CH717054A2 - Procedure for diagnosing a bearing. - Google Patents

Abstract

A method according to the invention for diagnosing one or more bearings (10, 20) of a machine from vibration data from a vibration sensor comprises the creation and feeding of digitized vibration data (46) into a processor (50). A frequency spectrum is calculated from the digitized vibration data (46) in the processor (50). The frequency spectrum is divided into a plurality of frequency ranges, each of the frequency ranges comprising a plurality of frequency lines. A data vector is determined from the frequency lines, a signal strength being determined from each data vector, the signal strength corresponding to the bearing defect or the position of the bearing (10, 20) in the machine and therefore clearly assigned to the corresponding bearing (10, 20) of the machine can be so that the operating state or the installation location of the bearing (10, 20) in the machine can be recorded at any time. A curve is formed from the data vector, using polynomial coefficients which describe the non-linearity of the curve and allow conclusions to be drawn about the type and severity of a bearing defect.

Description

background

The present invention relates to a method for diagnosing a bearing, in particular a roller bearing, ball bearing, roller bearing. The method can be used, for example, in a gearbox, a pump, a fan or another machine that contains rotating components.

In practice, defective bearings can lead to production stoppages, accidents, and consequential damage to machines and systems. For this reason, efforts have been made for decades to detect impending bearing damage before a defect can even occur in the bearing.

Proven measures to achieve a high level of reliability are oversizing and age-related and increasingly also condition-dependent replacement of the bearing, the replacement of the bearing must take place early in order to avoid a sudden failure of the bearing.

Due to the technical development, new possibilities for the diagnosis, control and condition monitoring of such bearings have arisen. With regard to the crack testing of bearings, these were, for example, optical, metallurgical, magnetic and radiological methods. The disadvantages of these methods are the dismantling and assembly costs, the lack of accessibility, the limited informative value (for example regarding the depth of the crack) and a limitation of the period for diagnosis, control or monitoring to service intervals.

By means of vibration measurement technology, monitoring of rotating parts is nowadays easily possible and is also used successfully in mechanical engineering to monitor the vibration amplitudes of roller bearings and gears.

State of the art

From JP2009020090 A a method and a device for diagnosing the condition of a ball bearing is known. Vibrations that are generated during operation of the ball bearing are detected by means of an acceleration sensor. The vibration signals are transmitted to a signal processing unit and converted into a digital signal. The level of the frequency spectrum is determined by means of a frequency analysis. Abnormal vibration frequencies resulting from a defective ball bearing part are determined on the basis of a predetermined mathematical relationship, whereby the level of the frequency spectrum which is associated with the abnormal vibration frequency is obtained. The level of the frequency spectrum determined in this way is compared and compared with a threshold value which is specified individually for each frequency of the fundamental wave with the abnormal vibration frequency and its harmonics, the size of which is determined so that the existence of an abnormality of a ball bearing part can be determined.

In a further development of this method according to JP2009020090 A, according to US2014214363 A1, instead of a threshold value, a data set from a database is used which contains the information for the frequencies of characteristic vibrations of a plurality of bearing manufacturers, in particular information on the error frequencies of the vibrations that occur when a certain bearing defect occurs. This information about the error frequencies can come from the manufacturers of the bearings or can be determined during operation of the bearing and stored in a memory unit. The information on the error frequencies of the vibrations from different manufacturers is available in the database for the known types of bearing error frequency. These bearing error frequency types therefore include the error frequencies for an outer shell bearing error, an inner shell bearing error, a cage error or a rolling element error. These bearing error frequencies for different manufacturers, for different types of bearings and for different mounting locations of the bearings in the relevant machine are stored in the database.

With the measuring method according to US2014214363 A1, the installation location of a bearing from a specific manufacturer can be determined if the bearing manufacturer is known. Conversely, if the installation location of the bearing is known, it can be determined which bearing has been installed by which manufacturer at a specific installation location. The measuring method can also be used to determine which bearing of the machine has a defect by superimposing the measured frequency ranges on the frequency ranges specified by the manufacturers and comparing them with them. For this purpose, the maximum vibration amplitudes are determined for each frequency range and compared with the corresponding frequency ranges of the various manufacturers. The frequency ranges that show the greatest agreement also have the highest vibration amplitude. In this way, the installation point of a defective bearing from a specific manufacturer can be identified on the machine or, if the installation point is known, each defective bearing from each manufacturer can be identified.

Finally, US2018 / 0347548A1 describes a method according to which the amplitudes of bearing defect frequencies and their harmonics are used to determine an increase in the signal energy in order to calculate a defect factor therefrom. The defect factor is empirically determined with a threshold value, in order to infer the state of the bearing. Methods for comparing a signal energy or an amplitude with threshold values are widespread, since a signal analysis can be carried out easily. The present invention is based on the pattern of distribution of signal energy or amplitude and not a correlation between defect frequencies and signal energy.

In the condition monitoring (condition monitoring) and the condition prediction (predictive maintenance) of rolling, roller or ball bearings, vibration signals are thus used. By means of an acceleration sensor, e.g. piezo or MEMS-based, which is ideally attached to a bearing, the vibration signals are recorded and fed as an analog electrical signal to a digitizing unit. The vibration signal digitized in this way is then broken down into its frequency parts by means of a Fourier analysis (FFT). The resulting frequency spectrum is used to analyze whether there is a defect in the bearing. A bearing manufacturer provides the frequencies that indicate a bearing defect. A bearing defect can be diagnosed if a peak is found at the frequency specified by the bearing manufacturer. The relevant bearing defect frequencies can be found online on a manufacturer website of the bearing manufacturer. These bearing defect frequencies can also be calculated for the bearing if the bearing geometry, diameter and number of rolling elements are known, but this bearing manufacturer information is not always available. In the following, a peak is understood to mean a peak value, in particular a local peak value in a data series.

