Vibration monitoring for fault detection in process automation

Description

[0001] The invention relates to a computer-implemented monitoring method for vibration analysis for the detection of machine or hardware faults in process automation.

[0002] It is known from the prior art to monitor machine-executed processes to detect errors or failures in the machines, computers, etc. used. For this purpose, sensors are typically employed that continuously record data during the process. This data is then evaluated according to a defined criterion to determine whether an error has occurred. For example, DE 10 2018 222 562 A1 discloses a method for detecting a fault condition in the excitation circuit of an electric machine, whereby the excitation current and the control signal are filtered and compared with reference values ​​as sensor-acquired comparison signals. EP 0 934 567 A1 discloses a method for classifying the statistical dependence of a measurable time series, which describes iterative tests with different null hypotheses.From DE 10 2019 107 363 A1, a computer-implemented method for outputting a wear signal from a machine tool using a so-called "change point detection" with distance metric is known. However, this method is generally not reliable enough for long-term, 24 / 7 monitoring.

[0003] Furthermore, the assessment of data quality based on null hypotheses is also known from L. Shalalfeh et al., Fractional Dynamics of PMU Data (IEEE Transactions on Smart Grid, IEEE, USA, Vol. 12, No. 3, pp. 1-11, May 2021), XP011850407, ISSN: 1949-3053, DOI: 10.1109 / TSG.2020.3044903.

[0004] From EP 0 907 913 B1 a method for diagnostic differentiation is also known in which deviations from the rule are measured with a histogram, whereby conclusions can be drawn about different sources of disturbance.

[0005] The object of the invention is to propose a monitoring method for fault detection which enables particularly reliable fault detection.

[0006] The problem is solved, starting from previous monitoring methods known from the prior art, by the features of claim 1.

[0007] The measures mentioned in the dependent claims make advantageous embodiments and further developments of the invention possible.

[0008] The monitoring method according to the invention is intended for use in process automation for vibration analysis. For example, a machine with a rotor, such as a fan, can be examined for possible defects such as slowly developing, trending deviations or the like by detecting the vibration and its deviations. Accordingly, the monitoring method according to the invention for detecting a machine fault or a hardware fault resulting from a vibration analysis in a machine-executed process also initially comprises the acquisition of at least one series of sensor data measured during the process. These sensor-acquired data are often recorded sequentially. However, it is conceivable, for example, to record them in a time sequence.also that the series of sensor data represents a local sequence or a local progression, for example, when it is to be sensorily detected whether the same distance is always maintained between two parts or whether a component does not leave its intended storage position due to heating, imbalances or other irregularities.

[0009] The invention is characterized by the fact that the desired fault detection can also include fault prediction. The invention recognizes that machine or hardware faults, particularly in the monitoring of vibrations or rotations, often "announce" themselves. That is, a sudden, spontaneous failure of the machine without any warning rarely occurs; rather, symptomatic phenomena appear that indicate the fault and are measurable because, for example, measured variables related to the machine / hardware gradually change. These changed measured variables can generally be detected by sensors, such as an output current, an input current, a rotational speed (e.g., in a fan), etc. Not infrequently, the machine or hardware can continue to operate, sometimes initially even without any noticeable loss of production quality, until the machine then suffers a major failure.

[0010] In this respect, the invention also makes it possible to optimize maintenance intervals, i.e. to predict, for existing machine inventories, when maintenance will be due and how long or short the maintenance intervals should be.

[0011] According to the invention, the difficulty was recognized that the measured sensor data are covered by statistical errors, in particular statistical noise, which makes it difficult to perform an evaluation of individual measured sensor data, since these are subject to statistical errors and can deviate from predetermined target values, even though no error is present.

[0012] It was recognized that, for example, such a sensorily detected quantity does not change on average or only changes slowly, but the statistical fluctuations of this quantity around an expected value increase.

