Methods for equipment monitoring

Abstract

The present invention relates to a method for equipment monitoring, which includes: a) providing multiple fleet measurement data for at least one equipment, wherein the multiple fleet measurement data include the number of fleet equipment data measurement "Nfleet"; b) providing multiple equipment measurement data for the monitored equipment, wherein the multiple equipment measurement data include the number of monitoring equipment data measurement "Ndata"; c) determining a fleet distribution "Fdist" determined as the distribution of differences between consecutive measurements of the multiple fleet measurement data; d) determining an equipment distribution "Ddist" determined as the distribution of differences between consecutive measurements of the equipment measurement data; e) determining the number of valid data measurements "Neff"; f) determining the failure probability of the monitored equipment, which includes using the fleet distribution "Fdist", the equipment distribution "Ddist", the number of monitoring equipment data measurements "Ndata" and the number of valid data measurements "Neff".

Description

Technical Field

[0001] The invention relates to a method for monitoring an equipment, a device for monitoring an equipment, and a computer program element and a computer-readable medium. Background Art

[0002] Degradation models track the changes in relevant variables in order to measure or predict their future evolution until a certain limit is reached, at which point the device is out of specification or fails to perform its intended function, which is often referred to as a "soft failure". An important aspect of modeling the lifetime of a device is the need to calculate the future evolution of these degradation variables as accurately as possible. The individual behavior of a device needs to be simulated or predicted, and there may not be much existing measurement data or observations. There is usually more measurement data or observations for one or more similar devices, where the sum of this measurement data is called fleet data. The predicted failure, failure time or probability of failure (PoF(t)) distribution can then be used in conjunction with other data to infer the state of the device and from this possible service or maintenance activities to avoid failures or outages or to schedule maintenance based on the predicted lifetime.

[0003] Incorporating prior knowledge to make parameter estimates more accurate for additional measurements is a well-known approach in statistics, using for example Bayesian statistics. This is often used to determine parameters for individual devices or to improve knowledge of individual devices by combining prior data and measurements. Past data is also used to predict future evolution and from this the likelihood of failure.

[0004] US 2015 / 0100284 describes a system which uses a fleet of machines and measurement data in order to calculate the risk rate of the system and thus the future reliability.

[0005] NZGebraeel et al. describe the combination of cluster / previous data with individual degradation measures in "Residual Life Distributions from ComponentDegradation Signals: A Bayesian Approach" (IIE Transactions 37, 543-557 (2005)).

[0006] S. Tang et al., “Remaining Useful Life Prediction of Lithium-Ion Batteries based on the Wiener Process with Measurement Error” (Energies 7, 520-547 (2014)) describes a method of combining fleet data and increasing measurement data.

[0007] GA Whitmore "Estimating Degradation By a Wiener Diffusion Process Subject to Measurement Error" (Lifetime Data Analysis 1, 307-319 (1995)) describes the background of MLE parameter estimation in one of the embodiments. As part of the improvement, it is also described how to improve the prediction by incorporating only some parameters (measurement error).

[0008] However, the authors found that existing techniques are not able to accurately model new or relatively new equipment to determine its probability of failure.

[0009] It is necessary to solve this problem. Summary of the invention

[0010] Therefore, it would be advantageous to have an improved capability for monitoring a device to provide information related to its expected failure lifetime.

[0011] The objects of the invention are solved by the subject-matter of the independent claims, wherein further embodiments are incorporated in the dependent claims.

[0012] In a first aspect, a device monitoring method is provided, the method comprising:

[0013] a) providing a plurality of cluster measurement data for at least one device, wherein the plurality of cluster measurement data includes a cluster device data measurement number “N” fleet ”;

[0014] b) Providing a plurality of device measurement data for the monitored device, wherein the plurality of device measurement data includes the number of times the monitoring device data is measured “N” data ”;

[0015] c) determining a fleet distribution "F" determined as a distribution of differences between successive measurements of the plurality of fleet measurement data dist ”;

[0016] d) determining a device distribution "D" determined as the distribution of differences between consecutive measurements of the device measurement data dist ”;

[0017] e) Determine the number of valid data measurements "N" eff ”;

[0018] f) Determine the failure probability of the monitored equipment, which includes using the cluster distribution "F dist ”、Equipment Distribution “D dist "、N" data" and the number of valid data measurements "N eff ”.

[0019] In one example, in step a), the plurality of cluster measurement data are for a plurality of devices, and in step c), the cluster distribution “F dist ” includes determining the difference between consecutive measurements for each device in the device.

[0020] In one example, in step e), the number of valid data measurements “N eff "Including the use of cluster distribution "F dist ” and / or equipment distribution “D dist ”.

[0021] In one example, in step e), the number of valid data measurements “N eff ” is globally determined as a fixed value. This value can be pre-selected based on actual experience or determined using the distribution of measured values, i.e., input N fleet and / or F dist Fleet data of different devices

[0022] In one example, step f) comprises:

[0023] X1) obtaining start measurement data from measurement data of multiple devices;

[0024] X2) Select the cluster distribution "F dist ” or device distribution “D dist ”, including the use of at least one selection probability;

[0025] X3) selecting a data value from the distribution selected in step X2); and

[0026] Therein the failure probability of the monitored equipment is determined based on a statistical simulation technique starting from the value selected in X1 and using successive values ​​selected in a plurality of steps X3).

[0027] In one example, the statistical simulation technique is a Monte Carlo method involving simulation of a large number of possible future paths.

