Storage device, and associated machine and process
By using the Hjorth complexity parameter to analyze vibration measurements and set a detection threshold, the method effectively predicts bearing failures in warehouse devices, overcoming the limitations of existing technologies.
Patent Information
- Application Number
- DE102023211019
- Authority / Receiving Office
- DE · DE
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-11-07
- Publication Date
- 2025-05-08
AI Technical Summary
Existing methods for detecting failures in warehouse devices, particularly in the vibration analysis of bearings, are prone to false alarms due to sensitivity to speed variations and fail to recognize certain defects.
A procedure that involves determining the Hjorth complexity parameter from vibration measurements during a training period, modeling these values with a normal distribution, setting a detection threshold, and comparing subsequent measurements to this threshold to accurately predict bearing failures.
This approach allows for precise prediction of bearing damage by being insensitive to rotation speed fluctuations and enables the detection of various types of defects, enhancing diagnostic safety and the accuracy of warehouse monitoring.
Smart Images
Figure 00000000_0000_ABST
Abstract
Description
[0001] The present invention relates to a bearing device and a method for monitoring a bearing device in a machine.
[0002] In particular, the invention is concerned with the detection of a failure of a bearing of the bearing device.
[0003] A machine, e.g. a truck, comprises several axles.
[0004] Each axle is mounted in bearings.
[0005] The bearings are monitored to detect bearing failure.
[0006] Generally, each bearing is equipped with a vibration sensor that provides a signal representative of the bearing's vibrations.
[0007] A spectral analysis is performed on the signal, using, for example, fast Fourier analysis, to obtain a spectrum to identify harmonics representative of bearing defects, such as a defect in the raceway of the bearing's inner ring.
[0008] However, some bearing defects are not detected from the spectrum.
[0009] It is also known to determine the root mean square of the set of vibration measurements comprising the signal.
[0010] However, the quality of the root mean square from the set of vibration measurements is sensitive to bearing speed variations, which can lead to false alarms.
[0011] Consequently, the present invention aims to improve the detection of bearing failure using vibration measurements.
[0012] According to one aspect, a method for determining a method for monitoring a bearing device in a machine is provided, wherein the bearing device comprises a bearing provided with an inner ring and an outer ring adapted to rotate concentrically with respect to each other, and a vibration sensor measuring vibrations of the bearing.
[0013] The procedure includes the following steps: a. Determining values of at least one first statistical parameter corresponding to the Hjorth complexity parameter from at least one set of vibration measurements provided by the sensor during a training period following the implementation of the sensor in the machine and when the rotational speed of the inner ring or the outer ring relative to the other is included in a predetermined interval, b. Modelling the values of at least the first statistical parameter with a normal distribution during the training period, c. Determining a detection threshold at least from the normal distribution during the training period, d. Determining at least a first value of the first statistical parameter from a set of vibration measurements provided by the sensor during normal operation of the machine and when the rotational speed of the one ring is included in a predetermined interval, e. Determining an evaluation during normal operation of the machine from at least the first value of the Hjorth complexity parameter and the normal distribution modelling the values of the first statistical parameter, f. comparing the value with the detection threshold, and g. Determination of the bearing failure according to the result of the comparison.
[0014] Since the determination of the Hjorth complexity parameter is insensitive to rotation speed variations of the bearing, the first statistical parameter, which is equal to the Hjorth complexity parameter of the set of vibration measurements, allows an accurate prediction of the bearing damage by analyzing the evolution of only the first statistical parameter.
[0015] Preferably, - step (a) further comprises determining values of at least one second statistical parameter from the set of vibration measurements provided by the sensor during the training period, - step (b) further comprises modelling the values of the second statistical parameter with a normal distribution during the training period, - step (c) comprises determining the detection threshold based on the normal distribution modelling the values of the first statistical parameter and the normal distribution modelling the values of the second statistical parameter during the training period, - step (d) further comprises determining a first value of the second statistical parameter from the set of vibration measurements provided by the sensor during normal operation, and - step (e) comprises determining the result during normal operation of the machine from the first value of the Hjorth complexity parameter, the first value of the second statistical parameter, the normal distribution modelling the values of the first statistical parameter and the normal distribution modelling the values of the second statistical parameter.