[0013] Furthermore, the invention takes into account for the first time that errors can often affect the acquired sensor data in various ways. Therefore, the invention considers statistical irregularities during evaluation and investigation, initially regardless of their cause, because simple comparisons and recording of individual value deviations have proven insufficient for reliable error detection. The method according to the invention can thus be used universally for a wide variety of applications, as no special requirements regarding the monitored machines are necessary and, as a rule, only the general expected behavior of a sensor-acquired measurement needs to be known, even if, for example, it is only a value that remains constant over time.

[0014] According to the invention, at least three criteria are checked before a warning signal is issued. The basic criterion, which is considered fundamental to error detection according to the invention, is whether the measurement data is constant within a predetermined tolerance range. In order to largely parallelize the processing, at least three criteria are checked, optionally even determined and verified simultaneously, before a warning is issued.

[0015] Since the basic criterion is of central importance and also represents the more general test for other criteria, such as trend analysis or investigations into abrupt changes, and is the fundamental condition for a trend deviation or abrupt deviation, the basic criterion is checked first. Only if this reveals a deviation are further criteria checked. These subsequent criteria are also evaluated in parallel for faster analysis.

[0016] Advantageously, one embodiment of the invention thus also provides a multi-stage, at least three-stage, warning system in which a separate warning level is displayed depending on the number of fulfilled criteria during error detection. In this way, a monitoring method is simultaneously provided that weights the occurring errors, so that the operator is made aware of the severity of the irregularities and can also more effectively consider whether and with what measures intervention should be made in a process or the operation of a machine.

[0017] If, in a first step, the basic criterion is checked and, in a further step, other criteria are checked, the warning system can also be advantageously adapted to this, i.e., issue at least a two-stage warning that takes this classification into account.

[0018] The invention is particularly suitable for the process automation of systems that operate continuously and whose vibration behavior or characteristics are especially relevant to their operation. Examples include the monitoring of vacuum pumps for cleanroom applications or the monitoring of fans that operate 24 / 7. The inventive analysis of the vibration signal advantageously provides a very early indication of incipient damage, allowing the machine or system to be repaired or maintained in a timely manner before major damage and failure occur. Based on several such indications, maintenance intervals can also be adjusted by estimating how frequently such damage occurs during operation and when it is likely to occur.The system can be adapted very quickly to various applications for long-term monitoring. In a preferred advanced configuration, vibration sensors are used to monitor, for example, fans, bearings, pumps, vacuum pumps, or other motors in continuous operation.

[0019] One advantageous application could be in the field of medical technology, where technical and biological processes interact, promoting statistical deviations and making it more difficult to distinguish between error detection requiring intervention and simple statistical anomalies. Another advantageous application lies in the area of ​​cooling production processes (fans, blowers, etc.).

[0020] According to the invention, a statistical evaluation of at least one recorded series or a part thereof is therefore provided. In an advantageous embodiment, at least three criteria are also checked in order to assign corresponding warning levels depending on how many of these criteria are met simultaneously.

[0021] The monitoring method according to the invention is generally implemented as a computer-based method in which the sensor data is evaluated electronically or computer-aided. It thus overcomes the technical prejudice that signals containing statistical errors or noise cannot be statistically evaluated for error and plausibility.

[0022] The statistical evaluation according to the invention is carried out under a null hypothesis. If the null hypothesis is disproven or the probability of disproving the null hypothesis exceeds a predefined significance level α, a warning is issued.

[0023] In particularly preferred embodiments of the invention, at least one of the following three criteria can be checked during the statistical evaluation: A typical error detectable using statistical methods is a jump in the mean sensor data level over time or space. This jump must be distinguished from individual deviations caused by noise and thus purely statistically, without any error in the true sense. Such a jump error can occur, for example, if the measured quantity is subject to a sudden but continuous change, i.e., deviates from steady-state behavior. However, a sensor-related error leading to this level deviation is also conceivable. The mean value, at least when averaged over larger temporal or spatial intervals, therefore changes over time. Another error detectable using statistical methods can occur when the measured sensor data continuously change according to a consistent trend ("drift").In this case, too, there is a deviation from steady-state behavior. The mean value changes continuously over time. Even if the mean remains constant when averaged over larger time or spatial intervals, the susceptibility to fluctuations or volatility can vary greatly, or in particular increase, in the event of an error.

