Rolling bearing abnormality detection device and logistic regression model generation method

Abstract

An abnormality detection device for a rolling bearing is provided with a detection device body. On the basis of vibration data acquired from an acceleration sensor provided in the rolling bearing, the detection device main body calculates the overall value and the peak height of the frequency caused by damage to the rolling bearing in the frequency waveform. Furthermore, the detection device main body calculates an abnormality probability, which is a probability of occurrence of damage inside the rolling bearing, on the basis of a plurality of reference values including the overall value and the peak height, and a logistic regression model.

Description

Technical Field

[0001] This disclosure relates to an anomaly detection device for rolling bearings and a method for generating a logistic regression model. Background Technology

[0002] In conventional state monitoring methods, measurement data obtained from sensors installed on the object are filtered and multiple feature quantities are calculated. Then, machine learning is performed by using these multiple feature quantities as learning data, thereby determining an algorithm for determining whether the object is abnormal (for example, see Patent Document 1).

[0003] Patent Document 1: Japanese Patent Application Publication No. 2019-45484

[0004] In conventional condition monitoring methods, only measurement data obtained when the object is in a normal state are used as learning data for machine learning. Therefore, damage to the object may be developing at the time when an anomaly is detected. For example, in rolling bearings used at relatively low speeds, such as those in elevator traction machines, the vibration is small in the early stages of damage and cannot be identified as an anomaly; by the time an anomaly is detected, the damage may be quite severe. Summary of the Invention

[0005] This disclosure is made to solve the aforementioned problems, with the aim of providing a rolling bearing anomaly detection device and a method for generating a logistic regression model that can detect rolling bearing anomalies earlier.

[0006] The abnormality detection device for rolling bearings disclosed herein includes a detection device body. This detection device body calculates the overall value and the peak height of the frequency caused by damage to the rolling bearing in the overall value and frequency waveform based on vibration data obtained from an accelerometer installed on the rolling bearing. Based on multiple reference values ​​including the overall value and the peak height and a logistic regression model, it calculates the probability of damage occurring inside the rolling bearing, i.e., the probability of abnormality. The logistic regression model is a logistic function generated by performing logistic regression analysis using multiple indicators including the overall value and the peak height, based on learning data from multiple samples including normal data and abnormal data. The normal data is vibration data obtained by rotating multiple rolling bearings in a normal state, and the abnormal data is vibration data obtained by rotating multiple rolling bearings in a damaged state.

[0007] The method for generating the logistic regression model disclosed herein includes the following steps: obtaining training data from multiple samples including normal data and abnormal data, wherein the normal data is vibration data obtained by rotating multiple rolling bearings in a normal state, and the abnormal data is vibration data obtained by rotating multiple rolling bearings in a damaged state; binary classifying the training data into normal data and abnormal data; and selecting multiple indicators from the binary classified training data to perform logistic regression analysis, deriving a logistic function for calculating the probability of damage occurring inside the rolling bearing, i.e., the probability of abnormality, as the logistic regression model.

[0008] According to this disclosure, abnormalities in rolling bearings can be detected earlier. Attached Figure Description

[0009] Figure 1 This is a structural diagram showing the rolling bearing of Embodiment 1.

[0010] Figure 2 It means Figure 1 Block diagram of an abnormality detection device for rolling bearings.

[0011] Figure 3 It means by Figure 2 A graph of an example of vibration data obtained by the data acquisition department.

[0012] Figure 4 It means by Figure 2 A graph showing an example of the frequency waveform output by the envelope resolution section.

[0013] Figure 5 It means Figure 2 A diagram illustrating the calculation method of peak height in the peak height calculation unit.

[0014] Figure 6 It means based on Figure 2 The flowchart shows the process of determining the condition of rolling bearings performed by the main body of the detection device.

[0015] Figure 7 It means based on Figure 2 The flowchart shows the model generation process performed on the main body of the detection device.

[0016] Figure 8 This is a table representing an example of the learning data used in the model generation process of Implementation 1.

[0017] Figure 9 This is a flowchart illustrating the model generation process performed on the main body of the detection device based on Embodiment 2.

[0018] Figure 10 This is a block diagram illustrating the abnormality detection device for the rolling bearing in Embodiment 3.