[0016] In order to increase the diagnostic reliability, as shown above, the early prediction of damage of the bearing 4 can be achieved by determining several statistical parameters from recorded vibration measurements of the bearing 4 by analyzing the evolution of these parameters, where one of the several statistical parameters is equal to the Hjorth complexity parameter.
[0017] Since several statistical parameters are used to predict bearing failure, different types of bearing defects can be detected.
[0018] In addition, the detection threshold is determined based on vibration measurements on the bearing in the machine, which allows the threshold to be defined depending on the application of the machine in order to obtain accurate monitoring of the bearing.
[0019] Advantageously, the values of each statistical parameter are determined during the training period and during normal operation of the machine when the kurtosis of the set of vibration measurements is less than a predetermined kurtosis threshold if the machine is a mobile machine.
[0020] Preferably, steps (d), (e), (f), (g) are repeated for a predetermined period of time.
[0021] Advantageously, the determination of a detection threshold comprises summing the values of each normal distribution contained in a first interval having a lower limit equal to one minus a predetermined first quantile, the detection threshold being equal to the sum.
[0022] Preferably, the determination of a score for each statistical parameter comprises - performing a Z-score normalization of the first value of each statistical parameter from the mean and variance of the normal distribution that models the values of the statistical parameter to obtain a standardized value of the statistical parameter, and - Summing each standardized value encompassing a second interval with a lower limit equal to one minus a predetermined second quantile, the sum being equal to the score.
[0023] The comparison of the score with the detection threshold advantageously comprises the comparison of the score with the detection threshold less the number of standardized values contained in the second interval.
[0024] Preferably, the bearing is considered defective if the value is greater than the detection threshold.
[0025] The second statistical parameter advantageously includes: - the root mean square of the set of vibration measurements, or - the sum of the areas of equal squares, each square being defined by a diagonal connecting two vibration measurements of the set of vibration measurements, or - the entropy of scattering of the set of measurements, or - the permutation entropy of the set of measurements, or - the mean value of a predetermined number of the strongest harmonics of a spectrum of the series of measurements, where the first and second statistical parameters are different.
[0026] Preferably, the method further comprises defining a third statistical parameter equal to the square of the sum of the first and second statistical parameters.
[0027] According to one aspect, a bearing device is proposed which comprises a bearing provided with an inner ring and an outer ring which are adapted to rotate concentrically with respect to each other, and a vibration sensor which measures vibrations of the bearing.
[0028] The storage device also includes: - first determining means for determining values of at least a first statistical parameter equal to the Hjorth complexity parameter from vibration measurements provided by the sensor when the rotational speed of one of the inner and outer rings relative to the other ring is comprised within a predetermined interval, - means for modelling the values of at least the first statistical parameter with a normal distribution during a training period after the implementation of the sensor in the machine, - second determination means for determining a detection threshold from at least the normal distribution during the training period, - third determining means for determining a value during normal operation of the machine from at least a first value of the Hjorth complexity parameter and the first normal distribution modelling the values of the first statistical parameter, and - a comparison device for comparing the value with the detection threshold and for determining the failure of the bearing according to the result of the comparison.
[0029] According to a further aspect, a machine is proposed which comprises a storage device according to the above definition.
[0030] Further advantages and features of the invention will become apparent from the detailed, in no way limiting, description of embodiments and the accompanying drawings, in which: Fig. Figure 1 schematically represents a rotating machine according to the invention; Fig. Figure 2 schematically illustrates an example of a monitoring device for monitoring a storage device according to the invention; Fig. Figure 3 schematically illustrates an example of a statistical parameter according to the invention; Fig. 4 schematically shows an example of a further statistical parameter according to the invention, Fig. 5 schematically shows an algorithm for calculating another statistical parameter; Fig. 6 and Fig. 7 shows an example of a method for monitoring the storage device according to the invention.