[0028] In one example, in step X2), the cluster distribution “F dist The probability of "is based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N eff / (N eff +N data )” to calculate, and wherein in step A2), the device distribution “D distThe probability of " is based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N data / (N eff +N data )” to calculate.

[0029] In one example, step X3) comprises randomly selecting a data value from the distribution selected in step X2).

[0030] In one example, step f) comprises:

[0031] Y1) obtaining start measurement data from measurement data of a plurality of devices;

[0032] Y2) Select the cluster distribution "F dist ” or device distribution “D dist ”, including the use of at least one selection probability; and

[0033] The determination of the failure probability of the monitored device includes utilizing a mathematically known first passage distribution for the initial value determined in Y1) and the distribution determined in Y2).

[0034] In one example, in step Y2), the cluster distribution “F dist The probability of " is based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N eff / (N eff +N data )” to calculate, and wherein in step Y2), select the device distribution “D dist The probability of " is based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N data / (N eff +N data )” to calculate.

[0035] In one example, step f) comprises:

[0036] A1) obtaining initial measurement data from measurement data of multiple devices;

[0037] A2) Select the cluster distribution "F dist ” or device distribution “D dist ”, including the use of at least one selection probability;

[0038] A3) selecting data values ​​from the distribution selected in step A2);

[0039] A4) determining a calculated device measurement value, the calculated device measurement value being determined as the data value selected in step A3) plus the start measurement value of the first iteration or as the data value selected in step A3) plus an existing calculated device measurement value of an iteration subsequent to the first iteration;

[0040] A5) Iterating steps A2) to A4) until after a certain number of iterations the calculated device measurement value reaches a threshold value, and wherein the determination of the probability of failure comprises utilizing the number of iterations to reach the threshold value.

[0041] In one example, the method includes:

[0042] A6) performing step A5) a plurality of times, and wherein the determination of the probability of failure comprises, for each of the plurality of performances of step A5), utilizing a number of iterations to reach a threshold value.

[0043] In one example, in step A2), the cluster distribution “F dist The probability of " is based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N eff / (N eff +N data )” to calculate, and wherein in step A2), the device distribution “D dist The probability of " is based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N data / (N eff +N data )” to calculate.

[0044] In one example, step A3) comprises randomly selecting data values ​​from the distribution selected in step A2).

[0045] In one example, step f) comprises:

[0046] B1) Using cluster distribution “F dist ” and equipment distribution “D dist "Determine the combined distribution.

[0047] In one example, the data value in the combined distribution is determined as the number of valid data measurements multiplied by the fleet distribution plus the number of monitoring device data measurements multiplied by the device distribution and the sum of the above divided by the number of valid data measurements and the number of monitoring device data measurements "(N eff *F dist +N data *D dist ) / (N eff +N data )”.

[0048] In one example, the data values ​​in the combined distribution are determined by applying statistical inference, where the number of valid data measurements “N eff " and the number of monitoring equipment data measurements "N data ” is an integral part of the inference.

[0049] In one example, step f) comprises:

[0050] B2) obtaining starting measurement data from measurement data of a plurality of devices;

[0051] B3) selecting a data value from the combined distribution determined in step B1);

[0052] B4) determining a calculated device measurement value, the calculated device measurement value being determined as the data value selected in step B3) plus the start measurement value of the first iteration or as the data value selected in step B3) plus an existing calculated device measurement value of an iteration subsequent to the first iteration;

[0053] B5) Iterating steps B3) to B4) until after a certain number of iterations the calculated device measurement value reaches a threshold value, and wherein the determination of the failure probability comprises utilizing the number of iterations to reach the threshold value.

[0054] In one example, the method includes:

[0055] B6) performing step B5) a plurality of times, and wherein the determination of the probability of failure comprises, for each of the plurality of performances of step B5), utilizing a number of iterations to reach a threshold value.

[0056] In one example, step B3) comprises randomly selecting a data value from the combined distribution.

[0057] In one example, step f) comprises:

[0058] D1) Determine the first failure probability “Prob1”, including the cluster distribution “F dist ”Use;

[0059] D2) Determine the second failure probability "Prob2", including the equipment distribution "D dist ”Use;

[0060] D3) Determining a failure of the monitored device, including utilizing a first failure probability "Prob1" and a second failure probability "Prob2".

[0061] In one example, step D1) includes modifying the parameters of the composite distribution until the composite distribution matches the cluster distribution “F dist ”, and wherein step D2) comprises modifying the parameters of the composite distribution until the composite distribution matches the device distribution “D dist”.

[0062] In other words, a parametric approach is used, i.e. the parameters of the assumed distribution are fitted so that it represents the empirical distribution of the fleet / equipment as well as possible. This parameter can then be used as an indication of the probability of failure.

[0063] In one example, step D3) includes calculating the total of the valid data measurement times the first failure probability plus the monitoring device data measurement times the second failure probability and dividing the total by the sum of the valid data measurement times and the monitoring device data measurement times “(N eff *Prob1+N data *Prob2) / (N eff +N data )” to determine the combined probability.

[0064] In a second aspect, an apparatus for equipment monitoring is provided, wherein the apparatus is configured to perform the method of the first aspect.

[0065] According to another aspect, a computer program element is provided for controlling an apparatus or a system as described previously, which computer program element is adapted to perform the method steps as described previously when the computer program element is executed by a processing unit.

[0066] According to another aspect, a computer readable medium storing the above-mentioned computer unit is also provided.