A distinction is usually made between four types of bearing defect frequencies: rolling elements, cages, inner bearing shells or outer bearing shells, see also FIG. 3 in this regard.

Bearing defect frequency, determined for the outer bearing shell BPFO = ballpass frequency, outer race bearing defect frequency, determined for the bottom bracket shell BPFI = ballpass frequency, inner race bearing defect frequency of the bearing cage FTF = fundamental train frequency, cage speed bearing defect frequency of the rolling element BSF = Ball (roller) spin frequency where: n number of balls, rollers or rolling elements of the bearing for rotational frequency of the shaft in [1 / s] d ball diameter [m] D mean ball track diameter [m] ϕ bearing contact angle [-]

An expert can use the frequency spectrum and the bearing defect frequencies to visually determine whether there is a bearing defect. To do this, he looks at the frequency spectrum, as shown for example in FIG. 1, and compares the position of the peaks in the frequency spectrum with the bearing defect frequencies characteristic of each of the above-mentioned defects. If the peaks are at the positions of one of the bearing defect frequencies typical for the corresponding defect, then the corresponding defect is presumably present in the bearing.

In the automated bearing defect monitoring, a computer program performs the same analysis that an expert does, but without his ability to interpret. The current bearing defect frequencies must be derived from the bearing defect frequencies specified by the manufacturer. These are normalized to 60 RPM or 1 Hz, so they are scaled with the speed. If the outer bearing shell is defective, this is BPFO '= BPFO * RPM / 60. Because only the harmonic frequencies are relevant for an outer bearing shell defect, the relevant bearing defect frequencies for an outer bearing shell defect are: BPFO ', 2 * BPFO' and 3 * BPFO '.

The computer program now compares whether there are significant peaks in the frequency spectrum at the specified (relevant) bearing defect frequencies, or measures the strength of the peaks or outputs their vibration velocity value. Depending on the signal strength, the bearing is then classified as harmless, conspicuous, defective or critical, e.g. for RMS values according to the ISO-10816-3 standard. Analogous to this assessment, the computer program provides a report on the condition of the warehouse or makes a recommendation for a service technician.

The quality of an automated bearing defect monitoring therefore depends largely on the quality of the manufacturer's information on the relevant bearing defect frequencies. There are several sources of error that influence or even impair the accuracy or correctness of these bearing defect frequencies: Frequencies supplied by the bearing manufacturer are incorrect or inaccurate. The current natural frequencies of the bearings no longer match the frequencies of the bearing defects, for example due to wear and tear or aging processes that change the bearing geometry. An inaccurate or incorrect measured value of a speed measurement leads to the bearing defect frequencies being sought at a different location in the frequency spectrum.

Such sources of error cannot be identified by means of the previously known computer programs, since these computer programs do not automatically recognize and correct such deviations, whereas a trained expert recognizes these deviations during visual analysis.

Object of the invention

It is the object of the invention to develop a method which can deal with age-related frequency deviations or frequency deviations caused by inaccurate measured values, so that the output of incorrect measured values or recommendations can be prevented.

This makes it possible to determine bearing defects for any bearing at an early stage, so that failure of the bearing can be avoided with a high degree of reliability.

A further object of the invention is to provide a device for monitoring or diagnosing the condition of a bearing, which makes it possible to detect a defect before a possible failure of the bearing can occur in the event of a further load.

Description of the invention

The object of the invention is achieved by a method of diagnosing a bearing according to claim 1. Advantageous exemplary embodiments of the method are the subject matter of claims 2 to 13. A preferred use of the method is the subject matter of claim 14.

If the term “for example” is used in the following description, this term relates to exemplary embodiments and / or embodiments, which is not necessarily to be understood as a more preferred application of the teaching of the invention. The terms “preferably”, “preferred” are to be understood in a similar manner by referring to an example from a set of exemplary embodiments and / or embodiments, which is not necessarily to be understood as a preferred application of the teaching of the invention. Accordingly, the terms “for example”, “preferably” or “preferred” can relate to a plurality of exemplary embodiments and / or embodiments.

The following detailed description contains various exemplary embodiments for the method according to the invention and the device according to the invention for monitoring or diagnosing the condition of a bearing. The description of a specific device of a specific method is only to be regarded as an example. In the description and claims, the terms “contain”, “comprise”, “have” are interpreted as “including, but not limited to”.

The term “method for diagnosis” can also be understood to mean a method for monitoring or a method for condition diagnosis of the bearing. The following characteristics can therefore apply to any of the processes mentioned.

A method for diagnosing a bearing of a machine comprises a recording of vibration data by a vibration sensor attached to or in the vicinity of the bearing, the vibration data being characteristic of the operating behavior of the bearing, feeding the vibration data into a digitizing unit for digitizing the vibration data and Creating digitized vibration data, feeding the digitized vibration data into a processor, with frequency spectra being calculated from the digitized vibration data in the processor. In particular, the processor can perform an FFT (Fast Fourier Transformation) or an envelope spectrum from voltage, current, acceleration or vibration velocity measurement values to calculate the frequency spectrum.

The frequency spectrum is divided into a plurality of frequency ranges 1 to N, each of the frequency ranges 1 to N comprising a plurality of frequency lines 1 to K.