[0031] It will be Fig. 1, which schematically shows a partial longitudinal section of a machine 1.
[0032] The machine 1 comprises a housing 2 and a shaft 3, which is mounted in the housing 2 via a roller bearing 4.
[0033] The machine 1 may be a mobile machine, e.g. a truck, wherein the shaft 3 is an axle of the truck supported by the roller bearing 4.
[0034] In one variant, machine 1 can be a stationary machine, e.g. a machine tool.
[0035] The roller bearing 4 is provided with an inner ring 5, which is fixed to the shaft 3, and an outer ring 6, which is mounted in the bore of the housing 2. The outer ring 6 radially surrounds the inner ring 5. The inner and outer rings 5, 6 rotate concentrically with each other.
[0036] The roller bearing 4 is further provided with a row of rolling elements 7, which are radially inserted between the inner and outer raceways of the inner and outer rings 5, 6. In the example shown, the rolling elements 7 are balls. Alternatively, the roller bearing can also comprise other types of rolling elements 7, for example, rollers. In the example shown, the roller bearing comprises a row of rolling elements 7. Alternatively, the roller bearing 4 can also comprise multiple rows of rolling elements.
[0037] A sensor 8 is mounted in the housing 2 to measure the rotation speed of the bearing 4.
[0038] The sensor 8 can be attached to a hole in the housing 2.
[0039] In one variant, the sensor 8 can be attached to the outer ring 6 or to the outer ring 6 and to the bore of the housing 2.
[0040] The sensor 8 delivers a signal representative of the vibrations of the bearing 4 to a device 9.
[0041] In one variant, the sensor 8 is located in the monitoring device 9.
[0042] The bearing 4 and the monitoring device 9 form a bearing device.
[0043] Fig. 2 shows an example of the monitoring device 9.
[0044] The monitoring device 9 comprises first determination means 10 for determining, from the vibration measurements provided by the sensor 8, values of at least one statistical parameter P1 corresponding to the complexity parameter Hjorth.
[0045] A set of vibration measurements comprises a predetermined number of samples x i of the signal supplied by sensor 8, where i is an integer between one and an integer M, where M corresponds, for example, to eight thousand samples.
[0046] The monitoring device 9 further comprises modeling means 11, second determination means 12, third determination means 13, comparison means 14, a memory 15 and implementation means 16.
[0047] In one variant, the memory is located outside the monitoring device 9.
[0048] The first determination means 10 implement a first algorithm ALGO1 which determines the first statistical parameter P1 equal to the Hjorth complexity parameter of the set of vibration measurements such that: P1=∑i=1MVar(d2(xi))Var(d(xi))1Mobility where Var(X) is the variance of the variable X, d(X) is the first derivative of the variable X, d2(X) is the second derivative of the variable X and the mobility is equal to: Mobility=Var(d(xi))Var(xi)
[0049] The first determining means 10 can implement further algorithms to determine further statistical parameters.
[0050] It is assumed that the first determining means 10 further implements a second algorithm ALGO2, a third algorithm ALGO3, a fourth algorithm ALGO4, a fifth algorithm ALGO5, a sixth algorithm ALGO6, a seventh algorithm ALGO7 and an eighth algorithm ALGO8.
[0051] The second algorithm ALGO2 determines a second statistical parameter P2, the third algorithm ALGO3 a third statistical parameter P3, the fourth algorithm ALGO4 a fourth statistical parameter P4, the fifth algorithm ALGO5 a fifth statistical parameter P5, the sixth algorithm ALGO6 a sixth statistical parameter P6, the seventh algorithm ALGO7 a seventh statistical parameter P7 and the eighth algorithm ALGO8 an eighth statistical parameter P8 from a set of vibration measurements provided by the sensor 8.