[0024] In one embodiment of the invention, the expected behavior can be described by a model function composed of a sum of a function describing steady-state behavior, a function representing a random walk to model noise, and a linear function describing trend behavior with time / spatial distance, in order to represent typical effects such as background noise or drift of the measured values. The measured values ​​can therefore be described as follows: y t = c t + δt + u t , where ytthe modeled measurement data as a function of time t (or of the place) represent ut describes stationary behavior, ct a random walk, for example to simulate noise and δt a trend, i.e. a "running away" or "drifting" of the measured values.

[0025] This model description allows for a statistical evaluation of expected errors, making them mathematically tangible in terms of potential errors. Statistical quantities such as the expected value, standard deviation, and variance can be calculated using this model description. In the case of steady-state behavior, the expected value can be determined by averaging. The deviations of the measured values ​​from the model function, the residuals, can also be calculated from the model.

[0026] Since a trend in the measured values, especially a continuous drift of the measured values, would represent an error, the model initially... δ Assuming a value of 0. The first criterion is a check to see if a jump occurs suddenly in the mean level of the measurement data, a so-called level shift. Ideally, with steady-state behavior, the variance is 0. If the measurements are subject to noise, the variance lies around a certain expected value (ideally the mean). The volatility of the random walk component ctAccording to a null hypothesis, the expected value is zero and only differs from zero if the overall behavior is non-stationary (even when considering underlying noise). In the case of non-stationary behavior, the expected value essentially changes if it is recalculated over successive time intervals. Statistically speaking, this represents a jump that shifts the measured values ​​so significantly that any deviation from the expected value at a given variance due to underlying noise can no longer be explained by the noise behavior.

[0027] For the null hypothesis, a stationary KPSS test can be used (KPSS: Kwiatkowski, Phillips, Schmidt and Shin): For a number T of measurements, a test statistic for the null hypothesis can be calculated. ∑ t = 1 T S t 2 s 2 T 2 assumed, whereby S t the sum of the residuals et S t = e 1 + e 2 + e 3 + … + e t At individual measurement points t (usually at specific time points t) relative to the model function or regression curve, s2 represents the variance relative to the expected value. When checking the first criterion, whether a level shift exists, it is generally advantageous to use a relatively small amount of measurement data to calculate the sum in the test function. This is because, when comparing the results, it is not necessary for the point at which a potential jump in the data occurs to lie within the time or spatial interval being tested. This test statistic follows a certain distribution. To test the null hypothesis, the probability of the given test statistic can be examined. Below a certain probability or significance level, it can be assumed that the null hypothesis is false. The significance level α, which should not be exceeded, can be, for example,It should be assumed to be 1%, otherwise the deviation from the test statistics is too great and a warning must be issued.

[0028] If the volatility of the model function = 0, then the behavior is stationary, possibly with some underlying noise.

[0029] To verify the second criterion, whether a trend exists or whether the measurement data continuously "drift away," it is generally necessary to evaluate and compare more measurement points. Alternatively, the KPSS test can also be applied for this criterion. The model function still requires the following conditions: δ = 0, because otherwise a trend behavior would have to be assumed, and autocorrelations are neglected.

[0030] If the significance is less than a certain, predetermined value and a large number of measurement data were used, a trend can be assumed; the second criterion is met.

[0031] The verification of the first and second criteria can only differ in the number of test data points or the measured time interval. The verification checks whether: the volatility of the random walk term is either zero or greater in absolute value than the absolute value of a given value, and / or the test statistic of the stationary KPSS test is met or not met with a certain predetermined probability.

[0032] To verify the third criterion, whether the range of fluctuations or susceptibility to fluctuations increases steadily, it suffices to compare the variances in two different test phases or measurement ranges. The null hypothesis to be tested is that the variances always match or at least differ only with a predetermined probability. If the range of fluctuations increases, the variance in a subsequent time or measurement interval also increases. The amount of test data can be smaller than when verifying the second criterion regarding the trend.

