

Now, such a calculation of the remaining life can be disadvantageous and inaccurate as described herein, and therefore the present invention can determine remaining usage cycles.

Instead of X (t), X (c) can be used, where c stands for a (future) cycle of use and in formula (1), each t can be replaced with a c without loss of generality, so that the formula is not repeated here .

Furthermore, a remaining usage cycle C can be defined as in formula (4):

The remaining usage cycle, as described herein, can be indicative of the future

Usage cycle, or an indication of how many usage cycles with a Production device are still possible before the device fails or is no longer usable.

In such exemplary embodiments, the future usage cycle can be or correspond to a predetermined number of usage cycles before the remaining usage cycle.

This can result in a probability density function, which can be seen in formula (5): cQ, which can also be briefly defined as n, describes the number of past cycles since a previous determination of residual usage cycles or since the determination of the first status value.

After each cycle, a method according to the invention can be applied again, so that advantageously a more precise prediction of the remaining usage cycles can take place as long as a wear threshold value (healthy threshold) has already been exceeded.

In some exemplary embodiments, the Wiener process model is based on a drift, as described herein.

As is well known, a Wiener process model represents a Markovian property whereby the Wiener process model is memoryless, i.e. if the second state value were determined exclusively based on the Wiener process model, a prediction would always be based on the first state value x 1, but no historical first state values (from previous measurements) would be taken into account.

Therefore, in some exemplary embodiments, the degradation model comprises a Bayesian forecasting model.

In some exemplary embodiments, the method further comprises: determining a drift in the Wener process model based on the Bayesian prediction model. It is generally known that the drift coefficient l (t) is subject to a process that changes over time. Therefore, the first state value can be the only known value, which is why the drift must be determined.

As described above, the drift is generally considered to be deterministic. However, in some exemplary embodiments it can be modeled as a random variable (or stochastic variable).

In some exemplary embodiments, the drift is determined based on a Bayesian filter and / or a Kalman filter.

In such a way, as is well known, the drift can in a

State space model can be constructed as shown in formulas (6) and (7):

h can be proportional to a normally distributed noise e can be proportional to a further normally distributed noise be as a variance to represent a Brownian motion. In addition, formula (6) represents a system equation and formula (7) represents an observation equation. H and e may be uncorrelated. However, the present invention is not limited to this.

If h and e are uncorrelated, formula (7) can be adapted according to the invention, from which equation (8) results:

Here e ~ N (0, n), i.e. proportional to a normal distribution around zero, is the first

State value with n as the variance, where n can originate from the natural numbers and comprises the past cycles since a method according to the invention was previously carried out. c _i corresponds to a current cycle or a cycle of a last measurement by a sensor and C _in corresponds to a cycle of the last execution of a

method according to the invention.

Typically, in known methods, such as, for example, in equation (7), the variance of e can increase over time, although no usage cycle has taken place, since no usage breaks and the like are taken into account.

In a method according to the invention which takes equation (8) into account, however, the variance advantageously only increases with each usage cycle, so that there is the advantage that an unnecessary addition of uncertainty in the determination of the second

State value is avoided.

In some exemplary embodiments, a strand tracking filter algorithm is used to detect sudden signal changes (or sudden (short-term) changes in the first

State value) and to determine the drift.

The following first algorithm can be implemented for this:

iv) State estimation based on the Bayesian predictive model for a single variable:

Update the drift estimate and the variance:

In some exemplary embodiments, the degradation model includes a Monte Carlo simulation.

In this case, with a sufficiently high number N of simulations based on the first algorithm, an accuracy of the determination of the second can advantageously be achieved

State value can be increased.

This allows a number of remaining usage cycles, for example on the basis of the following second

Algorithm to be determined.

Input: first status value Estimated Drift Failure Threshold FT

Initialization: S = 0

Simulate N degradation paths as follows:

v) end while

vi) end for

vii) Generate a normal distribution based on the set S _f

Here i <j <f. In addition, c _{i is} a usage cycle on which a past (or the last) sensor measurement was carried out and c _f is a future usage cycle on which the FT is achieved.