[0052] The seven statistical parameters P1, P2, P3, P4, P5, P6, P7 are different from each other.
[0053] For example, the second algorithm ALGO2 determines the second statistical parameter P2, which is equal to the root mean square of the set of vibration measurements, such that: P2=1N∑i=1Nxi2
[0054] For example, the third algorithm ALGO3 and a fourth algorithm ALGO4 determine the third statistical parameter P3 and the fourth statistical parameter P4, respectively, from the sum of the areas of equal squares.
[0055] Each square is defined by a diagonal connecting two vibration measurements of the vibration measurement set.
[0056] For example, the third statistical parameter P3 is equal to the sum of the identical areas, where each area A1 is defined by a diagonal D1, which represents a sample S1 of the measurement set and a third sample S3 of the measurement set following the sample S1, as in Fig. 3, where ti, i varies between 1 and M, is the sampling time, so that: P3=∑j=1M−3(tj+3−tj)(Si+3−Si)
[0057] The fourth statistical parameter P4 is, for example, equal to the sum of identical areas, where each area A11, A12, A13, A14 is defined by a diagonal D11, D12, D13, D14 connecting a sample S1, S2, S3, S4 of the series of measurements and the next sample S2, S3, S4, S5 of the series of measurements following said sample S1, S2, S3, S4, as in Fig. 4, so that: P4=∑j=1M−1(tj+1−tj)(Si+1−Si)
[0058] A fifth algorithm ALGO5 determines the fifth statistical parameter P5, which corresponds, for example, to the dispersion entropy of the measurement series.
[0059] The document entitled "Dispersion Entropy: A measure for time series analysis", M. Rostaghi and H. Azami, IEEE Signal processing letters, vol. 23, n. 5, pp. 610-614, 2016, discloses an algorithm for determining the dispersion entropy value DE of a univariate signal X of length N, X = {X1, X2, ...,XN}, where N is an integer.
[0060] For example, the univariate signal X corresponds to the set of vibration measurements that contains the given number M of samples x i includes.
[0061] The algorithm includes four main steps, which are Fig. 5 are shown.
[0062] During a step 17, the samples Xj of the univariate signal X, j varying from 1 to N, are mapped to c classes labeled 1 to c.
[0063] The cumulative normal distribution function (NCDF) is used to map the univariate signal X into a signal Y={Y1, Y2,...,YN}.
[0064] For each sample Yj of the signal Y, where j is an integer from 1 to N, a value Zjc defined, where: Zjc=round(c⋅Yj+0.5) where the round() operator either increments or decrements a number to the nearest integer.
[0065] During step 18, embedding vectors zkm,c with m as the embedding dimension, so that each embedding vector zkm,c corresponds to a time series. zkm,c={zjc,zj+dc,...zj+(m−1)dc}
[0066] Here, d is a time delay and k varies between 1 and N-(m-1)d.
[0067] Then each time series zkm,c on a dispersion pattern π v0v1 ...v m-1shown, where zkc=v0,zk+dc=1,...zk+(m−1)dc=vm−1.
[0068] The number of possible dispersion patterns associated with each time series zkm,c can be assigned is equal to c m , since the signal has m elements and each element can be one of the integers from 1 to c.
[0069] During step 19, for each of the c m potential propagation patterns the relative frequency is as follows: p(πv0v1−vm−1)=Number{iIi≤N−(m−1)d,zkm,c has type πv0v1…vm−1}N−(m−1)d where the Number() operator converts a value into a number. p(πv0v1…vm−1) gives the number of dispersion patterns π v0v1 ...v m-1 to the zkm,c divided by the total number of embedding signals with embedding dimension m.