[0067] The above aspects and examples will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Exemplary embodiments will be described below with reference to the following drawings:

[0069] Figure 1 An example of a method for determining a probability of equipment failure is shown;

[0070] Figure 2 An example of a method for determining a probability of equipment failure is shown; and

[0071] Figure 3 An example of a method for determining a probability of equipment failure is shown. DETAILED DESCRIPTION

[0072] Figure 1-Figure 3 The invention relates to a device monitoring method and apparatus. One example relates to a device monitoring method. The method comprises:

[0073] a) providing a plurality of cluster measurement data for at least one device, wherein the plurality of cluster measurement data includes a cluster device data measurement number “N” fleet ”;

[0074] b) Providing a plurality of device measurement data for the monitored device, wherein the plurality of device measurement data includes the number of times the monitoring device data is measured “N” data ”;

[0075] c) determining a fleet distribution "F" determined as a distribution of differences between successive measurements of the plurality of fleet measurement data dist ”;

[0076] d) determining a device distribution "D" determined as the distribution of differences between consecutive measurements of the device measurement data dist ”;

[0077] e) Determine the number of valid data measurements "N" eff ”;

[0078] f) Determine the failure probability of the monitored equipment, which includes using the cluster distribution "F dist ”、Equipment Distribution “D dist "、N" data " and the number of valid data measurements "N eff ”.

[0079] According to an example, in step a), a plurality of cluster measurement data are for a plurality of devices, and in step c), the cluster distribution “F dist ” includes determining the difference between consecutive measurements for each device in the device.

[0080] According to one example, in step e), the number of valid data measurements “N eff "Including the use of cluster distribution "F dist ” and / or equipment distribution “D dist ”.

[0081] According to one example, in step e), the number of valid data measurements “N eff ” is globally determined as a fixed value.

[0082] According to one example, step f) comprises:

[0083] X1) obtaining start measurement data from measurement data of multiple devices;

[0084] X2) Select the cluster distribution "F dist ” or device distribution “D dist ”, including the use of at least one selection probability;

[0085] X3) selecting a data value from the distribution selected in step X2); and

[0086] Therein the failure probability of the monitored equipment is determined based on a statistical simulation technique starting from the value selected in X1 and using successive values ​​selected in a plurality of steps X3).

[0087] According to one example, the statistical simulation technique is a Monte Carlo method involving simulation of a large number of possible future paths.

[0088] According to an example, in step X2), the cluster distribution “F dist The probability of "is based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N eff / (N eff +N data )” to calculate, and wherein in step A2), the device distribution “D dist The probability of " is based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N data / (N eff +N data )” to calculate.

[0089] According to one example, step X3) comprises randomly selecting data values ​​from the distribution selected in step X2).

[0090] According to one example, step f) comprises:

[0091] Y1) obtaining start measurement data from measurement data of a plurality of devices;

[0092] Y2) Select the cluster distribution "F dist ” or device distribution “D dist ”, including the use of at least one selection probability; and

[0093] The determination of the failure probability of the monitored device includes utilizing a mathematically known first channel distribution for the initial value determined in Y1) and the distribution determined in Y2).

[0094] According to an example, in step Y2), the cluster distribution “F dist The probability of "is based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N eff / (N eff +N data )” to calculate, and wherein in step Y2), select the device distribution “D dist The probability of " is based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N data / (N eff +N data )” to calculate.

[0095] According to one example, step f) comprises:

[0096] A1) obtaining initial measurement data from measurement data of multiple devices;

[0097] A2) Select the cluster distribution "F dist ” or device distribution “D dist ”, including the use of at least one selection probability;

[0098] A3) selecting data values ​​from the distribution selected in step A2);

[0099] A4) determining a calculated device measurement value, the calculated device measurement value being determined as the data value selected in step A3) plus the start measurement value of the first iteration or as the data value selected in step A3) plus an existing calculated device measurement value of an iteration subsequent to the first iteration;

[0100] A5) Iterating steps A2) to A4) until after a certain number of iterations the calculated device measurement value reaches a threshold value, and wherein the determination of the probability of failure comprises utilizing the number of iterations to reach the threshold value.

[0101] According to one example, the method includes:

[0102] A6) performing step A5) a plurality of times, and wherein the determination of the probability of failure comprises, for each of the plurality of performances of step A5), utilizing a number of iterations to reach a threshold value.

[0103] According to an example, in step A2), the cluster distribution “F dist The probability of "is based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N eff / (N eff +N data )” to calculate, and wherein in step A2), the device distribution “D dist The probability of " is based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements" N data / (N eff +N data )” to calculate.

[0104] According to one example, step A3) comprises randomly selecting a data value from the distribution selected in step A2).

[0105] According to one example, step f) comprises:

[0106] B1) Using cluster distribution “F dist ” and equipment distribution “Ddist "Determine the combined distribution.

[0107] According to one example, the data value in the combined distribution is determined as the number of valid data measurements multiplied by the fleet distribution plus the number of monitoring device data measurements multiplied by the device distribution and the sum of the above divided by the number of valid data measurements and the number of monitoring device data measurements "(N eff *F dist +N data *D dist ) / (N eff +N data )”.

[0108] According to one example, the data values ​​in the combined distribution are determined by applying statistical inference, where the number of valid data measurements “N eff " and the number of monitoring equipment data measurements "N data ” is an integral part of the inference.