[0033] Statistically speaking, the significance can thus be determined in relation to a fixed probability value, whereby if the significance falls below this probability value, the range of variation increases and the third criterion is met. Examples of implementation

[0034] Exemplary embodiments of the invention are shown in the drawings and are explained in more detail below, including further details and advantages. Specifically, the drawings show: Fig. 1: a schematic representation of a monitoring method for fault detection and fault prediction according to the invention, Figs. 2-4: various measurement data series superimposed with noise to illustrate fault detection, Fig. 5: a measurement data series over a longer period, which is subject to different faults, including a representation of the fault terms, Fig. 6: a schematic representation of a system used for machine monitoring of several vacuum pumps, Fig. 7: a schematic representation of the parallel evaluation of three criteria before output of the warning signal, and Fig. 8: a schematic representation of the staged evaluation before output of the warning signal.

[0035] Figure 1Figure 1 shows a schematic representation of a computer-implemented monitoring method 1 for fault detection and fault prediction according to the invention. A system 2 is monitored by means of a sensor 3, which sends the measured data to a computer-controlled evaluation unit 4. This evaluation unit 4 can be a control unit or controller, for example, within the machine, or it can run separately on a computer.

[0036] Evaluation unit 4 statistically analyzes the measured data: It first specifies a statistical significance level, e.g., 1%, and then at least one criterion. Here, evaluation unit 4 checks the basic criterion as well as three further criteria I, II, and III to determine: According to the basic criterion, is there any error-causing deviation because the measurement data are not constant within a predefined tolerance? First criterion I: Is there a level shift in the measurement data? Second criterion II: Is there a trend? Third criterion III: Is there a change in volatility or susceptibility to fluctuations?

[0037] Whether one of the criteria is met is determined using a so-called null hypothesis.

[0038] A model function is used as a basis, which maps statistical errors, for example via a random walk, but also has a term that can describe the basically unwanted trend behavior.

[0039] The trend behavior is initially considered non-existent, i.e., the term describing the trend is set to zero.

[0040] To determine the increase in volatility III, it suffices to examine the variances of two (temporally successive) sensor-detected subsequences. Their ratio would ideally be 1. Test statistics can also be used to assess the behavior and verify whether the measured behavior closely matches the test statistic.

[0041] In the case of level-shift behavior (criterion I), the expected value changes. Here, too, a test statistic can be used. Fewer measurement data points are needed because the level shift is expected to occur rather quickly and therefore should not be averaged over a longer period in order to reliably capture the signal change.

[0042] Trend behavior, which can also be very slow, requires measurement over a longer period of time to determine the change.

[0043] The evaluation of the significance and the determination of the deviation from the respective null hypothesis can be carried out in execution block 4a.

[0044] If the measurement data is not sufficiently constant within a tolerance limit (basic criterion K) and one of criteria I, II, or III is met, the evaluation unit 4 issues a first warning level WI. Similarly, a second, more critical warning level WII is signaled if two criteria are met, and a particularly high warning level WIII is signaled if three criteria are met. Figure 1 The time course of the measurement data is displayed, showing the ranges WI, WII, and WIII, in which one to three criteria are met. It is also conceivable that a measure of the deviation from the respective null hypothesis is determined and the warning levels WI, WII, or WIII are selected accordingly.

[0045] Warning levels WI, WII, and WIII can either be displayed, for example via a traffic light or signal tower 6, or, depending on the design, the evaluation unit 4 can send corresponding control commands to the system 2. These commands can also be adapted to the types of errors that typically occur with certain criteria I, II, and III.

[0046] Typically, it must be assumed that recorded measurement data from a measurement data series is superimposed with noise, as is the case in the Figures 2-4This is always the case. Due to this superimposed noise, the measurement data in individual sections of the measurement series are so widely distributed that identifying errors such as a trend or a level shift becomes difficult. In particular, the susceptibility to fluctuations or volatility is more difficult to discern graphically when the noise-induced variance of the measurement data is comparatively high and the measured values ​​are equally likely to vary to higher and lower values ​​compared to a mean or expected value.