[0019] Figure 11 It means based on Figure 10 The flowchart shows the model generation process performed on the main body of the detection device.

[0020] Figure 12 This is a table representing an example of the learning data used in the model generation process of Implementation Method 3.

[0021] Figure 13 This is a flowchart illustrating the model generation process performed on the main body of the detection device based on Embodiment 4.

[0022] Figure 14 This is a flowchart illustrating the model generation process performed on the main body of the detection device based on Embodiment 5.

[0023] Figure 15 This is a flowchart illustrating the model generation process performed on the main body of the detection device based on Implementation Method 6.

[0024] Figure 16 This is a block diagram showing the main parts of the abnormality detection device for rolling bearings according to Embodiment 7.

[0025] Figure 17 This is a structural diagram of the first example of the processing circuit that implements the functions of the main body of the detection device in embodiments 1 to 7.

[0026] Figure 18 This is a structural diagram of a second example showing the processing circuit that implements the functions of the main body of the detection device in embodiments 1 to 7. Detailed Implementation

[0027] The embodiments will now be described with reference to the accompanying drawings.

[0028] Implementation method 1.

[0029] Figure 1 This is a structural diagram showing the rolling bearing of Embodiment 1. In the diagram, the rolling bearing 11 has an outer ring 12, an inner ring 13, and a plurality of rolling elements 14. The outer ring 12 is assembled to a housing 15. The inner ring 13 is disposed inside the outer ring 12.

[0030] A plurality of rolling elements 14 are sandwiched between the outer ring 12 and the inner ring 13. Furthermore, the plurality of rolling elements 14 are arranged at equal intervals around each other in the circumferential direction of the inner ring 13. Balls or rollers are used as each rolling element 14.

[0031] The inner ring 13 is mounted on the rotating shaft 16. The inner ring 13 rotates together with the rotating shaft 16 with the axis of the rotating shaft 16 as the center.

[0032] Furthermore, although the rolling bearing 11 in Embodiment 1 is of the type in which the inner ring 13 rotates, it can also be of the type in which the outer ring 12 rotates.

[0033] When the inner ring 13 rotates, vibration is generated in the rolling bearing 11. An acceleration sensor 17 is provided in the housing 15. The acceleration sensor 17 generates a signal corresponding to the vibration of the rolling bearing 11.

[0034] Figure 2 It means Figure 1 A block diagram of an anomaly detection device for the rolling bearing 11. The anomaly detection device 20 has a detection device body 21 and a monitor 22.

[0035] The main body 21 of the detection device serves as a functional module, comprising a vibration data acquisition unit 23, a filtering processing unit 24, an overall value calculation unit 25, an envelope analysis unit 26, a peak height calculation unit 27, a multiplication unit 28, a model generation unit 29, a model storage unit 30, a probability calculation unit 31, a judgment unit 32, and a display unit 33.

[0036] The vibration data acquisition unit 23 acquires vibration data output from the acceleration sensor 17 when the rotating shaft 16 rotates at a constant speed.

[0037] Figure 3 This is a graph representing an example of vibration data acquired by the vibration data acquisition unit 23. Figure 2 The time variation of the vibration acceleration of the rolling bearing 11 is shown.

[0038] The filtering processing unit 24 performs filtering processing on the vibration data acquired by the vibration data acquisition unit 23 to generate noise-removed data, which is the data after noise removal. Examples of filtering processing include band-pass filtering, low-pass filtering, and high-pass filtering.

[0039] The overall value calculation unit 25 calculates the overall value based on the noise removal data generated by the filtering processing unit 24. The overall value is the sum of the magnitudes of all frequency components in the full frequency band after FFT (Fast Fourier Transform) analysis.

[0040] When the roundness of the rolling bearing 11 deteriorates due to wear or other factors, or when damage occurs inside the rolling bearing 11, the overall vibration value of the rolling bearing 11 increases. In addition, if damage occurs inside the rolling bearing 11 and a period of time passes, the corners of the damaged area become rounded and the peak value decreases, but the overall value usually remains relatively large.

[0041] The envelope analysis unit 26 performs envelope processing on the noise removal data generated by the filtering unit 24 and outputs the frequency waveform after Fourier transform.