[0109] According to one example, step f) comprises:

[0110] B2) obtaining starting measurement data from measurement data of a plurality of devices;

[0111] B3) selecting a data value from the combined distribution determined in step B1);

[0112] B4) determining a calculated device measurement value, the calculated device measurement value being determined as the data value selected in step B3) plus the start measurement value of the first iteration or as the data value selected in step B3) plus an existing calculated device measurement value of an iteration subsequent to the first iteration;

[0113] B5) Iterating steps B3) to B4) until after a certain number of iterations the calculated device measurement value reaches a threshold value, and wherein the determination of the failure probability comprises utilizing the number of iterations to reach the threshold value.

[0114] According to one example, the method includes:

[0115] B6) performing step B5) a plurality of times, and wherein the determination of the probability of failure comprises, for each of the plurality of performances of step B5), utilizing a number of iterations to reach a threshold value.

[0116] According to one example, step B3) comprises randomly selecting a data value from the combined distribution.

[0117] According to one example, step f) comprises:

[0118] D1) Determine the first failure probability “Prob1”, including the cluster distribution “F dist ”Use;

[0119] D2) Determine the second failure probability "Prob2", including the equipment distribution "D dist ”Use;

[0120] D3) Determining a failure of the monitored device, including utilizing a first failure probability "Prob1" and a second failure probability "Prob2".

[0121] According to one example, step D1) comprises modifying the parameters of the composite distribution until the composite distribution matches the cluster distribution “F dist ”, and wherein step D2) comprises modifying the parameters of the composite distribution until the composite distribution matches the device distribution “D dist ”.

[0122] According to one example, step D3) includes determining the combined probability based on the number of valid data measurements multiplied by the first failure probability plus the number of monitoring device data measurements multiplied by the second failure probability and the whole divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "(Neff*Prob1+Ndata*Prob2) / (Neff+Ndata)".

[0123] It can be seen from the above that a device can be configured to execute the above method, thereby providing a device monitoring device.

[0124] Thus, the merging of the two data (fleet data and equipment) takes place directly in the determination of the process characteristics before simulating / determining the future evolution. Another advantage is that the combination of non-homogeneous data (equipment data / failure rate data) is suitable for hybrid models.

[0125] Recall that the simulation or calculation of the future evolution of the degradation variable x requires a good understanding of the underlying processes, which are usually not deterministic but have a stochastic character. The properties of this underlying stochastic process may vary from device to device, and in particular these individual properties need to be used in order to improve the predictions compared to pure fleet or ensemble statistical methods. On the other hand, especially for new devices, only very limited data are available and the determination of the underlying process is therefore usually very uncertain. This has been addressed in the present approach by using knowledge from the fleet (“prior knowledge”) in combination with device data to improve the predictions. Over time and as the amount of device data increases, the relevance of the fleet data is gradually replaced by better device data.

[0126] The development method described here combines fleet data and equipment data in a consistent manner, and selects parameters that can specify how much trust should be placed in the fleet data compared to the equipment data. In addition, statistical data (from service reports, durability testing, maintenance assessments, etc.) can be combined with the fleet data / equipment data for use in parallel to improve the accuracy of the predictions and the robustness of the prediction algorithms, especially when the amount of individual data is not large at the beginning.

[0127] So, in summary, a hybrid strategy is used to determine the characteristics of the process, rather than combining them on a lifecycle probability distribution, which is often used. In this new approach, a generic / fleet probability distribution is mixed with a probability distribution from the equipment in a predetermined way that changes the weight or emphasis of each model in a predetermined way. Here, the merging of the two data occurs directly in the determination of the process characteristics before simulating / determining the future evolution. This allows a more mathematically consistent approach to be adopted, and has the advantage of capturing deviations from general behavior in a dynamic manner based on equipment data. Another advantage is that the combination of non-homogeneous data (equipment data / failure rate data) is suitable for hybrid models. This enables more accurate predictions of the remaining life of the equipment, allowing customers to better plan service and maintenance, and detect problems before failures occur. In order to provide the expected life in the initial stages of equipment operation, fleet data is used, but as more equipment data becomes available, the relevance of the fleet data gradually decreases, so that the fleet data does not hinder future predictions.

[0128] In more detail, the degradation variable x(t) or x_i is under consideration, and of particular interest is the time evolution of the degradation variable x(t) or x_i, i.e., how the parameter or variable of the device changes over time until it reaches a critical value, which is then defined as a "failure" of the device. Therefore, in relation to the present disclosure reference is made to distributions and parameters, which describe this time evolution.

[0129] As a specific example, the variables being observed (or measured) are typically those related to the operation of the device: timing information (i.e., e.g., how long it takes to fully open or close a device such as a circuit breaker), information about the speed of a specific component at a specific time, distance, such as overshoot relative to normal operation or bounce levels. Other variables might be physical properties, such as the energy stored in a spring, the pressure in an air chamber, the amplitude of some (acoustic) signal at a specific frequency, etc. In other fields, this might be battery capacity, resistor values, etc. The important point is that these correspond to the state of the device, in contrast to external variables such as temperature, current, external vibrations, etc.

[0130] Usually, the time evolution of these variables is described by looking at the difference of two consecutive measurements of them. Here, continuous can mean two consecutive operations if the variable (such as timing information) can only be observed during this operation, or with a fixed time interval if the variable (such as the pressure inside a container) can be measured independently. Since the process is not of deterministic type in most cases, it needs to be described as a random process, that is, given the (multiple) values ​​(until) now, the distribution of the value of the next measurement needs to be specified. This is the distribution F(dist) referred to. It is the distribution of the differences of consecutive measurements, for example, Deltatx_i = (x_i-x_(i-1)). The distribution D(dist) of the monitored equipment is determined similarly.

