Data extraction device and abnormality monitoring apparatus

Abstract

To provide technique capable of improving the accuracy of abnormality determination compared to conventional techniques.SOLUTION: In a data extraction device 30 according to the present disclosure, since an envelope spectrum being a spectrum of an envelope-processed wave obtained by performing envelope processing on a vibration waveform is used as data for abnormality determination, the device is less susceptible to noise in comparison to conventional methods that use the spectrum of the vibration wave itself as the data for abnormality determination and improves accuracy in abnormality determination. Moreover, instead of using the entire envelope spectrum of the vibration wave as the data for abnormality determination, the device extracts the envelope spectrum within a limited vibration frequency range that includes a vibration frequency of a specific wave in plural types of waves included in the envelope-processed wave as the data for abnormality determination. This excludes data not used in abnormality determination and facilitates abnormality determination. Furthermore, even if the vibration frequency of the specific wave changes due to the change in the rotational speed of a bearing 94, since the device calculates the vibration frequency of the specific wave from the rotational speed of the bearing 94 and determines the limited vibration frequency range, the abnormality determination can be easily performed without being affected by the change in the rotational speed of the bearing 94.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a data extraction device that extracts data for determining an abnormality in a rotating part of a monitored machine, and an abnormality monitoring device having such a data extraction device. [Background technology]

[0002] Known abnormality monitoring devices of this type are those that perform abnormality determination based on the spectrum of vibration waves generated during rotation of a rotating part of a monitored machine (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2001-99757 A (Claim 1, paragraph

[0005] ) Summary of the Invention [Problem to be solved by the invention]

[0004] However, in conventional anomaly monitoring devices, it has been difficult to accurately determine anomalies due to the large influence of noise contained in vibration waves. Therefore, this application discloses a technique that can improve the accuracy of anomaly determination compared to conventional techniques. [Means for solving the problem]

[0005] One aspect of the invention of the present disclosure is a data extraction device that extracts data for abnormality determination of a rotating part of a monitored machine, the data extraction device comprising: an envelope processing unit that performs envelope processing on a vibration wave of the rotating part during rotation to obtain an envelope processed wave; a spectrum calculation unit that calculates an envelope spectrum, which is the spectrum of the envelope processed wave; a basic data memory unit that stores ratio data for specifying a ratio between the frequency of a specific wave among multiple types of waves contained in the envelope processed wave and the reference rotation speed when the rotating part rotates at a predetermined reference rotation speed; a frequency calculation unit configured to calculate the frequency of the specific wave as a real special frequency when the rotating part rotates at the actual rotation speed based on an actual rotation speed, which is the actual rotation speed of the rotating part, and the ratio data; and a data extraction unit configured to extract the envelope spectrum of a limited frequency range including the real special frequency from the envelope spectrum when the rotating part rotates at the actual rotation speed, as the abnormality determination data. Effect of the Invention

[0006] In the data extraction device of the present disclosure, the envelope spectrum, which is the spectrum of the envelope processed wave obtained by enveloping the vibration wave, is used as the data for abnormality judgment, so that it is less susceptible to noise and the accuracy of abnormality judgment is higher than that of the conventional device that uses the spectrum of the vibration wave itself. In addition, instead of using the entire envelope spectrum of the vibration wave as the data for abnormality judgment, the envelope spectrum of a limited frequency range including the frequency of a specific wave among the multiple types of waves included in the envelope processed wave is extracted as the data for abnormality judgment, so that data not used for abnormality judgment is eliminated and abnormality judgment is made easier. Moreover, even if the frequency of the specific wave changes with the change in the rotation speed of the rotating part, the frequency of the specific wave is calculated from the rotation speed of the rotating part to determine the limited frequency range, so that by using the data extraction device of the present disclosure, abnormality judgment can be easily performed without being affected by the change in the rotation speed of the rotating part. [Brief description of the drawings]