[0070] During step 20, the dispersion entropy value DE is determined: DE(X,m,c,d)=−∑πcmp(πv0v1…vm−1)ln(p(πv0v1…vm−1))
[0071] The sixth algorithm ALGO6 determines the sixth statistical parameter P6, which corresponds, for example, to the permutation entropy of the measurement series.
[0072] The document entitled “Efficiently measuring complexity on the basis of real-world data”, VA Unakafova, K. Keller, Entropy, 15(10), 4392-4415.
[0073] Permutation entropy provides a quantification measure for the complexity of the vibration measurement by capturing the order relationships between the values of the vibration measurements and extracting a probability distribution of the ordinal patterns.
[0074] The seventh algorithm ALGO7 determines the seventh statistical parameter P7, which corresponds, for example, to the mean value of a given number of the strongest harmonics of a spectrum of the measurement series.
[0075] For example, a Fast Fourier algorithm is implemented to determine the spectrum of the measurement series, and the mean is determined from the hundred strongest harmonics of the spectrum.
[0076] The eighth parameter P8 is determined from the eighth algorithm ALGO8, which is implemented by the first determining means 10.
[0077] For example, the eighth parameter P8 is equal to the square of the sum of the statistical parameters: P8=∑k=16Pk2
[0078] An example of a method for implementing the monitoring device 9 is presented below.
[0079] The method comprises a first part for determining a detection threshold Sd during a training period after the implementation of the sensor 8 in the machine 1, wherein the detection threshold Sd is used to determine a failure of the bearing 4, and a second part for determining a failure of the bearing 4 during normal operation of the machine 1 after the training period.
[0080] Fig. Figure 6 shows an example of the first part of the procedure.
[0081] It is assumed that one ring of the inner and outer rings 5, 6 rotates relative to the other ring of the inner and outer rings 6, 5.
[0082] During a step 21, the first part of the method is started when a predetermined duration between two consecutive method executions is reached.
[0083] The predetermined duration is, for example, equal to 24 hours, so that the training period is performed once a day during, for example, one month after the implementation of the sensor 8 in the machine 1.
[0084] While the first part of the method is initialized (step 21), during a step 22, when the rotational speed of one ring is comprised in a predetermined interval, in a step 23 a group of vibration measurements comprising xi samples of the signal supplied by the sensor 8 is stored, for example, in the memory 15, where i varies between one and the integer M.
[0085] For example, M is equal to 8000, so the set of vibration measurements includes 8000 samples xi.
[0086] The predetermined interval is defined by a lower limit and an upper limit.
[0087] The lower limit value and the upper limit value are determined according to the type of machine 1 (mobile or stationary) and the application of machine 1.
[0088] For example, for a mobile machine, the lower limit is 50 miles per hour and the upper limit is infinity.
[0089] If the rotation speed of one ring is less than the lower limit, the first part of the process goes back to step 21 and waits for the next start of the first part of the process during the training period.
[0090] When the xi samples are stored in the memory 15 (step 24), the implementation means 16 determine, during a step 25, the value of the kurtosis of the set of vibration measurements when the machine 1 is a mobile machine.
[0091] If machine 1 is a stationary machine, the first part of the method proceeds from step 24 to step 26.
[0092] If the determined kurtosis value is greater than a predefined kurtosis threshold, the first part of the procedure returns to step 21.
[0093] If the determined value of the kurtosis is less than the predetermined kurtosis threshold, the first determining means 10 implements, during step 26, the eight algorithms ALGO1, ALGO2, ALGO3, ALGO4, ALGO5, ALGO6, ALGO7, ALGO8 to determine the value of each of the eight statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 from the samples xi.
[0094] For example, the eight values of the eight statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 are stored in memory 15.
[0095] The kurtosis threshold, for example, is equal to six.
[0096] Steps 22, 23, 24, 25, 26 are repeated until a predetermined number of values of each statistical parameter P1, P2, P3, P4, P5, P6, P7, P8 are stored in the memory 15, for example until forty values of each statistical parameter P1, P2, P3, P4, P5, P6, P7, P8 are stored in the memory 15.