In Figure 1, a frequency spectrum with N = 9 frequency ranges is shown, the fundamental frequency (4) and 2 harmonics (5) and (6) and sidebands (8). Section (7) in Fig. 4 shows ± 15 frequency lines, i.e. K = 31, around the frequency range (6), whereby the frequency difference between two frequency lines corresponds to the resolution of the FFT. A data vector with N elements is determined for each of the K frequency lines, so that K data vectors are obtained for the N frequency ranges, a signal strength being determined from each of the K data vectors, a curve being formed from the signal strengths which can be described by means of a polynomial and their non-linearity, determined by assessing the level of the individual polynomial coefficients and their relationships to one another, allow statements about the bearing defect and its size, so that these K signal strengths allow statements about the bearing defect or correspond to the position of the bearing in the machine and therefore unambiguously can be assigned to the corresponding bearing of the machine so that the operating status or the installation location of the bearing in the machine can be recorded at any time. A curve is thus formed from the data vector, the polynomial coefficients being used, which describe the non-linearity of the curve and allow conclusions to be drawn about the type and severity of a bearing defect.

In particular, the characteristic curve of the signal strength corresponds to the bearing defect.

Modern processors process data efficiently when they are present as arrays (row, sequence), sequentially in the memory. An array (in the syntax of C, Pascal, Python etc.) can be one-dimensional or multi-dimensional. In the present application, a data vector is understood to be a one-dimensional array, the data vector containing a plurality of (amplitude) values from a Fourier frequency spectrum. A multi-dimensional array is called a matrix. A data vector must contain several frequency values and not just one scalar (amplitude) value. In addition, according to the invention, a plurality of data vectors must be provided, i.e. N> 1 and K> 1, so that amplitude variations caused by noise or impulsive interference frequencies can be minimized. K can be a maximum of 1024. N can be a maximum of 256.

It is known to the person skilled in the art that an assignment, as mentioned above, is most easily possible by means of a comparison of the signal strengths. It is also known to the person skilled in the art that the operating state or the installation location of the bearing can be recorded at any point in time. The greater the signal strength, the more likely it is that the correct assignment will be selected, in analogy to the maximum likelihood (ML) method, which is known per se to the person skilled in the art and should be mentioned here as an example for similar methods.

In particular, an index can be determined in each frequency range and an amplitude value associated with this index can be determined. An index corresponds to the position of an element in a data vector with K elements. The index of the greatest of the K signal strengths can be referred to as the optimal index and can allow conclusions to be drawn about the effective positions of the bearing defect frequencies.

The vibration sensor can in particular be designed as an acceleration sensor or speed sensor.

The frequency spectrum is thus divided into several frequency ranges. A frequency range is a range in the spectrum that extends around a fundamental frequency or harmonics or a sideband. The frequency range comprises a plurality of frequency lines. A BPFO defect has three frequency ranges, for example, a BPFI defect has nine frequency ranges, for example. The three frequency ranges of the BPFO defect consist of the fundamental frequency and two harmonics. The nine frequency ranges of the BPFI defect consist of the basic frequency, two harmonics and one sideband each. A frequency range can contain, for example, 31 frequency lines (± 15 + 1). For example, for the fundamental frequency, the frequency line with index -15 for the first harmonic corresponds to the frequency line with index -30, because the first harmonic has a frequency that is twice as high as the fundamental frequency. Starting from index -15 of the first fundamental frequency or its frequency range, all 9 amplitude values, one per frequency range in the case of the present BPFI example, are combined in a data vector, e.g. added up according to Fig. 5 and Fig. 6, and a measured value is formed from them. This results in 31 measured values in this example. If you have then calculated all 31 measured values in this way, you can see which measured value has the highest value and its corresponding position. The position corresponds to the index of the frequency line with the highest value in relation to the frequency range. This position can be viewed as the optimal position within any frequency range.

If there is no bearing defect, then the K measured values averaged over N frequency ranges are similar in value because it is noise and not a signal which is caused by an excitation or natural frequency.

Then, for example, a polynomial of the P th degree can be calculated based on all measured values. The bearing defect is assessed by considering this polynomial. For this purpose, the polynomial is compared with a linear characteristic. If the deviation between the polynomial and the linear characteristic curve exceeds a threshold value, this deviation is interpreted as a bearing or gearbox defect. This method has the great advantage that it is more robust than the established comparison of amplitude values with threshold values and can be automated, i.e. no vibration expert has to assess the measured values.

According to one exemplary embodiment, the frequency spectra of the bearing can be compared with a plurality of error frequency spectra, for example from different bearing manufacturers. The error frequency spectra can be stored in a database. The error frequency spectra can be clearly assigned to different bearing defects.

According to one embodiment, an expanded measurement window is determined for each frequency line, which takes into account several frequency lines lying next to one another. In particular, the expanded measurement window can consist of several frequency lines lying next to one another. With the extended measurement window, not only one frequency line per frequency band is taken into account in the data vector, but 3, 5, 7 frequency lines or up to 5% of the frequency lines, based on a selected highest frequency in the frequency spectrum. The adjacent frequency lines are viewed and applied as a single frequency line. As a result, each frequency range expands in the same proportion as the width of the measurement window increases. By using an extended measurement window, the length of the data vector with K elements increases proportionally to the number of additionally considered frequency lines.

According to one embodiment, a search method for signal strengths is used to determine the signal strengths, as shown for example in FIGS. 2 and 5.

The signal strengths can be one or more elements from the group of SNR, FNR (fault frequency-to-noise ratio), amplitudes, sum of amplitudes, peak values, RMS measured values, MA6 *, FM4 *, NB4, MA6, FM4, NA4, PAPR, crest factor, energy ratio, kurtosis include. For example, a signal strength can be formed from the N measured values, e.g. SNR, the strongest being selected from the K SNR values. Alternatively, to improve the quality of the position estimation, the SNR and, for example, MA6 can be calculated from the N measured values, and then combine them from twice the number of signal strengths (one half SNR, other half according to this example MA6) to determine the position of the bearing defect frequency with the highest probability to be determined.