[0131] With respect to the fleet data associated with the individual data of the monitored devices, the time evolution of the degradation variables is a device-specific process, that is, no two devices will behave the same. This is due to the randomness of the process itself (i.e., even identical devices will not behave the same), and also because the parameters defining the process dynamics will vary (some devices will have stronger trends than others, and some devices will have greater dispersion than others). The fleet of devices then tells us something about the “average” or “expected” parameters or distributions. This is the best information that can be used if no further information from the monitored devices is available. On the other hand, if a large number of operations of the monitored devices have been measured, these operations can be analyzed to obtain the parameters or distributions of individual devices (such as circuit breakers). This is the other extreme, and the method currently described enables a gradual transition from one to the other.

[0132] Therefore, data from a fleet of potentially large numbers of devices (i.e., a large number of data) are analyzed to determine the distribution of the degradation variables, F(dist). Since there is a large number of data, called N(fleet), this is an accurate representation of the "true" distribution. An analysis is then performed to determine the effective number of observations, Neff. This "true" distribution is therefore related to the "average" or "expected" distribution or parameters. Neff can be derived from this, it is not equal to the total number of measurements, or it can be fixed.

[0133] Then, for the monitored equipment, there is equipment data (called N(data)) for one device (where the device is similar to the one or more devices that provide fleet data), and for these devices, there is not much data, but the amount of data is increasing. This data is analyzed to determine the distribution of the degradation variables D(dist). However, since there is not much data, especially in the initial stages of the equipment life cycle, the distribution is not very good and it is difficult to use it to determine the probability of failure (PoF) or remaining useful life (RUL). However, the data is always increasing because more data is obtained from more and more operations over time. Therefore, as mentioned above, only this data is considered, which is D(dist).

[0134] Typically, the number of "observations" (i.e., the number of measurements taken for the entire fleet) is much larger than the number of individual devices. Therefore, in a simple combination, the data from the monitored device will overwhelm the data from the fleet, and the data from the device itself will never be seen. Therefore, it is necessary to reduce the importance of the fleet data. In the present method, this is achieved by using Neff as the "effective" number of observations.

[0135] One way to calculate Neff is to compare the overall variability of the fleet and see how this compares to the overall variability of the equipment since only a finite number of Ndata measurements are observed. If the two are the same size (assuming the overall variability of the equipment decreases with N), then this is called Neff. Another way to express this is as prior knowledge in a Bayesian interpretation. The knowledge before the equipment is put into operation is not zero. This can be based on the knowledge of the fleet, but still remember that individuals can be very different from ordinary individuals. In Bayesian statistics, the amount of knowledge can be expressed as the number of experiments that are assumed to be performed to obtain the current knowledge of the equipment. While these are more mathematical concepts, it is also practical to choose a fixed value for Neff as a value where information about the fleet becomes equally important as information from the equipment measurements.

[0136] The purpose of the developed prognostic analysis is to understand the future evolution of the degradation variable. Since this is a stochastic process, this is not a deterministic evolution, but needs to be described by a probability distribution. PoF is such a distribution. It records the probability that the degradation variable will exceed a certain (predefined) limit in the future. As the process deviates more and more from the current value, this probability increases over time until the device fails with a probability of 1.

[0137] There are generally two ways to calculate PoF: In the case of a known process, such as the so-called Wiener process, there are analytical expressions that can be used to directly calculate PoF. A more general and also more intuitive approach is to use the Monte Carlo method: Here, starting with the current value x_i, an increment \Delta x_(i+1) is then randomly drawn from the distribution to obtain the next value x_(i+1) = x_1 + \Delta x_(i+1). This operation is repeated again, and one possible path that the degraded variable may take in the future is obtained. These steps can be repeated until a barrier (threshold), i.e. a critical value, is reached. Then, the failure time t_fail for this path can be obtained. In general, this is then repeated many times. Due to the randomness, each path will be slightly different. Therefore, a large number of possible failure times are determined overall. The histogram of these is the predicted PoF.

[0138] Therefore, the distribution of changes in the degradation variables of the fleet data D(fleet) and the monitored equipment data D(dist) are combined using the effective number of observations Neff and the monitored equipment observations Ndata. There are a number of ways to do this, as explained in more detail below.

[0139] Figure 1 A method is shown that uses a statistical combination of fleet and equipment data. The fleet data is collected and converted into a distribution F(\Delta), also called F(dist), which describes the probability of obtaining a specific value. The same operation is done using all the data available so far from the equipment D(\Delta), also called D(dist). In order to combine the two, a two-step process needs to be completed. The first process randomly selects one of the two distributions with a certain probability, which is calculated as a function of Neff, the equipment data and the fleet data, in the simplest case as a function of Neff and Ndata. This combined distribution is then used as the basis for calculating the further evolution of the degradation variable x (for example, using a Monte Carlo method). The distribution of failure time, i.e., PoF(t) and all other KPIs derived from it, is thus obtained.

[0140] Therefore, we select from each distribution in a random manner. Assume that both F(dist) and D(dist) are in the form of known distributions, so we can sample values ​​of \Delta x_i from each of them (either because they are in the form of histograms or because we know how to draw random numbers from the distributions). We then set up a two-step process:

[0141] First, randomly choose one of the two distributions. The probability of choosing F should be Neff / (Neff+Ndata), and the probability of choosing D should be Ndata / (Neff+Ndata);

[0142] Then, a delta \Delta x is selected from either F(dist) or D(dist). This value is updated and the process is repeated to generate a large number of possible paths until failure.