[0097] As long as the predetermined number of values of each statistical parameter P1, P2, P3, P4, P5, P6, P7, P8 is not stored in the memory 15 (step 27), the first part of the method returns to step 21.
[0098] When the predetermined number of values of each statistical parameter P1, P2, P3, P4, P5, P6, P7, P8 is stored in the memory 15 (step 27), the modeling means 11, during a step 28, model the stored values of each statistical parameter P1, P2, P3, P4, P5, P6, P7, P8 with a normal distribution defined by the mean of the values of said statistical parameter and the variance of the values of said statistical parameter.
[0099] The values of the first statistical parameter P1 are modeled by a first normal distribution, which has a mean µ1 and a variance σ1 2 has.
[0100] The values of the second statistical parameter P2 are modeled by a second normal distribution having a mean µ2 and a variance σ2 2 has.
[0101] The values of the third statistical parameter P3 are modeled by a third normal distribution, which has a mean µ3 and a variance σ3 2 has.
[0102] The values of the fourth statistical parameter P4 are modeled by a fourth normal distribution, which has a mean µ4 and a variance σ4 2 has.
[0103] The values of the fifth statistical parameter P5 are modeled by a fifth normal distribution, which has a mean µ5 and a variance σ5 2 has.
[0104] The values of the sixth statistical parameter P6 are modeled by a sixth normal distribution, which has a mean µ6 and a variance σ6 2 has.
[0105] The values of the seventh statistical parameter P7 are modeled by a seventh normal distribution, which has a mean µ7 and a variance σ7 2 has.
[0106] The values of the eighth statistical parameter P8 are modeled by an eighth normal distribution, which has a mean µ8 and a variance σ8 2 has.
[0107] During a step 29, the second determining means 12 determine the detection threshold Sd by summing the values of the first to seventh distributions contained in a first interval having a lower limit equal to one minus a predetermined first quantile.
[0108] The predetermined first quantile is chosen according to the sensitivity of the bearing failure detection.
[0109] The lower the predetermined first quantile, the greater the sensitivity of the method for detecting failures.
[0110] For example, the predetermined first quantile is equal to 80%.
[0111] Fig.Figure 7 shows an example of the second part of the method during normal operation of machine 1 after the training period.
[0112] It is assumed that one ring of the inner and outer rings 5, 6 rotates relative to the other ring of the inner and outer rings 6, 5, that the first to eighth normal distributions are determined and the detection threshold Sd is set.
[0113] The second part of the method comprises steps 21, 22, 23, 24, 25 and 26 if the machine 1 is a mobile machine.
[0114] If machine 1 is a stationary machine, the second part of the method does not include step 25, the second part of the method proceeding from step 24 to step 26.
[0115] When the value of each of the eight statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 is determined from the samples xi, the third determining means 13 determines in a step 30 a score Sc from the value of each of the eight statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 and the one to seventh normal distributions determined during the training period.
[0116] The third determining means 13 performs a Z-score normalization of each value of the statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 from the mean and the variance of the normal distribution of the statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 to obtain a standardized value of each value of the statistical parameters P1, P2, P3, P4, P5, P6, P7, P8.
[0117] The value of each of the statistical parameters P1, P2, P3, P4, P5, P6, P7, P8 is denoted by Val1, Val2, Val3, Val4, Val5, Val6, Val7, Val8.
[0118] The standardized value of each value Val1, Val2, Val3, Val4, Val5, Val6, Val7, Val8 is denoted as ValS1, ValS2, ValS3, ValS4, ValS5, ValS6, ValS7, ValS8.