[0007] [Figure 1] Figure 1 shows a schematic diagram of the monitored machine. [Diagram 2] FIG. 2 is a circuit diagram showing the configuration of the abnormality monitoring device. [Diagram 3] FIG. 3 is a block diagram showing the configuration of the abnormality monitoring device. [Figure 4] Figure 4 shows a graph of the envelope spectrum of the vibration of the monitored machine. [Diagram 5] FIG. 5 is a graph showing an example of a degraded shock wave. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] An anomaly monitoring device 10 according to an embodiment of the present disclosure will be described with reference to Fig. 1 to Fig. 5. Fig. 1 shows a monitored machine 90 monitored by the anomaly monitoring device 10. The monitored machine 90 is, for example, a grinding machine, and includes a rotating shaft 91 extending in the vertical direction, a disk-shaped grindstone 92 fixed to the upper end of the rotating shaft 91, a support base 93 that rotatably supports the rotating shaft 91 at two points in the vertical direction via a pair of bearings 94, a motor 89 that is a drive source of the monitored machine 90, and a controller 88 (see Fig. 2) that controls the motor 89.

[0009] The multiple rotating parts included in the monitored machine 90, such as the motor 89, the rotating shaft 91, and the pair of bearings 94, rotate at an arbitrary number of rotations (rotational speed) according to the operation of the controller 88. Then, due to the misalignment between the center of gravity of each rotating part and the center of rotation, vibration waves (hereinafter, appropriately referred to as "eccentric waves") with a specific vibration frequency are generated for each rotating part. Furthermore, the vibration waves generated during operation of the monitored machine 90 include multiple types of waves (hereinafter, appropriately referred to as "machining distortion waves") due to the non-circularity and surface roughness at the contact points between the rotating parts. Furthermore, the vibration waves generated during operation of the monitored machine 90 include waves caused by deterioration of the rotating contact points.

[0010] Specifically, the pair of bearings 94 described above are general ball bearings of the same type, and as shown in FIG. 1, for example, are composed of an inner ring 95, an outer ring 96, a plurality of balls 97 arranged between the circumferential surfaces thereof, a holder 98 that holds the plurality of balls 97 at equal intervals, and a pair of seal rings 99 that hold grease between the inner ring 95 and the outer ring 96, and these components are in rotational contact with each other. If the inner ring 95 or the outer ring 96 has a deteriorated portion such as a scratch, a dent, or peeling, a shock wave (hereinafter, appropriately referred to as a "deterioration shock wave") is generated every time the ball 97, the holder 98, or the seal ring 99 collides with the deteriorated portion. Such deterioration shock waves are generated at intervals of a generation period T10 according to the relative rotation speed between the inner ring 95 and the outer ring 96 (hereinafter, simply referred to as the "rotation speed of the bearing 94") as conceptually shown in FIG. 5, and have a vibration frequency specific to a certain part such as the inner ring 95 or the holder 98. Incidentally, even unevenness caused by normal processing errors generates shock waves with vibration frequencies specific to each part, but these are smaller than the deterioration collision waves.

[0011] Incidentally, the position of the deteriorated part when the bearing 94 begins to deteriorate is approximately the same, and for bearings 94 of the same model, the above-mentioned collision period T10 is the same, and the natural frequency of some of the components of the bearing 94 that resonate in collisions is also the same. Therefore, if the model of the bearing 94 is known, it is possible to check the progress of deterioration in advance by checking the change in the magnitude of the deterioration shock wave.

[0012] However, as described above, the vibration waves during operation of the monitored machine 90 include various types of waves such as eccentric waves and processing distortion waves in addition to the deterioration shock waves. In the power spectrum obtained by FFT processing of the vibration waves during operation of the monitored machine 90, the information of the deterioration shock waves as exemplified in FIG. 5 does not appear prominently, and it is difficult to extract the information. Furthermore, the rotation speed of the bearing 94 is also arbitrarily changed, and the generation period T10 of the deterioration shock waves also changes accordingly. As a result, even if the information of the deterioration shock waves appears prominently in the power spectrum, it is difficult to extract the information of the deterioration shock waves from it. That is, it is difficult to monitor the abnormality of the bearing 94 of the monitored machine 90 with the conventional device. Therefore, the abnormality monitoring device 10 of this embodiment is used to monitor the abnormality of the bearing 94 of the monitored machine 90.