[0042] Figure 4 This is a graph representing an example of the frequency waveform output by the envelope resolution unit 26. For example... Figure 4 As shown, when damage occurs inside the rolling bearing 11, a peak value is generated at the bearing damage frequency and at a frequency that is an integer multiple of the bearing damage frequency. The bearing damage frequency is the frequency caused by damage inside the rolling bearing 11, and is determined by the size of the rolling bearing 11 and the number of rolling elements 14, etc.

[0043] The peak height calculation unit 27 calculates the peak height of the bearing damage frequency in the output frequency waveform.

[0044] Figure 5 This is an explanatory diagram illustrating the calculation method of the peak height in the peak height calculation unit 27. When damage occurs inside the rolling bearing 11, a peak value is generated at the bearing damage frequency. Alternatively, sometimes a peak value is generated at a frequency close to the bearing damage frequency, depending on the sampling frequency.

[0045] In calculating the peak height, multiple sampling points are set at the frequencies where the peak occurs, i.e., on both the side lower than the peak frequency and the side higher than the peak frequency. Then, the lowest point (X1, Y1) is selected from the multiple sampling points on the side lower than the peak frequency. Additionally, the lowest point (X2, Y2) is also selected from the multiple sampling points on the side higher than the peak frequency.

[0046] The peak height calculation unit 27 uses the lowest point (X1, Y1) and the lowest point (X2, Y2) to calculate the height H up to the peak frequency (Xp, Yp) using any one of the following three formulas. In addition, A and B are constants.

[0047]

[0048]

[0049]

[0050] The multiplication unit 28 calculates the value obtained by multiplying the output value of the overall value calculation unit 25 and the output value of the peak height calculation unit 27, which is the value of overall value × peak height.

[0051] The model generation unit 29 performs machine learning on normal and abnormal data to generate a logistic regression model. The machine learning method is, for example, supervised learning. Details of the method for generating the logistic regression model will be described later.

[0052] The model storage unit 30 stores the logistic regression model generated by the model generation unit 29.

[0053] The probability calculation unit 31 calculates the probability of anomaly in the rolling bearing 11 based on multiple reference values ​​and a logistic regression model. The probability of anomaly in the rolling bearing 11 is the probability of damage occurring inside the rolling bearing 11, which is the object of inspection. In the probability calculation unit 31 of Embodiment 1, the overall value and the peak height are used as multiple reference values.

[0054] The determination unit 32 determines whether the rolling bearing 11 is abnormal by comparing the abnormal probability calculated by the probability calculation unit 31 with a threshold for the abnormal probability. The threshold for the abnormal probability is preset in the determination unit 32.

[0055] If the probability of an abnormality is above a threshold, the determination unit 32 determines that there is an abnormality, that is, that damage has occurred in the rolling bearing 11. Conversely, if the probability of an abnormality is below the threshold, the determination unit 32 determines that the rolling bearing 11 is in a normal state.

[0056] The display unit 33 displays the determination result of the determination unit 32 on the monitor 22. Additionally, the display unit 33 may also display the anomaly probability calculated by the probability calculation unit 31 on the monitor 22.

[0057] Figure 6 It means based on Figure 2 The flowchart shows the state determination process of the rolling bearing 11 performed by the main body 21 of the detection device. The main body 21 of the detection device performs the state determination process when the rotating shaft 16 rotates at a constant speed.

[0058] When the initial state determination process is initiated, the detection device body 21 obtains vibration data from the acceleration sensor 17 in step S101.

[0059] Next, in step S102, the detection device body 21 performs filtering processing on the acquired vibration data. After that, in step S103, the detection device body 21 calculates the overall value.

[0060] Next, in step S104, the detection device body 21 performs envelope analysis to obtain the frequency waveform after Fourier transform. Then, in step S105, the detection device body 21 calculates the peak height. Additionally, in step S106, the detection device body 21 calculates the value of the total value multiplied by the peak height.

[0061] Next, the detection device body 21 calculates the anomaly probability in step S107. Then, in step S108, the detection device body 21 compares the anomaly probability with a threshold to determine whether the rolling bearing 11 has an anomaly. Following this, in step S109, the detection device body 21 displays the determination result on the monitor 22 and ends the status determination process.