In some exemplary embodiments, the method further comprises: processing the first status value. The first state value can, for example, be noisy, an outlier from a statistic, and the like.

The first status value can therefore be processed, i.e. processed to filter and / or normalize the first state value.

Therefore, in some exemplary embodiments, the processing includes filtering.

For example, a Kalman filter can be used to filter a noise.

In general, noise can make a prediction or the determination of the second status value and / or the future usage cycle incorrect. This can be caused, for example, by an indirect measurement.

To remove noise from the first state value, a Kalman filter can be implemented in the form of the following third algorithm:

Input: first status value x _;

Output: Filtered first status value

Update of the signal estimate and the variance:

Noise can thus advantageously be removed and the determination of the second state value can be carried out based on a more stable first state value.

In some exemplary embodiments, the processing includes a rolling window regression.

The rolling window regression can be applied after the Kalman filter in order to filter a fluctuation of the first state value on a short time scale. For this purpose, a mean value is formed for several (consecutive) first status values. For example, a first mean value is formed for the first through the nth first state value (in a first window). The window is then shifted by one position so that a second mean value is formed for the second to n + 1-th first state values. Then the window can be shifted one more place so that for the third to n + 2th

State value a third mean value is formed.

As a result, the stability of the first state value and thus the determination of the future state value can advantageously be increased.

In some exemplary embodiments, the processing is therefore also based on one

moving average as described herein.

In some exemplary embodiments, the method further comprises: determining at least one threshold value based on the first status value, which is indicative of the second status value.

The first threshold value can include, for example, a healthy threshold, a failure threshold, and the like, as described herein.

As a result, a (predetermined) standard can advantageously be met and a

Reliability of the device can be guaranteed.

To be able to meet a standard and ensure reliability

typically several tests are performed before a device is released, for example for use in production. In such tests, the device can be tested in such a way that its behavior is checked under extreme conditions. Based on such tests, standards can typically be generated which the

Define extreme conditions as thresholds for specific device characteristics.

By means of a specification which is set, for example, by a standard, it is possible, for example, to know in advance a maximum permitted operating condition of a device or a part of a device.

For example, in the case of a gluing machine, a pressure can be an indirect measurement of an amount of current flowing through a motor. Furthermore, an FT can be based on a standardized The (maximum) value of such a current flow rate can be given, which is expressed by pressure values in a chamber of the motor.

Operating or operating the engine above such a maximum value can be dangerous for the intactness of the engine. Therefore, if the engine is operated above the maximum value, the engine can be downtime in order to avoid further damage.

In such a context, the FT can represent a value at which the device can no longer be operated.

In addition, based on a test or a check and / or based on a standard (or several standards), a threshold value can be established up to which the device can still be operated without damage having typically already occurred. Such a threshold value can include security until the FT is reached and can be referred to as HT (Healthy Threshold). Furthermore, a BU (Business as Usual) threshold value can be defined, which can designate a state in which the device is operated.

The HT can be defined based on the BU and using a standard deviation s from first state values (or raw data) during a BU operating mode. Without limiting the present invention to this, the HT can be defined as follows:

The BU can be determined on the basis of filtered first state values (which have been filtered with a Kalman filter, as described herein), wherein advantageously a probability that the BU assumes a value above the HT is low as a result of the filtering as long as the BU is below of the HT lies. In addition, every time the threshold value (HT) is exceeded, a random (or stochastic) degradation or wear takes place.

If the filtered first state value exceeds the HT, in some

Embodiments, no more residual usage cycles are determined, so that

advantageously, the HT can be recognized before damage to a device occurs. Some exemplary embodiments relate to a remaining usage cycle determining circuit which is set up to carry out a method according to the invention.

The remaining usage cycle determination circuit can comprise, for example, a CPU (Central Processing Unit), GPU (Graphic Processing Unit) or any other type of at least one processor, an FPGA (Field Programmable Gate Array), and the like. The circuit can receive sensor data in that it is connected to at least one sensor or comprises at least one sensor. Furthermore, it can be connected to at least one sensor and at the same time comprise at least one (other) sensor.