[0047] Fig. 2The graph shows a signal that is clearly obscured by noise, recorded as a series of measurements. On average, however, the measurements remain at a constant level, represented by a dotted line. Despite the high variance of the measurements, this indicates constant volatility; the variance does not increase over time, but rather the measurements remain essentially constant around the mean / expected value during this period. Over the entire period, apart from statistical outliers in the measurements, there are no signal fluctuations on average. According to an evaluation based on the third criterion (III), constant volatility would be observed.

[0048] Fig. 3This also shows a signal superimposed with noise. However, over time, the measured values ​​increase continuously, indicating a trend. The dotted line, which represents the mean values ​​of the measured data, is a straight line with a positive slope and shows the increasing level. However, since the statistical fluctuations or variance of the data are also high here, this trend is difficult to recognize without evaluating a mean or expected value. This trend is identified according to the second criterion, II.

[0049] Fig. 4 shows an analogous situation regarding the Fig. 2-3This means that a noisy signal is present, and an evaluation of the mean or expected value reveals a sudden jump in the measured values ​​(level shift) to a higher level. The mean or expected value is represented by a dotted line that exhibits this sudden jump (level shift). Before and after this jump event, the measured data remain constant on average within these time intervals. Therefore, an evaluation according to the first criterion (I) is relevant in this case.

[0050] The first row shows the progression of a measurement data series over a longer period of 19 days. Figure 5 The first row represents the value. The following rows describe, on a scale from 0 to 1, the terms in the model function that describe the trend (criterion II), the level shift (criterion I), and the volatility (criterion III).

[0051] The measurement data trend ("Value") shows that the measured values ​​are superimposed with statistical noise throughout the entire period. This noise also fluctuates in intensity; for example, the variance is significantly higher around July 4th than between June 17th and approximately June 20th. On average, however, the values ​​are constant between June 17th and approximately June 20th, and the fluctuations are statistically significant to a lesser extent, as can be seen in the last line of Figure 5 This shows that the volatility reaches a value of 0 until June 20th. Similarly, the trend line in the second row remains constant at 0 until June 20th, as the average signal level does not change.

[0052] The behavior of the measurement data from June 30th to July 5th is different. Here, the deviations continuously increase, while the average signal level remains constant. The variance grows. This increasing deviations raise the volatility value from June 30th onwards. The constant average signal level is reflected in a trend value close to 0 during this period.

[0053] At two discrete points, namely around June 22nd and June 30th, the signal level changes abruptly, which is reflected in the third row by an isolated peak at precisely these times. Trend and volatility also show changes at these points, also as singular peaks in the signal curve, because the signal curve (trend) also changes with the jump, or rather, a deviation occurs that exceeds the variance.

[0054] Between June 22nd / 23rd and June 30th, the measurement signal increases continuously and rises slightly. Therefore, during this period, line 2 of the Figure 5 The trend value is in the range of 1 because there is an upward trend.

[0055] The third line of Fig. 5 This shows the level shift term, where, as described above, the peaks around June 22nd and June 30th are particularly noticeable. Outside of these two peaks, the level term remains close to 0, but shows irregular deviations. This is because the measured signal ("value") is not always statistically constant on average, depending on the range under consideration; that is, the signal level, which is inherently subject to noise, changes slightly.

[0056] Furthermore, higher statistical fluctuations and / or changes in the level occurring in between, such as around June 22nd and June 30th, can lead to the value in the volatility term changing briefly and being non-zero, because the signal undergoes a change in the form of a jump, which is also larger than the variance caused by noise, i.e. the susceptibility of the signal to fluctuations has also changed here.

[0057] Figure 6Figure 1 shows a schematic of a system for monitoring vacuum pumps with vibration sensors and diagnostic electronics for process monitoring. The vibration sensors used are of the manufacturer "ifm VSE". The signals are sent to a processor for evaluation. Several clock cycles of the clocked process are combined into a machine cycle. Continuous pattern monitoring, supported by numerical evaluation software, analyzes the measurement data to examine volatility, level shifts, and trends. Mean and expected values ​​are also calculated. Using probability methods, the system extrapolates from recent measurements to predict future data development or checks whether the measurements lie within or outside a defined confidence band with a certain probability.