[0013] In order to monitor for abnormalities using anomaly monitoring device 10, an acceleration sensor 70 for detecting acceleration in a direction perpendicular to the axial direction of a rotating shaft 91 is attached to a support base 93 of a monitored machine 90. In addition, a clamp meter 71 is attached to a power line 89A of a motor 89. The detection results of the acceleration sensor 70 and the clamp meter 71 are input to the abnormality monitoring device 10.

[0014] 2, the abnormality monitoring device 10 includes a CPU 11, a memory 12, an A / D converter 13, and an amplifier 14. Then, by the CPU 11 executing a program (not shown) stored in the memory 12, the entire abnormality monitoring device 10 becomes equipped with a plurality of functional units, such as a rotation speed detection unit 20, an input control unit 21, a vibration information acquisition unit 22, a vibration frequency calculation unit 23, a basic data storage unit 24, a temporary data storage unit 25, a filter generation unit 26, a spectrum calculation unit 27, a data extraction unit 28, and an envelope processing unit 29, which are shown in the block diagram of FIG.

[0015] The rotation speed detection unit 20 is composed of an A / D converter 13 and a CPU 11, and takes in the detection result of the clamp meter 71 into the CPU 11 through the A / D converter 13. Then, based on the change in the drive current to the motor 89 detected by the clamp meter 71, the actual rotation speed, which is the actual (real) rotation speed of the motor 89 during operation, is calculated. More specifically, the motor 89 of the monitored machine 90 in this embodiment is an AC motor, and rotates at a rotation speed according to the frequency of the AC current, which is the drive current to the motor 89. Then, the rotation speed detection unit 20 calculates the frequency of the drive current from the change in the drive current detected by the clamp meter 71, and calculates the actual rotation speed of the motor 89 from the frequency. In this embodiment, the rotation speed of the bearing 94 and the rotation speed of the motor 89 are the same.

[0016] The clamp meter 71 may be attached not to the power line 89A of the motor 89 but to a control line for controlling the switching of an inverter circuit that is a drive circuit for the motor 89. Furthermore, for example, if the controller 88 outputs information on the rotation speed of the motor 89, the information on the rotation speed of the motor 89 output by the controller 88 may be input to the abnormality monitoring device 10 instead of using the clamp meter 71. Furthermore, instead of the rotation speed of the motor 89, for example, a rotation sensor such as a resolver may be provided on the rotating shaft 91, and the detection result of the rotation speed of the rotating shaft 91 may be taken into the abnormality monitoring device 10.

[0017] The input control unit 21 determines whether the monitored machine 90 is in operation based on whether the actual rotation speed calculated by the rotation speed detection unit 20 is equal to or lower than a predetermined reference value. When the monitored machine 90 is in operation, the input control unit 21 issues a data acquisition command to the vibration information acquisition unit 22 and the vibration frequency calculation unit 23, and thereafter, while the monitored machine 90 is continuously operating, issues a data acquisition command to the vibration information acquisition unit 22 and the vibration frequency calculation unit 23 every time a predetermined data acquisition period (e.g., one hour) has elapsed.

[0018] The vibration information acquisition unit 22 is composed of the amplifier 14, A / D converter 13, and CPU 11. When the vibration information acquisition unit 22 receives a data acquisition command, the amplifier 14 amplifies the detection signal of the acceleration sensor 70, and the A / D converter 13 samples the signal at a predetermined sampling frequency (e.g., 15 to 30 kHz) to acquire a plurality of acceleration data for a predetermined time period (e.g., 10 to 40 s). The vibration information acquisition unit 22 also stores the acquired plurality of acceleration data in the temporary data storage unit 25 as an acceleration data file with acquisition date and time information.

[0019] The envelope processing unit 29 performs an envelope processing on the vibration wave specified by the data of the acceleration data file to obtain an envelope processing wave. Specifically, the multiple acceleration data specify the change in acceleration over time, and the multiple acceleration data constitute the vibration data. The envelope processing unit 29 then performs an envelope processing on the vibration wave specified by the vibration data to obtain an envelope processing wave.