[0143] During the Monte Carlo simulation, the values ​​of Neff and Ndata remain fixed because they represent the current amount of data.

[0144] This two-step approach is one possible approach because it is quite flexible and general. But of course, other approaches can be chosen. If the distributions are approximated in histogram form, then these distributions can be combined into a common histogram by adding the densities as (Neff*F(dist)+Ndata*D(dist)) / (Neff+Ndata).

[0145] Figure 2 A method is shown which relates to combining parameters of underlying statistical distributions. In a first step, the distribution of the parameter inferred from the fleet data is determined, and in a second step the same is done for the monitored device data. The distribution of this parameter embodies both a value and a certainty. Usually, the certainty of the fleet data is not really useful because it embodies device-specific variations. Therefore, a distribution is instead constructed using Neff which embodies the best estimate of the fleet parameter and the assumed certainty corresponding to the result of a valid number of measurements made on the device. In a last step, these two are combined into a single distribution, for example using Bayesian inference. In most cases, this is done by calculating the best parameter value given both the fleet and the data. This is then used as a basis for further calculations of the time evolution or for directly obtaining the PoF distribution.

[0146] Therefore, using parametric statistical methods, where the distribution F(dist) is given as some function F(\theta_fleet) and D(dist) is given as some function F(\theta_data), and by applying statistical inference, such as Bayesian statistics, \theta_fleet and \theta_data can be combined into a common \theta_combined. In this combination, Neff and Ndata are the components of the inference.

[0147] A parameter is therefore a value that describes the distribution of the degradation variable (the difference of continuous degradation variables). For example, a very common model for random processes is the Wiener process. One of them describes normally distributed deltas, i.e., each \Delta x_i is assumed to come from the same normal distribution. To specify the normal distribution, on the one hand, you need the mean value mu, which corresponds to the drift of the path, and on the other hand, the standard deviation sigma, which describes the variation around this drift. The mean and the standard deviation are the two parameters needed in this case and therefore need to be determined. In the Monte Carlo method, there is then a normal distribution whose parameters are determined by the fleet and another normal distribution whose parameters are determined by the measurements of the equipment monitored so far. Then, data are extracted from each of them. This is Figure 1 (assuming a histogram rather than a normal distribution). Figure 2 In , different approaches are shown. Here, the increments are taken from a normal distribution where the mean and standard deviation are a combination of the fleet parameters and the device values. As the simplest example, a weighted average can be used:

[0148] mu=(Neff mu_fleet+Ndata mu_data) / (Neff+Ndata),

[0149] sigma^2=(Neff sigma_fleet^2+Ndata sigma_data^2) / (Neff+Ndata)

[0150] There are of course more sophisticated methods in statistics on how to combine them, this is just an illustration of that method.

[0151] Figure 3 A method is shown where the combination is not done at the level of the time evolution of the degradation variables, but at the PoF. The fleet data is analyzed and the PoF is derived from it. This can be done using Monte Carlo methods or other methods, for example, by directly analyzing the failure time distribution. In a second step, an estimate of the PoF of the device considering only the device measurements is calculated. The combination of these two is then done at the two PoF levels, where the probability of choosing one of the two is calculated again based on Neff, the fleet and the device data.

[0152] Note that Figure 1 and Figure 3 In FIG. 1 , p_fleet=f(data, fleet) and p_data=1-f(data, fleet) are shown, which are the probabilities of selecting a fleet or data distribution. These probabilities are also called Prob1 and Prob2. They are used to select Figure 1The combined distribution of F_dist or D_dist shown in the histogram is used to select Figure 3 The failure probability distribution in PoF_fleet(t) or PoF_data(t).

[0153] Thus, in a non-parametric approach, the merging of the two data will occur in the following way: both the fleet data and the equipment data are collected and used to infer the distribution of the variation of the degradation variable, for example using a histogram or an empirical cumulative distribution function (ecdf) as a basis. The fleet data are more numerous and therefore closer to the true distribution of the average equipment, while the equipment monitored in the initial stage can only be based on a few observations made and the histogram is quite uncertain. The two distributions are then used to simulate the future evolution of the monitored equipment. To this end, the two distributions are merged based on an additional parameter (Neff) that is set to weight the two inputs based on the "valid number of observations" to which the fleet data are assumed to correspond. The future variation of the degradation variable is then derived from the two histograms, with a probability that depends on the ratio of Neff and the number of observations of the equipment Ndata. In this way, a gradual transition from the fleet to the equipment data can be guaranteed, where the larger deviations of the equipment are already taken into account early in the simulation. Alternatively, the probability does not need to be based on this, but can be done based on a "goodness of fit" criterion that compares the degree of similarity of the properties of the equipment data to the properties of the fleet data (for example, using the Kolmogorov Smirnov test criterion or a similar distribution test). In this way, device data is used to switch faster in situations where a specific device differs significantly from average or fleet behavior.

[0154] However, in the parametric approach, the merging of the two data is done at the parameter level. The first parameter distribution can be determined from the fleet data. Since there is a large amount of data available (Nfleet), the (average) parameters in this case can be determined quite accurately. This distribution is also static or usually only updated periodically. The second parameter distribution can be determined from the device data. Since the amount of data (Ndata) is small, it is quite uncertain. Using the number of valid observations (Neff), which is considered to be a typical value of the accuracy of the fleet data when applied to individual devices, as an additional input, the previous fleet data distribution of the parameter is determined. In the third step, the two distributions are combined and the connection parameter is determined from this, which can be regarded as a weighted value. By using two parameter distributions, a relative weighting based on the consistency of the two or the accuracy of the field data compared to the fleet data is automatically achieved. Finally, the first channel time distribution of the degradation change distribution is calculated by simulation or analysis, and the connection parameter is used as the basis for predicting future evolution.