[0119] The following equation connects each standardized value ValS1, ValS2, ValS3, ValS4, ValS5, ValS6, ValS7, ValS8 with the value Val1, Val2, Val3, Val4, Val5, Val6, Val7, Val8, the mean µ1, µ2, µ3, µ4, µ5, µ6, µ7, µ8 and the variance σ12,σ22,σ32,σ42,σ52,σ62,σ72,σ82: ValSj=Valj−μjσi2 where j varies between 1 and 7.
[0120] The score Sc is equal to the sum of the standardized values ValSj contained in a second interval having a lower bound of 1 minus a predetermined second quantile.
[0121] For example, the predetermined second quantile is equal to 90%.
[0122] During a step 31, the comparison means 14 compare the score Sc with the detection threshold Sd less the number of standardized values contained in the second interval determined in step 30.
[0123] The subtraction of the number of standardized values contained in the second interval determined in step 30 allows the detection threshold Sd to be weighted according to the number of standardized values contained in the second interval.
[0124] If the score Sc is less than the detection threshold Sd minus the number of normalized values contained in the second interval (step 32), bearing 4 is considered functional. The second part of the procedure returns to step 21.
[0125] If the score Sc is greater than the detection threshold Sd less the number of standardized values included in the second interval (step 32), the bearing 4 is considered defective and, during a step 33, the comparison means 14 issue an alarm to prevent the bearing 4 from being defective.
[0126] Since the determination of the Hjorth complexity parameter is insensitive to speed variations of bearing 4, the first statistical parameter P1, which is equal to the Hjorth complexity parameter of the set of vibration measurements, allows an accurate prediction of damage to bearing 4 by analyzing the evolution of only the first statistical parameter.
[0127] In order to increase the diagnostic reliability, as shown above, the early prediction of damage to the bearing 4 can be achieved by determining several statistical parameters from recorded vibration measurements of the bearing 4 by analyzing the evolution of these parameters, where one of the several statistical parameters is equal to the Hjorth complexity parameter.
[0128] Since several statistical parameters are used to predict a failure of the bearing 4, different types of defects of the bearing 4 can be detected.
[0129] Furthermore, the detection threshold Sd is determined based on vibration measurements on the bearing 4 in the machine 1, which makes it possible to define this threshold depending on the application of the machine 1 in order to obtain accurate monitoring of the bearing 4. QUOTES CONTAINED IN THE DESCRIPTION
[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited non-patent literature
[0000] Dispersion Entropy: A measure for time series analysis", M. Rostaghi and H. Azami, IEEE Signal processing letters, vol. 23, n.5, pp. 610-614, 2016
[0059] Efficiently measuring complexity on the basis of real-world data", VA Unakafova, K. Keller, Entropy, 15(10), 4392-4415
[0072]
Claims
[1] A method for monitoring a bearing device in a machine (1), the bearing device comprising a bearing (4) provided with an inner ring (5) and an outer ring (6) adapted to rotate concentrically with each other, and a vibration sensor (8) measuring vibrations of the bearing, the method comprising the following steps: Determining values of at least one first statistical parameter equal to the Hjorth complexity parameter (P1) from at least one set of vibration measurements provided by the sensor (8) during a training period following the implementation of the sensor (8) in the machine (1) and when the rotational speed of one of the inner and outer rings (4, 5) with respect to the other is comprised within a predetermined interval, Modelling the values of at least the first statistical parameter (P1,) with a normal distribution during the training period, Determining a detection threshold (Sd) from at least the normal distribution during the training period, Determining at least a first value of the first statistical parameter (P1) from a set of vibration measurements provided by the sensor during normal operation of the machine when the rotational speed of the one ring is included in the predetermined interval, Determining an evaluation (Sc) during normal operation of the machine (1) at least from the first value of the Hjorth complexity parameter (P1) and the normal distribution modelling the values of the first statistical parameter, Comparing the result (Sc) with the detection threshold (Sd), and Determining the failure of the bearing (4) according to the result of the comparison. [2] A method according to claim 1, wherein: Step (a) further comprises determining values of at least one second statistical parameter from the set of vibration measurements provided by the sensor (8) during the