According to one embodiment, the optimal position in the frequency ranges, that is to say an optimal index, can be determined by means of an optimization criterion. In particular, the optimization criterion can be one or more of the elements from the group SNR, FNR, amplitudes, sum of all amplitudes, peak values, RMS measured values, MA6 *, FM4 *, NB4, MA6, FM4, NA4, PAPR, crest factor, energy ratio , Include kurtosis. In particular, a combination or fusion of signal strengths can take place.

According to one exemplary embodiment, the signal strengths or the maximum measured values can be used for further analyzes, with further data, for example data from a plurality of days, being able to be taken into account.

In particular, a plurality of statistical calculations can be carried out, the result of which gives a probability for a specific frequency.

According to one embodiment, the measured values of the frequency ranges can be analyzed by means of a classification method which, based on the distribution function of the cumulative sum, results in a curve which can be described by means of a polynomial and compared with a linear curve, from which a deviation value can be determined. In particular, a classification method from the field of machine learning can be used. The classification method can be used to determine the type of bearing defect or its severity, for example whether there is a rolling element, cage, inner bearing shell or outer bearing shell defect by dividing the degree of deviation into classes and each class corresponding to a degree of severity. The classification method can thus be used to determine the stage at which the bearing defect is, that is to say how advanced the defect is. For this purpose, a polynomial of the P-th degree or an interpolation function or a curve fitting function can be calculated, whereby, depending on the size of the polynomial coefficients, conclusions can be drawn about the linearity or nonlinearity and thus the strength of the existing bearing defect. For this purpose, for example, a polynomial of the P th degree can be calculated based on all measured values. Depending on the size of the polynomial coefficients, conclusions can be drawn about the linearity or non-linearity and thus the strength, that is to say the degree of severity, of the existing bearing defect.

To automate the analysis, the data can be processed by machine learning methods, for example by means of a neural network, in particular a deep neural network, with no polynomial being analyzed here, but rather network coefficients.

The probabilities derived from the resulting measured values can be assigned to an assessment scheme on the basis of which the bearing is classified, for example, as good, conspicuous, critical or defective.

A method according to one of the preceding exemplary embodiments can be used to determine the rotational frequency (RPM) of a machine.

In particular, the method according to one of the preceding exemplary embodiments can be used to determine the rotational frequency (RPM) of a machine without a rotational speed sensor.

The method according to one of the preceding exemplary embodiments can be used to determine the bearing type or the bearing manufacturer. The method can also be used to prove that the wrong bearing was installed during maintenance or repair without having to dismantle the machine.

In order to be able to determine the peaks of the bearing defect frequencies, various methods for diagnosing, monitoring or diagnosing the condition of a bearing can be used, on the one hand a diagnostic method which uses an extended measurement window or a diagnostic method which includes a search method for peaks (peak Searcher).

The present method can also be used to determine the rotational frequency (RPM) of a machine, in which case an expensive rotational speed sensor can be dispensed with. Furthermore, the rotational frequency (RPM) of a machine can be determined with plain bearings, thereby eliminating the need for expensive distance measuring sensors, for which no rolling bearing defect frequencies can be determined. In particular, no speed sensor is required to determine the operating state of each of the bearings.

The present method is also suitable for determining the bearing type and bearing manufacturer, making use of the small differences in the bearing defect frequencies. Alternatively, it can be determined whether and, if so, which of the four typical bearing defects is present.

The method is therefore suitable for combination with other monitoring methods in the sense of expansion or as backup functions. Examples of the use of the method in the machine industry are bearing, gear, shaft breakage, axle breakage, unbalance, misalignment and looseness monitoring. In the present application, misalignment monitoring is understood to mean monitoring of an alignment error. In the present application, looseness monitoring is understood to mean monitoring of looseness, for example of screws.

Brief description of the drawings

The methods according to the invention are illustrated below with the aid of a few exemplary embodiments. 1 shows a representation of a frequency spectrum of a bearing with a defect, the x-axis describes the frequency in Hertz and the y-axis describes the amplitude, for example in volts, and FIG. 2 shows a flow diagram of the data flow from the bearing to the display of the results of the method according to the invention for diagnosing the bearing or bearings, FIG. 3 a visualization of the parameters required for the determination of a certain bearing defect, Creation of signal strengths in a calculation unit, FIG. 6 a representation of the method according to one of the preceding exemplary embodiments, FIG. 7 a graph of the amplitude as a function of the frequency.

Detailed description of the drawings

If elements are shown in dashed lines in the figures, these are options which are possible as an alternative or in addition to the embodiment shown in solid lines.

1 shows an illustration of a frequency spectrum 1 of a bearing with a defect. In Fig. 1, the example of an inner bearing shell defect is shown. The frequency spectrum contains a bearing defect fundamental frequency 4, a first harmonic 5, a second harmonic 6, which correspond to the defect frequencies characteristic of the inner bearing shell defect. The bearing defect fundamental frequency 4, the first harmonic 5 and the second harmonic 6 are also referred to below as main frequencies. Each main frequency has sidebands 8; one sideband each is shown on the left and right in FIG. The frequency in Hz is plotted on the abscissa of the diagram, the amplitude in V on the ordinate.