[0155] However, in another approach, there is a combination of two different data sources (device / fleet data and statistical failure rate data), and two different models are applied in parallel. A specific weighting function is applied to the KPIs of both models so that RUL and PoF have common KPIs.

[0156] So the summary is as follows:

[0157] 1. Use fleet data to obtain the distribution of changes in variables within the degradation model;

[0158] Alternatively, this first distribution can also come from expert evaluation;

[0159] 2. Obtaining equipment data online up to the current point and using it to obtain a second distribution representing the expected change of the variables within the degradation model;

[0160] 3. Use additional parameters to weight the extent to which the fleet data is applicable to the individual equipment. This can be in the form of valid measurement points.

[0161] 4. The two distributions are merged into a composite where the relative weights of each part are chosen in a way that, i.e., the added knowledge through additional measurements leads to a shift from a generic fleet distribution to a device-specific distribution;

[0162] 5. Calculate the future evolution of the degradation variable, in particular its distribution about the channel of one or more critical values, by simulation or other numerical / analytical tools.

[0163] So, in general, at the beginning of the system lifecycle, there is little device data from sensors, so the statistical data KPIs are more important (they have a decreasing weighting function). During the lifecycle, more data is collected for the monitored devices, and the KPIs in the device data increase their weight in the results.

[0164] So in the detailed workflow, the process starts with the fleet data where statistical parameter estimation techniques are used to determine the best choice for the parameters of the underlying distribution. MLE (maximum likelihood estimation) is a very common approach. The three parameters in the model are used along with Neff to construct a prior distribution, i.e., not just a single value of the three parameters, but the overall distribution. Neff can be used directly on one of the features of the model.

[0165] In addition, there is equipment data (e.g., circuit breakers). In Bayesian methods, the prior data is combined with the likelihood of the measurements to get the posterior distribution (this is the core of Bayesian methods). This is the combination step. From this posterior distribution, a value is again selected and a procedure is called. So this produces three values, which are the combined parameters of the distribution.

[0166] In order to use these values ​​to calculate PoF and other KPIs like RUL, there are two possible approaches, both of which are possible in this case: This can be done with Monte Carlo simulations, and as these simulations cross the barriers, the PoF can be constructed. Alternatively, in this case, there is an analytical expression that arrives directly at the PoF.

[0167] In a further exemplary embodiment, a computer program or a computer program element is provided, characterized in that it is configured to execute the method steps of the method according to one of the preceding embodiments on a suitable system.

[0168] Therefore, the computer program element can be stored on a computer unit, which can also be a part of an embodiment. The computing unit can be configured to execute or induce the steps of the above method. In addition, it can be configured to operate the components of the above device and / or system. The computing unit can be configured to automatically operate and / or execute the user's command. The computer program can be loaded into the working memory of the data processor. The data processor can therefore be equipped to execute the method according to one of the aforementioned embodiments.

[0169] According to another exemplary embodiment of the present invention, a computer readable medium, such as a CD-ROM, is provided, wherein the computer readable medium has a computer program element stored thereon, the computer program element being described in the preceding section.

[0170] Although the present invention has been described and illustrated in detail in the drawings and the foregoing description, such description and illustration are to be considered illustrative or exemplary rather than restrictive. The present invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments may be understood and effected by those skilled in the art in practicing the claimed invention by studying the drawings, the disclosure, and the appended claims.

Claims

1. A method for equipment monitoring, comprising: a) providing a plurality of fleet measurement data for at least one device, wherein the plurality of fleet measurement data includes a number of fleet device data measurement times "Nfleet"; b) providing a plurality of device measurement data for the monitored device, wherein the plurality of device measurement data includes the number of monitoring device data measurement times "Ndata"; c) determining a fleet distribution "Fdist" determined as the distribution of differences between consecutive measurements of said plurality of fleet measurement data; d) determining a device distribution "Ddist" determined as the distribution of differences between consecutive measurements of said device measurement data; e) Determine the number of valid data measurements "Neff"; f) determining the failure probability of the monitored device, the determination comprising using the cluster distribution "Fdist", the device distribution "Ddist", the monitoring device data measurement number "Ndata" and the effective data measurement number "Neff", and Wherein said step f) comprises: A1) acquiring initial measurement data from the plurality of device measurement data; A2) selecting the cluster distribution "Fdist" or the device distribution "Ddist", including utilizing at least one selection probability; A3) selecting a data value from said distribution selected in step A2); A4) determining a calculated device measurement value, the calculated device measurement value being determined as the data value selected in step A3) plus a start measurement value for a first iteration or as the data value selected in step A3) plus an existing calculated device measurement value for an iteration subsequent to the first iteration; A5) Iterating steps A2) to A4) until, after a certain number of iterations, said calculated device measurement value reaches a threshold value, and wherein said determination of said probability of failure comprises utilizing said number of iterations to reach said threshold value.

2. The method of claim 1 , wherein in step a), the plurality of fleet measurement data are for a plurality of devices, and wherein in step c), the determination of the fleet distribution “Fdist” comprises determining the difference between consecutive measurements for each of the devices.