training period, Step (b) further comprises modelling the values of the second statistical parameter (P2, P3, P4, P5, P6, P7) with a normal distribution during the training period, Step (c) comprises determining the detection threshold (Sd) from the normal distribution modelling the values of the first statistical parameter and the normal distribution modelling the values of the second statistical parameter during the training period, Step (d) further comprises determining a first value of the second statistical parameter (P2, P3, P4, P5, P6, P7) from the totality of the vibration measurements provided by the sensor (8) during normal operation, and Step (e) comprises determining the score (Sc) during normal operation of the machine (1) from the first value of the Hjorth complexity parameter (P1), the first value of the second statistical parameter (P2, P3, P4, P5, P6, P7), the normal distribution modelling the values of the first statistical parameter and the normal distribution modelling the values of the second statistical parameter. [3] A method according to claim 1 or 2, wherein the values of each statistical parameter (P1, P2, P3, P4, P5, P6, P7) are determined during the training period and during normal operation of the machine when the kurtosis of the set of vibration measurements is less than a predetermined kurtosis threshold when the machine (1) is a mobile machine. [4] Method according to one of claims 1 to 3, wherein the determination of a detection threshold (Sd) comprises the summation of the values of each normal distribution contained in a first interval having a lower limit equal to one minus a predetermined first quantile, the detection threshold being equal to the sum. [5] Method according to one of claims 1 to 4, wherein determining a score (Sc) for each statistical parameter comprises: Performing a Z-score normalization of the first value of each statistical parameter from the mean and variance of the normal distribution that models the values of the second statistical parameter to obtain a standardized value of said statistical parameter, and Summation of each standardized value contained in a second interval with a lower bound equal to one minus a predetermined second quantile, the sum being equal to the score. [6] Method according to claim 5, wherein the comparison of the score (Sc) with the detection threshold (Sd) comprises the comparison of the score (Sc) with the detection threshold (Sd) less the number of standardized values contained in the second interval. [7] Method according to one of claims 1 to 6, wherein the bearing (4) is considered to be defective if the score (Sc) is greater than the detection threshold (Sd). [8] Method according to one of claims 2 to 7 as dependent on claim 2, wherein the second statistical parameter (P2, P3, P4, P5, P6, P7) comprises: the root mean square of the set of vibration measurements, or the sum of the areas of equal squares, where each square is defined by a diagonal connecting two vibration measurements of the set of vibration measurements, or the entropy of scattering of the set of measurements, or the permutation entropy of the set of measurements, or the mean value of a predetermined number of the strongest harmonics of a spectrum of the measurement series, where the first and the second statistical parameter are different. [9] A method according to any one of claims 2 to 7 as dependent on claim 2, further comprising defining a third statistical parameter (P8) equal to the square of the sum of the first and second statistical parameters. [10] A bearing device comprising a bearing (4) provided with an inner ring (5) and an outer ring (6) adapted to rotate concentrically with each other, and a vibration sensor measuring vibrations of the bearing, the bearing device further comprising: first determination means (10) for determining values of at least one first statistical parameter equal to the Hjorth complexity parameter (P1) from vibration measurements provided by the sensor when the rotational speed of one of the inner and outer rings (4, 5) with respect to the other ring is within a predetermined interval, modeling means (11) for modeling the values of at least the first statistical parameter with a normal distribution during a training period after the implementation of the sensor (8) in a machine (1), second determining means (12) for determining a detection threshold (Sd) from at least the normal distribution during the training period, third determination means (13) for determining a score (Sc) during normal operation of the machine (1) from at least a first value of the Hjorth complexity parameter (P1) and the normal distribution modelling the values of the first statistical parameter, and Comparing means (14) for comparing the result (Sc) with the detection threshold (Sd) and for determining the failure of the bearing (4) according to the result of the comparison.