2 shows a flow diagram of the data flow from the warehouse to the display of the results of the diagnostic method according to the invention for a measurement i, where i denotes the sequence number of the measurement or of the spectrum. In the present illustration, a first bearing 10 and a second bearing 20, which are arranged on a shaft, are monitored. For the detection of vibrations of the first bearing 10 and the second bearing 20, a vibration sensor 15, 25 is attached to the bearing or in the vicinity of the bearing. Each of the vibration sensors 15, 25 thus serves to record vibration data from the corresponding bearing 10, 20. In the present exemplary embodiment, the vibration sensor 15, 25 is designed as an acceleration sensor. The vibration data 16, 26 are fed into a digitization unit 40 for digitizing the vibration data 16, 26. In the digitization unit 40, digitized vibration data 46 are created from the vibration data. The digitized vibration data 46 are fed into a processor 50. The processor 50 can comprise a local PC or a cloud solution. The processor 50 calculates frequency spectra 1 from the digitized vibration data 46. For this purpose, the processor 50 includes a calculation unit 51. The frequency spectra of the corresponding bearing are transmitted as result data 61 to a comparison unit 52. In the calculation unit, each frequency spectrum is divided into a plurality of frequency ranges. Each of the frequency ranges includes a plurality of frequency lines. An index is determined in each frequency range and an amplitude value associated with this index is determined. The index in the frequency domain references the position in the frequency spectrum (FFT), an amplitude value that can be determined (see also pages 9, 10). An index, i.e. a position in the frequency spectrum, is calculated according to round (f / fa), where f is the frequency for which you are looking for the index, and fa is the resolution of the spectrum, e.g. 0.1 Hz.

The amplitude value belonging to the index can be summarized in a data vector as follows: The amplitude value xi, k, n of the frequency lines of the nth frequency range, of the kth measured value, of the ith spectrum is determined in a first step collected. In a second step, the k-th data vector is created x (i, k) = [xi, k, 1, xi, k, 2, ..., xi, k, N]. In a third step, a new measured value is determined from the k-th data vector, the signal strength s (i, k), i.e. for example SNR, FNR amplitudes, sum of amplitudes, peak values, RMS measured values, MA6 *, FM4 *, NB4, MA6, FM4, NA4, PAPR, crest factor, energy ratio, kurtosis calculated. If all K signal strengths are available, a new data vector can be created in a fourth step, y (i) = [s (i, 1), s (i, 2), ... s (i, K)], as well whose largest value is found in y (i) and whose position is found in the data vector, which is carried out in the comparison unit 52. In a fifth step, the data vector y (i) is sent to the classification unit 53. In a sixth step, the data vector y (i) is analyzed in the classification unit 53, for example according to the following rule: if the data vector y (i) is linear or the result of a function f (y (i)), there is no bearing defect, otherwise there is a Bearing defect likely. In a seventh step, the type of bearing defect can be determined. In other words, a measured value is determined for the corresponding frequency range, a signal strength being determined from all measured values. The signal strength corresponds to the position of the bearing 10, 20 in the machine and can therefore be clearly assigned to the corresponding bearing 10, 20 of the machine, so that the operating state or the installation location of the bearing 10, 20 in the machine can be detected at any time, with in particular, the characteristic curve of the signal strength corresponds to the bearing defect.

According to one embodiment, the i-th frequency spectrum is subdivided into a plurality of frequency ranges by means of the calculation unit 51. A frequency line is successively taken from each frequency range, from left to right, i.e. from low to higher frequencies in the frequency range, in a frequency range 7 as shown in FIG. 4, for example a number of nine frequency lines for the BPFI shown in FIG Malfunction. The calculation unit 51 creates a data vector x (i, k) = [xi, k, 1, x1, k, 2, ..., xi, k, N] with k = from the frequency lines or their amplitude and phase 1,2, ... K, which in the case without an extended measurement window has a number K of data vectors. The calculation unit 51 creates a number of data vectors which corresponds to the number K of frequency lines in the frequency range of the fundamental frequency Fg, Fg being in the middle of the K frequency lines. For BPFI bearing defects, this number K is, for example, 31 or 5% of the frequency lines based on Fmax. Fmax can be the highest of all bearing defect frequencies or an empirically determined frequency. The number of frequency ranges depends on the type of bearing defect to be examined and the selected number of harmonics and sidebands. The calculation unit 51 calculates a signal strength from each data vector. The signal strength can be selected from a variety of possible parameters, the signal strength can be a scalar value, an RMS measured value, an SNR or FNR value, the largest value of all elements in the data vector, amplitudes, sum of amplitudes, peak values, MA6 *, FM4 *, Include NB4, MA6, FM4, NA4, PAPR, Crest-Factor, Energy Ratio, Kurtosis. The resulting K result data 61 are fed to the comparison unit 52. The comparison unit 52 collects all result data which the iterative calculation process delivers from left to right, that is to say from low to higher frequencies in the frequency range. The comparison unit 52 determines the greatest value of all signal strengths from all the result data obtained and forwards this maximum value and a new data vector containing all K maximum values or K * N measured values to the classification unit 53. From all the result data obtained, the classification unit 53 determines which bearing defect is most likely to be present or which bearing manufacturer is the bearing whose digitized vibration data have been processed. Alternatively or in addition, historical results data can also be used.

The result 63 processed in this way in the classification unit 53 can be visualized in a visualization unit 54. The visualization unit 54 can be designed, for example, as a display or as a storage medium for the data vector or tensor.