[0020] The spectrum calculation unit 27 performs spectrum analysis processing (FFT) on the envelope-processed wave to calculate an envelope spectrum, which is the spectrum of the envelope-processed wave.

[0021] The data extraction unit 28 extracts, from the envelope spectrum calculated by the spectrum calculation unit 27, an envelope spectrum in a limited frequency region from a lower limit frequency Fb to an upper limit frequency Ft, which is determined based on the data in the basic data storage unit 24 as follows, as data for abnormality determination.

[0022] The basic data storage unit 24 stores ratio data for specifying the ratio between the frequency of a specific wave among a plurality of types of waves included in the envelope processing wave when the bearing 94 rotates at a predetermined reference rotation speed and the reference rotation speed. Specifically, FIG. 4 shows an envelope spectrum when the bearing 94 rotates at the reference rotation speed. The envelope spectrum of the envelope processing wave includes information on a plurality of types of waves with frequencies f1 to f9. Among these waves, for example, the above-mentioned deterioration shock wave shown at frequency f6 is included. In addition, in the monitored machine 90 that is not deteriorated, the frequency of the deterioration shock wave when the bearing 94 rotates at the reference rotation speed can be known from the envelope spectrum based on the vibration wave of the monitored machine 90 using the same type of bearing 94. In the following description, even before deterioration occurs, a "deterioration shock wave" is assumed and data of a wave with the same frequency as the "deterioration shock wave" is extracted, so in this embodiment, a wave with the same frequency as the "deterioration shock wave" is referred to as a "specific wave". Then, assuming that the ratio data is K1, the reference rotation speed is N1, and the frequency of a specific wave at the reference rotation speed is f6, the ratio data K1 calculated by the following formula is stored in the basic data storage unit 24.

[0023] K1=f6 / N1...Equation 1

[0024] In addition, the above ratio data may be any data that can identify the ratio between the reference rotation speed and the frequency of the deterioration shock wave at the reference rotation speed. For example, the ratio data may be two values, the value of the reference rotation speed and the value of the frequency, or the ratio data may be the frequency of a specific vibration divided by the rotation speed of motor 89.

[0025] When the frequency calculation unit 23 receives a data acquisition command, it calculates the frequency of a specific wave when the motor 89 is rotating at the actual rotation speed as the actual special frequency. Specifically, if the actual rotation speed calculated by the rotation speed detection unit 20 is N2, the actual special frequency is Fr, and the ratio data is K1 as described above, the frequency calculation unit 23 calculates the actual special frequency Fr using the following formula 2.

[0026] Fr=K1·N2[Hz]...Equation 2

[0027] The filter generating unit 26 generates a filter by adding a preset allowable error to the real special frequency Fr. Specifically, the allowable error is stored in the basic data storage unit 24. Here, assuming that the allowable error is Fw, the upper limit frequency is Ft, and the lower limit frequency is Fb, the filter generating unit 26 calculates the upper limit frequency Ft and the lower limit frequency Fb using the following formulas 3 and 4.

[0028] Ft = Fr + Fw [Hz] Equation 3 Fb = Fr - Fw [Hz] Equation 4

[0029] Here, the upper frequency limit Ft and the lower frequency limit Fb are set to values ​​that allow only a wave of a specific frequency (frequency f6 in this embodiment) to be extracted from a plurality of waves having relatively large envelope spectrum values, such as the waves with frequencies f1 to f9 shown in Fig. 4. In other words, the "limited frequency region" refers to a region for extracting a wave of a specific frequency from a plurality of types of waves.