[0058] Selective parameterization can be applied, which provides for a more detailed analysis according to specific criteria or parameters (detailed parameterization options), e.g.: Selection of reference data from the data set and / or configuration of objects in vibration analysis and / or analysis using narrow confidence bands to detect even small deviations or stability of measurement data.

[0059] The user typically has two options, depending on the specific implementation of the invention, for training the Pattern Monitor: Option A: "Supervised": The user manually marks patterns present in historical data, for example, a trend that existed in a specific period (or analogous to volatility or a jump). The algorithm is then parameterized so that it would have recognized precisely this period and not all other periods, if possible with the testing procedures. This also checks whether the user has missed any markings or marked a period that shows no pattern at all. Option B: "Unsupervised": The user uploads a period that is stationary and contains none of the patterns. The algorithm is then trained so that it would not have issued an alarm / warning signal in the selected period, but is parameterized as sensitively as possible (e.g., a narrow confidence band), meaning it would trigger an alert at the slightest changes / patterns.This process also checks whether there are indeed patterns in the data that the user simply forgot to mark.

[0060] The Figures 7 and 8 Two variants for determining the criteria are shown. Fig. 7 Three criteria are determined in parallel: the basic criterion K, the analysis of a trend II, and the occurrence of jumps I. Depending on which criterion is relevant, one of the warning messages (or possibly several warning messages) WK, WI, or WII is issued.

[0061] Fig.8 This demonstrates a staged determination of criteria: First, the basic criterion K is checked; then, if a deviation is detected, the other two criteria I and II are checked in parallel. A two-stage warning signal WI, WII is sufficient as output, depending on whether the non-constant measurement data shows jumps or trends.

[0062] Advantageously, the monitoring method according to the invention enables a statistical evaluation of measurement data, which is partly subject to statistical errors, such as noise, and allows for the detection of any errors. The method according to the invention primarily utilizes statistical methods and is largely universal, independent of the machine / hardware type being monitored, because it essentially distinguishes between the occurrence of statistical errors and errors caused by actual malfunctions. According to the invention, this is achieved by determining deviations from a null hypothesis and evaluating their statistical significance relative to a predetermined significance level.The classification into different categories based on the number of fulfilled error criteria and / or the degree of deviation allows for the assessment of the seriousness of the situation during further training. Reference symbol list

[0063] 1 Monitoring procedure 2 System to be monitored 3 Sensor 4 Evaluation unit 4a Significance calculation with respect to null hypothesis 5 Measurement data with error categorization 6 Signal tower Criterion I (Volatility) Criterion II (Level shift) Criterion III (Trend behavior) First warning level WII Second warning level WIII Third warning level WK Base warning level α Significance level K Base criterion