The classification unit processes the data vector y (i) by checking f (y (i)) for linearity: the person skilled in the art has various options for this. If the values in y (i) are similar to a uniform distribution, which is the case if there is no bearing defect because the spectral values correspond approximately to white noise, then the values in y (i) are pretty much the same size (i.e. flat). If there is a bearing defect, then typically 1-5 values in y (i) are massively larger and you get a bell-like curve when you record y (i). Since these two cases are easy to distinguish, also graphically, it is easy for a person skilled in the art to differentiate between defective and good bearings. For automated analysis, the values on y (i) are shown as a cumulative sum. If the values in y (i) are the same, the result is a linear line curve, otherwise an S-curve. A difference between a linear (theory) and the effective curve based on y (i) can be determined. The effective curve is described using a polynomial of the P-th degree. The difference between the 1st polynomial and the higher order polynomial can be used to determine the severity of the manifestation of a bearing defect. The classification unit 51 calculates such a value, which is compared with empirical values, i.e. threshold values, and possibly triggers a warning or alarm. The user can then assess the severity of the defect in more detail in the visualization unit 54 and take appropriate measures.

3 shows the parameters required to determine a specific bearing defect. A bearing 10 is arranged on a shaft 5. The bearing 10 is visible in the left-hand part of FIG. 3 in a side view. The bearing 10 is shown enlarged on the right-hand side of FIG. 3 in a top view. Some of the balls 13 of the ball bearing shown by way of example have been omitted in order not to overload the illustration. Instead of the balls 13, other rolling elements can also be provided, which is not shown in the drawing. The balls 13 are arranged between an inner ring 11 and an outer ring 12. The ball diameter d or the rolling element diameter and the mean ball race diameter D or the mean rolling element path diameter, which are shown in FIG. 3.

The data vectors do not contain any arbitrary values, because the data vectors contain the resultant of a spectral analysis (amplitude values), the position (index range) of which is determined deterministically, based on the position of a bearing defect according to the calculations on page 4 for PBFI, BPFO etc. , or according to the statements above. Thus, a person skilled in the art can create the data vector and it is not arbitrary.

The installation location of a bearing cannot be determined if only the same bearing types from the same manufacturer are installed in a machine, which rarely occurs in practice and is only the case with small motors. US 2014/0214363 A1 presents an alternative method with which the idea of bearing identification and installation location can be determined. The method in this application also uses the bearing-specific defect frequencies, calculated according to the formulas on page 4, but the method differs fundamentally from US 2014/0214363 A1. In US 2014/0214363 A1, only the basic frequencies are used for bearing identification, amplitude values are summed up at the bearing defect frequencies and no measures are proposed against interfering signals. And if there is no bearing defect, which is not examined and assessed in US 2014/0214363 A1, reliable bearing identification cannot take place because the spectrum does not show any prominent peaks at the bearing defect frequencies.

4 shows a further example of a method for monitoring or diagnosing the condition of a bearing, for which an extended measurement window is used in the calculation unit 51. According to this exemplary embodiment, M spectral lines are used for the present expanded measurement window. e.g. M = 3 (± 1 around a spectral line). Not only one measured value per frequency range is taken into account, but M measured values and the M amplitudes of the M frequency lines of a frequency range are summarized, e.g. averaged or added up. In this example, the frequency range does not expand and retains its width K. Usually, the largest measured value within the measurement window is used, for example the maximum amplitude or the RMS value of the data in the measurement window, see also US20140214363 A1.

The method for monitoring or diagnosing the condition of a bearing according to an exemplary embodiment thus comprises an expanded measurement window by means of which not only the signal strength, i.e. the peak, of an individual specific frequency is determined or its harmonics, ie fdo ', 2 * fdo', 3 * fdo 'and possibly a number of sidebands, but rather the environment of each bearing defect frequency is taken into account, which is shown in FIG. 4. Fdo 'is particularly representative of BPFO', BPFI ', FTF', BSF '.

A spectrum according to the example shown in FIG. 4 has a frequency resolution of 0.25 Hz. The speed of the machine is 750 RPM, i.e. the rotational frequency is 12.5 Hz, and a maximum frequency error of ± 5% (± 0.625Hz) is expected. Then one considers left and right of the actual bearing defect frequency up to a further 3 frequency lines, i.e. ± 3. In this example, the extended measurement window would expand each frequency range by a factor of M = 7. If, however, the M amplitudes per calculation step are combined in order to calculate a signal strength, then the frequency range does not have to be expanded.

4 shows the FFT spectrum 3, which includes a bearing defect fundamental frequency 4, which is referred to below as Fg1. Furthermore, the FFT spectrum 3 comprises a first harmonic 5 and a second harmonic 6 as well as a sideband frequency each. The FFT spectrum 3 can in particular be designed as an envelope curve spectrum, an amplitude spectrum or a density spectrum. For the second harmonic 6, a frequency range 7 is shown enlarged by the frequency line 6 (FFT-bin), which shows a range around the second harmonic 6 with ± 15 FFT frequency lines. All relevant frequency lines in the envelope spectrum, which are shown with dashed lines, show the bearing defect frequencies. These bearing defect frequencies are specified by the bearing manufacturer for a specific bearing type. A window is defined around these frequency lines, for example with ± 15 FFT frequency lines or ± 5% of Fmax, in order to find the exact frequency, i.e. the position of the peak in the spectrum. Fmax is the highest of all possible defect frequencies or can be determined empirically or by the rotational frequency.

Fig. 5 shows a search method for signal strengths. In particular, the signal strengths can be in the form of peaks. According to this exemplary embodiment, the comparison unit 52 comprises a search unit. The search unit can be used to carry out a search method for determining the greatest signal strength based on the measured values supplied by the calculation unit 51. A method for monitoring or diagnosing the condition of a bearing according to this exemplary embodiment comprises, in particular, a search method for peaks (peak searcher).