[0030] In this embodiment, the data extraction device 30 is configured by the rotation speed detection unit 20, the input control unit 21, the vibration information acquisition unit 22, the vibration frequency calculation unit 23, the basic data storage unit 24, the temporary data storage unit 25, the filter generation unit 26, the spectrum calculation unit 27, the data extraction unit 28, and the envelope processing unit 29 of the abnormality monitoring device 10. Even if the operating speed of the monitored machine 90 is arbitrarily changed, the data extraction device 30 can select an envelope spectrum of a certain vibration frequency necessary for determining an abnormality of the bearing 94 from multiple types of vibration and extract it as data for abnormality determination. Then, the first feature amount acquisition unit 31, the second feature amount acquisition unit 32, the multidimensional data generation unit 33, the accumulated data storage unit 34, the abnormality determination unit 35, the function update unit 36, and the like of the abnormality monitoring device 10 can use the data for abnormality determination to determine an abnormality.

[0031] The first feature amount acquiring unit 31 acquires, as the feature amount 1, a standardized peak value of the abnormality determination data (that is, a peak value of the envelope spectrum in the limited frequency region).

[0032] The second feature amount acquiring unit 32 calculates, as feature amount 2, the kurtosis in the statistical distribution of the multiple acceleration data of the acceleration data file.

[0033] The multidimensional data generation unit 33 generates two-dimensional multidimensional data having components of feature amount 1 and feature amount 2 based on the common acceleration data file, stores the data in the accumulated data storage unit 34, and deletes the acceleration data file from which the multidimensional data was generated from the temporary data storage unit 25. In this way, multiple pieces of multidimensional data are accumulated in the accumulated data storage unit 34 as the operating time of the monitored machine 90 passes.

[0034] The abnormality determination unit 35 determines whether the newly generated multidimensional data is abnormal or not. To this end, the abnormality determination unit 35 uses an abnormality determination function that is generated in advance by machine learning using OCSVM (One Class Support Vector Machine) to determine whether the new multidimensional data is abnormal or not.

[0035] Here, OCSVM is a well-known technology described in, for example, the non-patent document "Support Vector Machine" (co-authored by Ichiro Takeuchi and Masayuki Toriyama, ISBN978-4-06-152906-9), and it acquires a plurality of pieces of multi-dimensional data (which can also be called "vector data" because of its multi-dimensionality) having a plurality of feature quantities as components, and generates a function for anomaly determination, which is a vector function. At this time, it is preferable that each feature quantity is standardized based on the accumulated feature quantities before use.

[0036] The abnormality determination unit 35 inputs new multi-dimensional data as a function for abnormality determination, and if the output of the function is "negative", it outputs a determination result that the new multi-dimensional data is abnormal, i.e., "the monitored machine 90 is abnormal", and if the output of the function is not "negative", it outputs a determination result that the multi-dimensional data is normal, i.e., "the monitored machine 90 is normal". The output of the abnormality determination unit 35 is also taken into the PLC 15 (see FIG. 2) of the system including the monitored machine 90, and if it is determined that there is an abnormality, the warning light 16 is turned on.

[0037] When the anomaly monitoring device 10 is shipped together with the monitored machine 90, it is preferable to operate the monitored machine 90 for a predetermined time or more before shipping so that the anomaly judgment unit 35 is provided with a function for anomaly judgment by machine learning, but it is also possible to copy an anomaly judgment function that has been learned by machine learning by operating another monitored machine 90 of the same structure to a new anomaly monitoring device 10. Furthermore, after the anomaly monitoring device 10 and the monitored machine 90 are shipped, the monitored machine 90 may be operated for a predetermined time or more to have the anomaly judgment unit 35 learn the function for anomaly judgment by machine learning before it is used for anomaly judgment in the anomaly monitoring device 10.

[0038] The function update unit 36 ​​adds the multi-dimensional data newly stored in the accumulated data storage unit 34 to the multiple multi-dimensional data already stored in the accumulated data storage unit 34, and recalculates and updates the abnormality determination function.