3. The method according to any one of claims 1 to 2, wherein in step e), determining the number of valid data measurements "Neff" includes using the cluster distribution "Fdist" and / or the device distribution "Ddist".

4. The method according to any one of claims 1 to 2, wherein in step e), the number of valid data measurements "Neff" is globally determined as a fixed value.

5. The method according to any one of claims 1 to 2, wherein step f) comprises: X1) acquiring start measurement data from the plurality of device measurement data; X2) selecting the cluster distribution "Fdist" or the device distribution "Ddist", including utilizing at least one selection probability; X3) selecting a data value from said distribution selected in step X2); and Wherein the failure probability of the monitored equipment is determined from a statistical simulation technique starting from the value selected in X1 and using successive values ​​selected in a plurality of steps X3).

6. A method according to claim 5, wherein the statistical simulation technique is a Monte Carlo method involving the simulation of a large number of possible future paths.

7. The method according to claim 5, wherein in step X2), the probability of selecting the cluster distribution "Fdist" is calculated based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "Neff / (Neff+Ndata)", and wherein in step A2), the probability of selecting the device distribution "Ddist" is calculated based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "Ndata / (Neff+Ndata)".

8. The method according to claim 5, wherein step X3) comprises randomly selecting the data value from the distribution selected in step X2).

9. The method according to any one of claims 1 to 2, wherein step f) comprises: Y1) obtaining start measurement data from the plurality of device measurement data; Y2) selecting the cluster distribution "Fdist" or the device distribution "Ddist", including utilizing at least one selection probability; as well as The determination of the failure probability of the monitored device includes: using a mathematically known first channel distribution for the initial value determined in Y1) and the distribution determined in Y2).

10. The method according to claim 9, wherein in step Y2), the probability of selecting the cluster distribution "Fdist" is calculated based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "Neff / (Neff+Ndata)", and wherein in step Y2), the probability of selecting the device distribution "Ddist" is calculated based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "Ndata / (Neff+Ndata)".

11. The method according to claim 1, wherein the method comprises: A6) performing step A5) multiple times, and wherein the determining of the failure probability comprises: for each of the multiple executions of step A5), using the number of iterations to reach the threshold.

12. The method according to any one of claims 1 or 11, wherein in step A2), the probability of selecting the cluster distribution "Fdist" is calculated based on the number of valid data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "Neff / (Neff+Ndata)", and wherein in step A2), the probability of selecting the device distribution "Ddist" is calculated based on the number of monitoring device data measurements divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "Ndata / (Neff+Ndata)".

13. The method according to any one of claims 1 and 11, wherein step A3) comprises randomly selecting the data value from the distribution selected in step A2).

14. The method according to claim 1, wherein step f) comprises: B1) Determine the combined distribution using the cluster distribution "Fdist" and the device distribution "Ddist".

15. The method according to claim 14, wherein the data value in the combined distribution is determined as the number of valid data measurements multiplied by the cluster distribution plus the number of monitoring device data measurements multiplied by the device distribution and the above whole divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "(Neff*Fdist+Ndata*Ddist) / (Neff+Ndata)".

16. The method of claim 14, wherein the data values ​​in the combined distribution are determined by applying statistical inference, wherein the number of valid data measurements "Neff" and the number of monitoring device data measurements "Ndata" are components of the inference.

17. The method according to any one of claims 14 to 16, wherein step f) comprises: B2) acquiring start measurement data from the plurality of device measurement data; B3) selecting a data value from said combined distribution determined in step B1); B4) determining a calculated device measurement value, the calculated device measurement value being determined as the data value selected in step B3) plus the start measurement value for a first iteration or as the data value selected in step B3) plus an existing calculated device measurement value for an iteration subsequent to the first iteration; B5) Iterating steps B3) to B4) until, after a certain number of iterations, said calculated device measurement value reaches a threshold value, and wherein said determination of said probability of failure comprises utilizing said number of iterations to reach said threshold value.

18. The method according to claim 17, wherein the method comprises: B6) performing step B5) multiple times, and wherein the determining of the failure probability comprises, for each of the multiple executions of step B5), utilizing the number of iterations to reach the threshold.

19. The method according to any one of claims 17, wherein step B3) comprises randomly selecting the data value from the combined distribution.

20. The method according to claim 1, wherein step f) comprises: D1) determining a first failure probability "Prob1", including utilization of said fleet distribution "Fdist"; D2) determining a second failure probability "Prob2" including utilization of said device distribution "Ddist"; D3) Determining the failure of the monitored device, including utilization of the first failure probability "Prob1" and the second failure probability "Prob2".

21. The method of claim 20, wherein step D1) comprises modifying parameters of a synthetic distribution until the synthetic distribution matches the cluster distribution "Fdist", and wherein step D2) comprises modifying parameters of a synthetic distribution until the synthetic distribution matches the device distribution "Ddist".

22. The method according to any one of claims 20 to 21, wherein step D3) includes determining the combined probability based on the number of valid data measurements multiplied by the first failure probability plus the number of monitoring device data measurements multiplied by the second failure probability and the whole divided by the sum of the number of valid data measurements and the number of monitoring device data measurements "(Neff*Prob1+Ndata*Prob2) / (Neff+Ndata)".

23. An apparatus for equipment monitoring, wherein the apparatus is configured to perform the method according to any one of claims 1 to 22.

24. A computer program element for controlling an apparatus according to claim 23, the computer program element being configured to perform the method according to any one of claims 1 to 22 when being executed by a processor.

25. A computer readable medium storing a computer program element according to claim 24.

Images