According to the exemplary embodiment shown in FIG. 4, a signal strength is determined in the calculation unit 51 from nine frequency ranges in FIG. The signal strength per FFT frequency line is calculated step by step over all frequency ranges. According to the present exemplary embodiment, this results in 31 measured values, which correspond to a frequency range with ± 15 FFT frequency lines. The same percentage tolerance includes fewer FFT frequency lines at lower frequencies. Thus, Fg1 with the FFT frequency lines -5, 2 * Fg1, i.e. the second harmonic, with the FFT frequency lines -10, 3 * Fg1 with the FFT frequency lines -15 each describe the same percentage deviation from Fg1, 2 * Fg1, 3 * Fg1. By means of interpolation, combination or summation of several frequency lines, all frequency lines can be brought to the number of 31 measured values and thus the multiplicative nature of the frequency deviation can be taken into account. The position of the largest of the 31 measured values most likely corresponds to the correct deviation of the bearing defect frequency in relation to the theoretical bearing defect frequency, for example a Fg specified by the manufacturer.

In this way, the search method for peaks (peak searcher) determines the most likely correct bearing defect frequencies or the rotational frequency of a shaft based on an optimization criterion. The optimization criterion can include one of the elements SNR, FNR, amplitude value, amplitudes, sum of amplitudes, peak value, RMS measured value, MA6 *, FM4 *, NB4, MA6, FM4, NA4, PAPR, crest factor, energy ratio, kurtosis. In this context, SNR is used as an abbreviation for the signal-to-noise ratio. FNR is used in this context as an abbreviation for the fault frequency-to-noise ratio.

The signal strength Ps is determined according to the following relationship. Here i is the serial number of the measurement (e.g. day i), N is the number of frequency ranges or all defect frequencies including harmonics and sidebands of a bearing defect type, such as BPFI or BPFO, M is the width of the extended measurement window, and k = 1..K is the number of frequency lines in each frequency range around the N defect frequencies:

The multiplicative nature of the frequency deviation and the absolute index in the spectrum are not taken into account in the formula. Alternatively, Ps (i, k) can also be formed from the variance of the relevant N or N * M frequency lines. The noise power is determined as follows. Half of the FFT spectrum, F (i), with a bandwidth of fs / 2 is used. In the i-th spectrum F (i), all N or N * M frequency lines are then zeroed: F (i) → F '(i, k).

The noise power Pn then results as

The FNR for the spectrum i is then calculated according to an exemplary embodiment.

In particular, M can be in the range of a maximum of +/- 10% of Fmax.

The measurement of the peaks or Ps of the bearing defect frequencies then takes place without taking into account further, possibly incorrect or disruptive, values in the frequency spectrum, as is shown in FIG. For this purpose, a plurality of frequency lines is determined. In the exemplary embodiment shown in FIG. 5, the signal strength is determined from nine frequency ranges.

Using the expanded measurement window, further frequency lines can be taken into account for calculating the signal strength, for example the frequency lines on the left and right (± 1) of each frequency line.

The resulting measured values of the bearing defect frequencies can be analyzed by means of a classification method, which in the following will also be referred to as the bearing defect classification method (BDC, bearing defect classifier). The resulting measured values can be used for further analyzes, further data, for example data from a plurality of days, being taken into account. The bearing defects can thus be identified and classified with a high degree of probability. A plurality of statistical calculations can be carried out by means of the BDC, the extended result of which gives a probability for a specific frequency. If a certain frequency range ± Fps is taken into account, which corresponds to K frequency lines, linear or non-linear characteristics are obtained. If the characteristic is linear or almost linear, the existence of a bearing defect is unlikely. The classification method can be used to determine the type of bearing defect or its severity, for example whether there is a rolling element, cage, inner bearing shell or outer bearing shell defect. In addition, it can be determined at what stage this bearing defect is, that is, how advanced the bearing defect is. For this purpose, for example, a polynomial of the P th degree can be calculated or another method from the field of interpolation or curve fitting can be used for the calculation. Depending on the size of the P coefficients, conclusions can be drawn about the linearity or non-linearity and thus the strength of the existing bearing defect. To automate the analysis, the data can be processed using machine learning methods, for example using a neural network. The legibility of the analysis or its interpretation can be further improved by displaying the results graphically, for example as a two-dimensional diagram, analogous to a spectrogram.

The probabilities derived from the resulting measured values can be assigned to an assessment scheme. For example, the probabilities can be communicated to the customer or expert in such a way that an assessment scheme is assigned to the statistical values or the polynomial coefficients. For example, a warehouse can be classified as good, conspicuous, critical or defective. A recommendation for action can be assigned to each of these assessments. The recommendation for action can include a graphic representation, for example a traffic light display, or it can be written in text form.

6 shows an illustration of the method according to one of the preceding exemplary embodiments. As described above, N = 9 frequency ranges 7 are evaluated by way of example. In a first step, the amplitude value xk, n of the k-th measured value of the frequency lines of the n-th frequency range is collected. In a second step, the k-th data vector is created according to x (k) = [xk, n, xk, 2, ... xk, N]

In a third step, a new measured value, which is referred to below as signal strength, is determined from the k-th data vector, that is, SNR, FNR, peak values, RMS measured values, MA6 *, FM4 *, NB4, MA6, FM4, NA4, PAPR, crest factor, energy ratio or kurtosis are calculated.

As soon as all K signal strengths are available, for example, as described above, 31 signal strengths, a new data vector can be created in a fourth step in the calculation unit 51 in accordance with the calculation rule below: y (i) = [x (1), x (2), ... x (31)] thus the largest value in y (i) and its position in the vector can be found.