[0039] The configuration of the abnormality monitoring device 10 of this embodiment has been described above. Next, the effects of the abnormality monitoring device 10 of this embodiment will be described. As described above, the data extraction device 30 of this embodiment uses the envelope spectrum, which is the spectrum of the envelope processed wave obtained by enveloping the vibration wave, as the data for abnormality determination. Therefore, compared with the conventional method of using the spectrum of the vibration wave itself as the data for abnormality determination, the data extraction device 30 is less susceptible to noise and has higher accuracy in abnormality determination. In addition, instead of using the entire envelope spectrum of the vibration wave as the data for abnormality determination, the envelope spectrum of a limited frequency range including the frequency of a specific wave among the multiple types of waves included in the envelope processed wave is extracted as the data for abnormality determination. Therefore, data not used for abnormality determination is eliminated, making abnormality determination easier. Moreover, even if the frequency of a specific wave changes with the change in the rotation speed of the bearing 94, the frequency of the specific wave is calculated from the rotation speed of the bearing 94 to determine the limited frequency range. Therefore, by using the data extraction device 30 of this embodiment, abnormality determination can be easily performed without being affected by the change in the rotation speed of the bearing 94.

[0040] Furthermore, by simply attaching a clamp meter 71 to the power line 89A of the motor 89, which is the drive source of the monitored machine 90, information on the actual rotation speed N2 of the rotational operation of the motor 89 can be acquired, so there is no need to significantly modify the monitored machine 90 in order to attach the abnormality monitoring device 10.

[0041] The abnormality determination unit 35 of the abnormality monitoring device 10 is provided with a function for abnormality determination that is machine-learned using a plurality of multi-dimensional data including the peak value of the envelope spectrum in the limited frequency region as one of the plurality of feature quantities. Then, based on the output of the function when new multi-dimensional data is input to the function, it is determined whether the monitored machine 90 is abnormal or not. This makes it easy to perform abnormality determination of the monitored machine 90 whose operating speed changes. In addition, the function update unit 36 ​​updates the function for abnormality determination using OCSVM with new multi-dimensional data, so that the accuracy of abnormality determination improves as the operating time of the monitored machine 90 becomes longer. Furthermore, the accuracy of abnormality determination can be improved by making the plurality of feature quantities include kurtosis that specifies the statistical distribution of the plurality of acceleration data measured as vibration data.

[0042] [Other embodiments] The data extraction device 30 in the above embodiment extracts, as data for abnormality determination, the envelope spectrum of one type of wave included in multiple types of vibration waves generated in the monitored machine 90 when the motor 89 is driven, but may be configured to extract, as data for abnormality determination, the envelope spectrum of some of multiple types of waves.

[0043] The abnormality monitoring device 10 of the embodiment has a function for abnormality determination generated by using OCSVM, but may have a function for abnormality determination generated by a multivariate analysis other than OCSVM. Also, abnormality determination may be performed by setting a predetermined reference value without performing multivariate analysis. Specifically, for example, an abnormality determination may be performed based on whether or not the peak value of the data for abnormality determination extracted by the data extracting device 30 is greater than a preset reference value.

[0044] The abnormality monitoring device 10 of the above embodiment performs abnormality determination using multivariate analysis (specifically, OCSVM) on multidimensional data having as its components feature quantities 1 and 2 related to the output of the monitored machine 90 (specifically, vibrations output by the monitored machine 90), but may be configured to perform abnormality determination using multivariate analysis on multidimensional data having as its components feature quantities related to both the output and input of the monitored machine 90. Specifically, abnormality determination may be performed using multivariate analysis on multidimensional data having as its components feature quantities related to the amount of power supplied to the motor 89, which is the input to the monitored machine 90, and feature quantities related to vibrations, which are the output of the monitored machine 90.

[0045] Furthermore, the multi-dimensional data in the above embodiment is two-dimensional, but it may be three or more dimensional.

[0046] The abnormality monitoring device 10 in the above embodiment judges abnormalities in the bearings 94 of the monitored machine 90, but it may also judge abnormalities in rotating parts other than bearings. Specifically, the abnormality monitoring device may judge abnormalities in rotating parts that are rotatably supported by fluid bearings or metal bearings, or rotating parts such as screws or propellers that rotate within a fluid.

[0047] In the above embodiment, a grinding machine is exemplified as the monitored machine 90, but it may be, for example, a press machine.

[0048] <Additional Notes> The following describes the feature groups extracted from the above-described embodiments, while indicating their effects, etc., as necessary. Note that, in the following, for ease of understanding, the corresponding configurations in the above-described embodiments are appropriately indicated in parentheses, etc., but these feature groups are not limited to the specific configurations indicated in parentheses, etc.