The at least one sensor can be set up to measure at least one status value of at least one maintenance variable or to generate measurement data that are indicative of the at least one status value, for example by means of a direct or indirect measurement.

Furthermore, the sensor can also be set up to measure several maintenance variables, such as a pressure and a humidity, a distance, a volume, and the like.

The remaining usage cycle determination circuit can also have one (or more)

Computers, servers, and the like, which are connected so that a

inventive method can be carried out.

Some exemplary embodiments relate to a remaining usage cycle determining device which comprises a remaining usage cycle determining circuit according to the invention.

The device for determining the remaining usage cycle can, for example, comprise a computer, server, and the like, and at least one sensor or be connected to it. In addition, the device for determining the remaining usage cycle can also comprise a device for which a number of remaining usage cycles is to be determined.

Embodiments of the invention will now be described by way of example and with reference to the accompanying drawings, in which: Fig. 1 schematically shows an embodiment of a method according to the invention for

Determining remaining usage cycles of a production device due to wear in a block diagram;

2 shows a further exemplary embodiment of a method according to the invention for determining remaining usage cycles of a production device due to wear in a block diagram;

3 shows a production device for which a number of remaining usage cycles can be determined;

4 shows a further exemplary embodiment of a method according to the invention for determining remaining usage cycles of a production device due to wear in a flow chart;

5 shows a graph for determining remaining usage cycles;

Fig. 6 is a graph showing degradation; and

7 shows a device for determining the remaining usage cycle according to the invention with a circuit for determining the remaining usage cycle according to the invention.

An embodiment of a method 1 according to the invention for determining

The remaining usage cycles of a production device due to wear and tear is shown in a block diagram in FIG. 1.

In FIG. 2, a first state value of a maintenance variable of a part of a production device is determined based on sensor data of a sensor for the production device, as described herein.

In FIG. 3, a future usage cycle of the part of the production device for which the maintenance variable has a predetermined second state value is determined, the future usage cycle being determined based on the first state value and based on a discrete stochastic degradation model, as described herein. FIG. 2 shows a further exemplary embodiment of a method 10 according to the invention for determining remaining usage cycles of a production device due to wear.

In FIG. 11, a first state value of a maintenance quantity of part of a

Production device based on sensor data from a sensor for the

Production device determined as described herein.

In FIG. 12, the first status value is processed as described herein.

In FIG. 13, at least one threshold value is determined based on the first state value, which is indicative of the second state value, as described herein.

In FIG. 14, a drift in a Wiener process model is determined based on a Bayesian prediction model as described herein.

In FIG. 15, a probability density function for determining the remaining usage cycles is generated as described herein.

16, a future usage cycle of the part of the production device for which the maintenance variable has a predetermined second status value is determined, the future usage cycle being determined based on the first status value and based on a discrete stochastic degradation model, as described herein.

In FIG. 17, a number of remaining usage cycles of the production device is determined based on the future usage cycle, as described herein.

Fig. 3 shows a production device for which a number of remaining usage cycles can be determined.

In this exemplary embodiment, the production device is a gluing machine 20 which has a motor 21 with a motor axis 22, a spindle 23, a nut 24 with rolling elements, guide rods 25, a piston 26, an adhesive chamber 27, an inlet valve 28, an outlet valve 29, and a Has nozzle 30.

With a high pressure pump and distributor, the gluing machine 20 can distribute glue. The manifold, as it operates under high pressure, can be prone to wear. About that In addition, the spindle 23 can be difficult to maintain, which can be expensive. A method according to the invention is therefore used for the spindle 23. A sensor system that monitors the gluing machine 20 can measure a pressure in the gluing chamber 27 in addition to eight further parameters. In this exemplary embodiment, the pressure is the relevant maintenance variable.

The (maximum) pressure (during a production cycle) is the main maintenance variable in this exemplary embodiment, according to which a possible failure of the spindle 23 can be predicted or estimated. The nozzle may have a momentary blockage while the adhesive is being applied. As a result, it may be necessary to increase the pressure in the glue chamber 27 in order to free the nozzle. In order to increase the pressure, as described above, a current of the motor 21 can be increased. The current, and thus the pressure, can, however, be subject to fluctuations, as a result of which a data point which reflects the maximum pressure which after a production cycle (measured indirectly) can be an outlier and / or noisy.