In a fifth step, the data vector y (i) is sent to the classification unit. In a sixth step, the data vector y (i) is analyzed, a linear distribution of the cumulative sum indicating that there is no bearing defect, otherwise a bearing defect is probable.

In a seventh step, the type of bearing defect is determined, for example BPFI, BSF.

7 shows a graph of the frequency spectrum F (i), which shows the amplitude as a function of the frequency. The amplitude is plotted on the ordinate, the frequency on the abscissa. In the frequency spectrum, three of the frequency ranges 7 are shown as examples, the frequency range (1), the frequency range (2) and the frequency range (N). Furthermore, the bearing defect frequencies Fg1, Fg2, FgN in the corresponding frequency ranges (1, 2, ..., N) are shown as examples. The bearing defect fundamental frequency 4 already described in connection with FIG. 4 is denoted by Fg1. The harmonics of the bearing defect frequency 4 or the sideband center frequency of the bearing defect frequency are denoted by Fg2 and FgN.

It is obvious to a person skilled in the art that many further variants are possible in addition to the exemplary embodiments described without deviating from the inventive concept. The subject matter of the invention is therefore not restricted by the preceding description and is determined by the scope of protection which is defined by the claims. The broadest possible reading of the claims is decisive for the interpretation of the claims or the description. In particular, the terms "contain" or "contain" are to be interpreted in such a way that they refer to elements, components or steps in a non-exclusive sense, which is intended to indicate that the elements, components or steps can be present or can be used, that they can be combined with other elements, components or steps that are not explicitly mentioned. When the claims relate to an element or component from a group which may consist of A, B, C to N elements or components, this formulation should be interpreted in such a way that only a single element of that group is required, and not one Combination of A and N, B and N, or any other combination of two or more elements or components of this group.

Claims (14)

A method for diagnosing a bearing (10, 20) of a machine, comprising recording vibration data (16, 26) by a vibration sensor (15, 25) attached to or in the vicinity of the bearing, the vibration data (16, 26) being characteristic for the operating behavior of the bearing (10, 20), feeding the vibration data (16, 26) into a digitization unit (40) for digitizing the vibration data (16, 26) and creating digitized vibration data (46), feeding in the digitized vibration data (46 ) in a processor (50), the processor (50) calculating a frequency spectrum (1) from the digitized vibration data (46), the frequency spectrum (1) being subdivided into a plurality of frequency ranges 1 to N, each of the Frequency ranges 1 to N comprises a plurality of frequency lines 1 to K, a data vector with N elements being determined for each of the K frequency lines, so that K data vectors for the N frequency ranges Ren can be obtained, whereby a signal strength is determined from each of the K data vectors, whereby a curve is formed from the signal strengths, which can be described by means of a polynomial and its non-linearity, determined by assessing the height of the individual polynomial coefficients and their relationships to one another Allow statements about the bearing defect and its size so that these K signal strengths permit statements about the bearing defect or correspond to the position of the bearing (10, 20) in the machine and can therefore be clearly assigned to the corresponding bearing (10, 20) of the machine, so that the operating state or the installation location of the bearing (10, 20) in the machine can be recorded at any time. 2. The method according to claim 1, wherein an expanded measurement window is used for each frequency line which takes into account a plurality of frequency lines lying next to one another. 3. The method according to any one of claims 1 or 2, wherein a search method for signal strengths is used to determine signal strengths. 4. The method according to any one of the preceding claims, wherein the signal strengths an element from the group of SNR, FNR, amplitudes, sum of amplitudes, peak values, RMS measured values, MA6 *, FM4 *, NB4, MA6, FM4, NA4, PAPR , Crest factor, energy ratio, kurtosis include. 5. The method according to any one of the preceding claims, wherein an optimal index is determined by means of an optimization criterion, the optimization criterion being an element from the group of SNR, FNR, amplitudes, sum of amplitudes, peak values, RMS measured values, MA6 *, FM4 *, NB4 , MA6, FM4, NA4, PAPR, Crest-factor, Energy-Ratio, Kurtosis includes. 6. The method according to any one of the preceding claims, wherein the signal strength is used for further analyzes, further data, for example data from a plurality of days, can be taken into account. 7. The method according to any one of the preceding claims, wherein the measured values of the frequency ranges are analyzed by means of a classification method which, based on the distribution function of the cumulative sum, results in a curve which can be described by means of a polynomial and compared with a linear curve, from which a deviation value is determined can be. 8. The method according to claim 7, wherein the type of bearing defect or its severity is determined by means of the classification method by dividing the degree of deviation into classes and each class corresponds to a specific degree of severity. 9. The method according to any one of claims 7 or 8, wherein the classification method is used to determine the stage in which the bearing defect is, wherein a polynomial P-th degree or an interpolation function or a curve fitting function is calculated, depending on the size of the polynomial coefficient on the Linearity or non-linearity and thus the strength of the existing bearing defect can be inferred. 10. The method according to any one of the preceding claims, wherein to automate the analysis, the data are processed by machine learning methods, for example by means of a neural network, in particular by means of a deep neural network. 11. The method according to any one of the preceding claims, wherein a plurality of statistical calculations is carried out, the result of which gives a probability for a specific error frequency. 12. The method according to any one of the preceding claims, wherein probabilities derived from the measured values are assigned to an assessment scheme, on the basis of which the bearing can be classified as good, conspicuous, critical or defective. 13. The method according to any one of the preceding claims, wherein no speed sensor is required to determine the operating state of each of the bearings. 14. Use of the method to determine the rotational frequency (RPM) of a machine or to determine the bearing type and bearing manufacturer.