[0049] [Feature 1] A data extraction device (30) for extracting data for abnormality determination of a rotating part (94) of a monitored machine (90), comprising: an envelope processing unit (29) for performing envelope processing on a vibration wave of the rotating part (94) during rotation to obtain an envelope processed wave; a spectrum calculation unit (27) for calculating an envelope spectrum which is the spectrum of the envelope processed wave; and a basic data storage unit (24) for storing ratio data (K1) for specifying a ratio between a frequency (f6) of a specific wave among a plurality of types of waves contained in the envelope processed wave and the reference rotation speed (N1) when the rotating part (94) rotates at a predetermined reference rotation speed (N1). a frequency calculation unit (23) configured to calculate a frequency of the specific wave as a real special frequency (Fr) when the rotating part (94) rotates at the real rotation speed (N2) based on an actual rotation speed (N2) that is the actual rotation speed of the rotating part (94) and the ratio data (K1); and a data extraction unit (28) configured to extract, as the abnormality determination data, the envelope spectrum of a limited frequency region including the real special frequency (Fr) from the envelope spectrum when the rotating part (94) rotates at the actual rotation speed (N2).

[0050] [Feature 2] The data extraction device (30) according to feature 1, wherein the monitored machine (90) includes a motor (89) as a drive source, a clamp meter (71) attached to a power line (89A) or a control line of the motor (89), and a rotation speed detection unit (20) that calculates the actual rotation speed (N2) of the rotating part (94) from a change in current detected by the clamp meter (71).

[0051] [Feature 3] An abnormality monitoring device (10) comprising an abnormality determination unit (35) that acquires a plurality of multi-dimensional data including a plurality of feature amounts based on an output or input of a monitored machine (90) having a rotating part (94) and has completed machine learning of an abnormality determination function that determines whether the multi-dimensional data is abnormal or not, and determines the presence or absence of an abnormality in the monitored machine (90) based on an output of the function when new multi-dimensional data is input to the function, the abnormality monitoring device (10) having the data extraction device (30) according to feature 1 or 2, wherein the plurality of feature amounts include a peak value of the envelope spectrum in the limited frequency region.

[0052] [Feature 4] The anomaly monitoring device (10) according to feature 3, wherein the function is machine-learned using OCSVM.

[0053] [Feature 5] The abnormality monitoring device (10) according to feature 4, further comprising a function update unit (36) that updates the function using the OCSVM with the newly acquired multidimensional data.

[0054] [Feature 6] The abnormality monitoring device (10) according to any one of Features 3 to 5, wherein the plurality of feature quantities include kurtosis, standard deviation, and other statistical values ​​that specify a statistical distribution of vibration waves during rotation of the rotating part (94).

[0055] In the data extraction device of feature 1, the envelope spectrum, which is the spectrum of the envelope processed wave obtained by enveloping the vibration wave, is used as the data for abnormality judgment, so that it is less susceptible to noise and the accuracy of abnormality judgment is higher than that of the conventional device that uses the spectrum of the vibration wave itself. In addition, instead of using the entire envelope spectrum of the vibration wave as the data for abnormality judgment, the envelope spectrum of a limited frequency range including the frequency of a specific wave among the multiple types of waves included in the envelope processed wave is extracted as the data for abnormality judgment, so that data not used for abnormality judgment is eliminated and abnormality judgment is made easier. Moreover, even if the frequency of the specific wave changes with the change in the rotation speed of the rotating part, the frequency of the specific wave is calculated from the rotation speed of the rotating part to determine the limited frequency range, so that by using the data extraction device of feature 1, abnormality judgment can be easily performed without being affected by the change in the rotation speed of the rotating part.

[0056] According to feature 2, information on the actual rotation speed of a rotating part can be acquired simply by attaching a clamp meter to the power line or control line of the motor that is the drive source of the machine to be monitored.