Therefore, as described herein, the pressure is filtered with a Kalman filter and a rolling window regression with a moving average.

4 shows a further exemplary embodiment of a method 40 according to the invention in a flowchart.

Condition monitoring is carried out in 41, whereby a first status value of a maintenance variable is obtained, i.e. a pressure data set is determined by a pressure sensor.

In 42, the obtained print dataset is filtered as described herein.

At 43 it is decided whether there has been wear, i.e. whether a maximum pressure value has been exceeded.

If the maximum pressure value has not been exceeded in 44, a new first is made

State value determined.

If the maximum pressure value has been exceeded, a drift estimate for a Wiener process based on a Bayesian network is carried out in 45, as described herein. The drift estimation is applied in such a way that a negative value or a zero value for the drift is avoided in order to advantageously avoid a divergence of a Monte Carlo simulation, which advantageously makes it possible to save computing power.

In 46, based on this, that is to say based on a Wiener process with drift using a Monte Carlo simulation, a future usage cycle is determined. This means that it is determined how often the pressure is likely to exceed the maximum pressure value in future usage cycles. The future usage cycle then indicates a failure threshold.

A number of remaining usage cycles is determined based on this. If in 47 the number of remaining usage cycles is less than the usage cycles per day (i.e. the gluing machine can then still be used on the production day), a new print data record is determined in 48. If the number of remaining usage cycles is greater, a report is generated in 49 and maintenance is initiated.

In this exemplary embodiment, it is assumed that there are sufficient first status values or that the sensor data set is large enough so that a

statistical / stochastic analysis can be performed, but the present invention is not limited thereto. For example, only a first status value can be present.

In such a method according to the invention, maintenance can be carried out when a production break occurs or when one is planned. For example, the production break can be planned based on the remaining usage cycles determined so that

advantageously costs can be saved, which by an unscheduled

Production break can arise.

FIG. 5 shows a graph 50 for determining remaining usage cycles, which shows on an ordinate 51 a measured pressure in an adhesive chamber (as described above), which is plotted as data points 52 against a time 53. Between the data points 52 there are empty spaces 54 which are created by production breaks (for example weekends). A healthy threshold 55 is also shown. The data points which are above the HT 55 correspond to second state values, have thus been determined using a method according to the invention. The data points above the HT 55 provide information about a number of remaining usage cycles up to a failure threshold 56. FIG. 6 shows a graph 60 which shows a degradation, with a multiplicity of probability density functions 61. The graph 60 has an ordinate 62

Probability (or probability density, English: likelihood) plotted and 63 future usage cycles on an abscissa. As can be seen, they are

Probability density functions 61 normally distributed. The mean values 64 of the

Probability density functions 61 indicate an expected number of remaining usage cycles, while crosses 65 represent an actual number of remaining usage cycles before a failure.

The advantage here is evident that when using an inventive

Procedure, an accurate prediction can be made. For example, here eighty percent of the determined remaining usage cycles remain below a predefined confidence interval, so that an accuracy is advantageously high, so that a

according to the invention can be used in production devices, and an uncertainty due to the continuous variable time can advantageously be disregarded.

7 shows a remaining usage cycle determining device 70 according to the invention for determining a number of remaining usage cycles of a production device 73

Remaining usage cycle determining device 70 comprises a sensor interface 71 which receives sensor data from a sensor 72. The sensor 72 is set up to generate sensor data which are indicative of a first state value (or a plurality of first state values) of a maintenance variable, as described herein.

The sensor interface 71 is set up to determine the first status value from the sensor data.

The remaining-use cycle determining device 70 further includes a

remaining use cycle determination circuit 74 according to the invention, which in this

Embodiment is designed as a CPU, and which is set up to a

carry out the inventive method.