Claims

1. Computer-implemented monitoring method (1) for vibration analysis for fault detection of a machine fault and / or hardware fault in process automation in a production process operated and / or assisted by machine, comprising: • acquiring at least one series of sensor data (5) measured in the process in a chronological and / or locational sequence, • wherein a statistical evaluation (4) of the at least one series or at least a part of the at least one series is performed in order to determine in each case a deviation of the measured sensor data (5) according to at least three criteria (K, I, II, III), wherein the statistical evaluation: • provides a null hypothesis, • provides an expected value for the measured sensor data, • provides the calculation of the statistical significance with respect to the deviation from the predetermined null hypothesis, and • outputs a warning if the statistical significance deviates from the null hypothesis and / or if at a predetermined significance level α the probability of a deviation from the null hypothesis is greater than the significance level α, characterized in that • in the statistical evaluation (4), the deviations of the measured sensor data are determined according to the following criteria (K, I, II; III) before the warning is output, wherein: i. in the statistical evaluation (4) for testing a basic criterion (K) it is checked whether the mean sensor data level in chronological and / or locational sequence remains constant within a tolerance predetermined by the basic criterion (K) and ii. in the statistical evaluation (4) to test the second of the criteria (II) it is checked whether a trend of a continuous change of the mean sensor data level is present in chronological and / or locational sequence, and / or iii. in the statistical evaluation (4) to test a first of the criteria (I), it is checked whether a jump of the mean sensor data level is present in chronological and / or locational sequence, wherein the jump has taken place when the jump shifts the measured values so strongly as a whole that deviations from the expected value with a specific variance as a result of underlying noise is no longer to be explained by the noise behaviour and / or iv. in the statistical evaluation to test a third of the criteria (III), it is checked whether the susceptibility to variation in the series changes in chronological and / or locational sequence, wherein v. in the statistical evaluation (4), the basic criterion (K) is always checked, wherein the statistical evaluation (4) of the deviations of the measured sensor data according to the basic criterion (K) is performed first, and then simultaneously or chronologically, a determination is made overlapping according to two of the further criteria (I, II, III) overlapping.

2. Computer-implemented monitoring method (1) for vibration analysis according to Claim 1, characterized in that the statistical evaluation provides the output of an at least three-level, in particular four-level warning (WK, WI, WII, WIII), wherein each number of criteria met of the criteria (K, I, II, III) is assigned a warning level (WK, WI, WII, WIII) in each case.

3. Computer-implemented monitoring method (1) for vibration analysis according to Claim 1, characterized in that the statistical evaluation provides the output of an at least two-level warning, depending on whether, in addition to the basic criterion, the first or second criterion also reveals a deviation.

4. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that the following function is used as the model function for the behaviour of the sensor data (5) in the series: y t = c t + δ t + u t , wherein yt represents the modelled measured data as a function of t, ct describes a random walk, δ describes a trend and δt describes a linear trend behaviour, and ut describes a stationary behaviour, when the sensor data of the series are plotted over t, wherein t describes the chronological and / or locational course.

5. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, to check the first criterion (I), the null hypothesis is used that the variance as the mean square deviation of the expected value and volatility of the random walk term ct is zero, wherein it is assumed in particular with respect to the model function that the trend δ = 0.

6. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, to check the first and / or second criterion (I), a stationary KPSS test is carried out, wherein it is checked as the null hypothesis whether a test statistic is met with a predetermined probability, wherein in particular the following is used as the test statistic: ∑ t = 1 T S t 2 s 2 T 2 as the sum of the squares of the sum St over the part T of the sequence, wherein St is in turn the sum of the residues in relation to the regression curve of the measured sensor data at individual points t, in relation to the product of variance s2 multiplied by the square of the number of the sequence part T, wherein a warning is output if the test statistic is not met with less than a predetermined probability.

7. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, to check the first and / or the second criterion and / or the basic criterion (I, II, K), autocorrelations are used, which are taken into consideration and / or calculated out and / or neglected in particular in signal delays in the form of sensor data influenced by delay.

8. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, to check the second criterion (II), a larger part of the sequence having more sensor data is used than to check the first criterion (I).

9. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, to check the third criterion (III), at least two partial sequences of sensor data are recorded.

10. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, to check the third criterion (III), a test statistic F is used as the null hypothesis, which forms the ratio of the variances of the two test sequences, and it is assumed as the null hypothesis that F =1.

11. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that the warnings are transmitted at a signal column, wherein: • each warning level is assigned a different light sign on the signal column as a function of the number of the warnings and / or • a different light sign on the signal column is assigned depending on the deviation from the null hypothesis.

12. Computer-implemented monitoring method for vibration analysis (1) according to any one of the preceding claims, characterized in that, upon the acquisition, the data of a period of time are loaded which is stationary and in particular does not contain any of the patterns, wherein subsequently the algorithm is trained so that it would not have output an alarm and / or warning signal in the selected period of time, but at the same time is parameterized as sensitively as possible, for example, by a narrow confidence band, thus would deflect in the event of correspondingly small changes and / or patterns.