[0057] The abnormality determination unit of the abnormality monitoring device of feature 3 includes an abnormality determination function that is machine-learned using a plurality of multi-dimensional data including the peak value of the envelope spectrum in the limited frequency region as one of the plurality of feature quantities. Then, based on the output of the function when new multi-dimensional data is input to the function, it is determined whether or not the monitored machine is abnormal. This makes it easy to perform an abnormality determination for a monitored machine whose operating speed changes.

[0058] Here, the function for determining anomaly may be one that has been machine-learned using OCSVM, as in the anomaly monitoring device of feature 4, or may be one that has been machine-learned using a multivariate analysis method other than OCSVM, or may be one that has been learned in advance by artificial intelligence equipped with a neural network.

[0059] In addition, according to feature 5, the function update unit updates the function for abnormality determination using OCSVM with new multi-dimensional data, so that the accuracy of abnormality determination improves as the operating time of the monitored machine becomes longer.

[0060] As in Feature 6, by including, among the multiple feature quantities, kurtosis, standard deviation, and other statistical values ​​that specify the statistical distribution of vibration waves during rotation of a rotating part, the accuracy of anomaly determination can be improved.

[0061] In addition, although this specification and drawings disclose specific examples of the technology included in the scope of the claims, the technology described in the claims is not limited to these specific examples, but includes various modifications and variations of the specific examples, as well as parts of the specific examples taken alone. [Explanation of symbols]

[0062] 10 Abnormality monitoring device 20 Rotational speed detector 23 Frequency calculation section 24 Basic Data Storage Unit 27 Spectral calculation section 28 Data Extraction Section 29 Envelope Processing Section 30 Data Extraction Device 34 Accumulated data storage unit 35 Abnormality determination section 36 Function Update Section 70 Acceleration Sensor 71 Clamp Meter 89 Motor 89A Power Line 90 Monitored Machines 94 Bearings (rotating parts)

Claims

1. A data extraction device that extracts data for determining an abnormality of a rotating part of a monitored machine, comprising: an envelope processing unit that performs envelope processing on the vibration wave generated during rotation of the rotating part to obtain an envelope processed wave; a spectrum calculation unit that calculates an envelope spectrum, which is the spectrum of the envelope processed wave; a basic data storage unit that stores ratio data for specifying a ratio between a frequency of a specific wave among a plurality of types of waves included in the envelope processing wave and the reference rotation speed when the rotating part rotates at a predetermined reference rotation speed; a frequency calculation unit configured to calculate a frequency of the specific wave when the rotating part rotates at the actual rotation speed based on an actual rotation speed, which is an actual rotation speed of the rotating part, and the ratio data as a real special frequency; a data extraction unit configured to extract, as the abnormality determination data, an envelope spectrum in a limited frequency region including the actual special frequency from the envelope spectrum when the rotating part rotates at the actual rotation speed; A data extraction device comprising:

2. the monitored machine includes a motor as a drive source, a clamp meter attached to a power line or a control line of the motor; 2. The data extraction device according to claim 1, further comprising a rotation speed detection unit that calculates the actual rotation speed of the rotating part from a change in current detected by the clamp meter.

3. An abnormality monitoring device comprising: an abnormality determination unit that acquires a plurality of multi-dimensional data including a plurality of feature quantities based on an output or input of a monitored machine having rotating parts as components, and that has completed machine learning of an abnormality determination function that determines whether or not the multi-dimensional data is abnormal, and that determines whether or not the monitored machine is abnormal based on an output of the function when new multi-dimensional data is input to the function, The data extraction device according to claim 1 or 2, The plurality of feature amounts include a peak value of the envelope spectrum in the limited frequency region.

4. The anomaly monitoring device according to claim 3 , wherein the function is machine-learned using OCSVM.

5. The anomaly monitoring device according to claim 4 , further comprising a function updating unit that updates the function using the OCSVM with the newly acquired multi-dimensional data.

6. 4. The abnormality monitoring device according to claim 3, wherein the plurality of feature quantities include kurtosis, standard deviation, and other statistical values ​​that specify a statistical distribution of vibration waves during rotation of the rotating part.

Images