The present invention, apart from gluing machines, can generally be used in

Find production devices that are affected by wear and tear, such as

for example in the auto industry. Such production devices can contain, for example, a spindle drive, a gear drive, a roller bearing, and / or other components. In addition, the present invention can be used for all components affected by wear, such as vehicle components that are monitored with appropriate sensors (for example vibration sensors). For example, a

Wheel bearing damage, gear damage, and the like can be predicted. In the case of an electric vehicle (with existing sensors), a prediction of damage to the

Drivetrain are carried out.

List of reference symbols

1 Procedure for determining remaining usage cycles

2 Determine first state value

3 Determine future usage cycle

10 Procedures for Determining Remaining Usage Cycles

11 Determination of the first state value

12 Processing of the first status value

13 determining at least one threshold value based on the first state value

14 Determine drift

15 Generate probability density function

16 Determine future usage cycle

17 Determining the number of remaining usage cycles

20 gluing machine

21 engine

22 motor axis

23 spindle

24 nut

25 management staff

26 pistons

27 glue chamber

28 inlet valve

29 outlet valve

30 nozzle

40 Procedure for determining remaining usage cycles

41 Performing condition monitoring

42 filters

43 Decision whether wear has taken place

44 Determination of new first status value

45 Performing drift estimation

46 Determining Future Usage Cycle

47 Analysis of whether the number of remaining usage cycles is greater than the number of usage cycles per day

48 Determination of new print data record 49 Creating a report

50 Graph for determining remaining usage cycles

51 ordinate with measured pressure

52 data points

53 time

54 Blank (production pause)

55 Healthy Threshold

56 Failure Threshold

60 graph showing degradation

61 Probability density function

62 ordinate with probability

63 abscissa with future usage cycles

64 Mean value of the probability density function

65 cross, which represent an actual number of remaining usage cycles

70 Remaining Usage Cycle Determination Device

71 Sensor interface

72 sensor

73 production device

74 Remaining Usage Cycle Determination Circuit

Claims

1. Method (1; 10; 40) for determining remaining usage cycles of a

A production device (20; 73) due to wear and tear, comprising:

Determining (2; 11; 41) a first status value of a maintenance variable of a part of the production device (20; 73) based on sensor data of a sensor for the production device; and

Determining (3; 16; 49) a future usage cycle of the part of the

Production device (20; 73) for which the maintenance variable has a predetermined second status value, the future usage cycle based on the first status value and based on a discrete

stochastic degradation model is determined.

2. The method (1; 10; 40) of claim 1, further comprising:

Determination (17) of a number of remaining usage cycles of the production device (20; 73) based on the future usage cycle.

3. The method (1; 10; 40) of claim 2, further comprising:

Generating (15) a probability density function for determining the

Remaining usage cycles.

4. The method (1; 10; 40) according to any one of the preceding claims, wherein the

Degradation model includes a Wiener process model.

5. The method (1; 10; 40) according to claim 4, wherein the degradation model comprises a Bayesian prediction model.

6. The method (1; 10; 40) of claim 5, further comprising:

Determination (14) of a drift in the Wiener process model based on the Bayesian prediction model.

7. The method (1; 10; 40) according to any one of claims 5 and 6, wherein the degradation model comprises a Monte Carlo simulation.

8. The method (1; 10; 40) according to any one of the preceding claims, further comprising:

Processing (12) the first status value.

9. The method (1; 10; 40) according to claim 8, wherein the processing comprises filtering

10. The method (1; 10; 40) according to claim 9, wherein the filtering is based on a Kalman filter.

11. The method (1; 10; 40) according to any one of claims 8 to 10, wherein the processing a

Includes rolling window regression.

12. The method (1; 10; 40) according to claim 11, wherein the processing further on a

moving average based.

13. The method (1; 10; 40) according to any one of the preceding claims, further comprising:

Determining (13) at least one threshold value based on the first state value, which is indicative of the second state value.

14. Remaining usage cycle determination circuit (74), which is configured to the

Method (1; 10; 40) to be carried out according to one of the preceding claims.

15. remaining usage cycle determining device (70) for determining

Remaining usage cycles comprising a remaining usage cycle determining circuit according to claim 14.