[0062] Figure 7 It means based on Figure 2The flowchart illustrates the model generation process performed by the detection device main body 21. The model generation process is the process of generating a logistic regression model. When the detection device main body 21 begins the model generation process, in step S201, it acquires learning data for supervised learning.

[0063] Supervised learning utilizes training data from multiple samples. These samples include multiple rolling bearings in normal states and multiple rolling bearings in abnormal states (i.e., damaged states). The training data also includes normal and abnormal data. Normal data consists of vibration data obtained by rotating the rolling bearings in their normal states. Abnormal data consists of vibration data obtained by rotating the rolling bearings in their damaged states.

[0064] After obtaining the learning data, in step S202, the detection device body 21 sets the normal data to "0" and the abnormal data to "1", and performs binary classification on the learning data of multiple samples.

[0065] Figure 8 This is a table representing an example of the learning data used in the model generation process of Implementation Method 1. Figure 8 The image shows the learning data after binary classification.

[0066] In the supervised learning of Implementation Method 1, for each sample, the overall value, peak height, and the value of overall value × peak height are used as input data, and the judgment result of whether it is normal is used as output data.

[0067] Next, the detection device body 21 performs logistic regression analysis in step S203. Logistic regression analysis is a statistical method that uses multiple indicators to predict the probability of a binary result occurring. In this example, the detection device body 21 selects the overall value and peak height as indicators based on the learned data after binary classification.

[0068] Then, in step S204, the detection device body 21 derives the following logic function P as a logistic regression model and ends the processing.

[0069]

[0070] z = A1 × Overall value + A2 × Peak height + A3

[0071] In addition, A1, A2, and A3 are constants, derived through logistic regression analysis.

[0072] The output of the logic function P is a value between 0 and 1. Therefore, the abnormal probability of the rolling bearing 11 can be calculated based on the values ​​of multiple reference values ​​obtained from the vibration data of the rolling bearing 11.

[0073] The method for generating the logistic regression model in Implementation 1 includes a first step, a second step, and a third step.

[0074] The first step is to obtain training data from multiple samples, including both normal and outlier data. The second step is to binary classify the training data into normal and outlier data. The third step is to select multiple indicators from the binary-classified training data to perform logistic regression analysis and derive the logistic function.

[0075] In this anomaly detection device 20, the main body 21 of the detection device acquires vibration data from the accelerometer 17 and calculates the overall value and peak height. Furthermore, the main body 21 of the detection device calculates the anomaly probability of the rolling bearing 11 based on multiple reference values, including the overall value and peak height, and a logistic regression model.

[0076] Therefore, abnormalities in the rolling bearing 11 can be detected earlier. Furthermore, even when the rolling bearing 11 is a type of rolling bearing used at relatively low speeds, such as the bearing in an elevator traction machine, abnormalities in the rolling bearing 11 can be detected earlier. This reduces elevator downtime.

[0077] Furthermore, a threshold for the probability of an anomaly is set in the main body 21 of the detection device. The main body 21 then determines whether the rolling bearing 11 is abnormal by comparing the calculated probability of an anomaly with the threshold. Therefore, anomalies in the rolling bearing 11 can be detected more quickly.

[0078] Additionally, the vibration data acquisition unit 23 acquires vibration data from the accelerometer 17. The filtering processing unit 24 generates noise-removed data by filtering the vibration data acquired by the vibration data acquisition unit 23. The overall value calculation unit 25 calculates the overall value based on the noise-removed data. The envelope analysis unit 26 performs envelope processing on the noise-removed data and outputs a frequency waveform after Fourier transform. The peak height calculation unit 27 calculates the peak height based on the frequency waveform output by the envelope analysis unit 26.

[0079] Therefore, the overall value and peak height can be easily calculated using the detection device body 21.

[0080] Furthermore, in the method for generating the logistic regression model, training data from multiple samples, including normal and abnormal data, is obtained and binary-classified into normal and abnormal data. Then, multiple indicators are selected from the binary-classified training data to perform logistic regression analysis, and a logistic function for calculating the probability of abnormality is derived as the logistic regression model.

[0081] Therefore, by using the generated logistic regression model, the abnormality of the rolling bearing 11 can be detected more quickly.

[0082] In addition, several metrics include overall values ​​and peak heights. Furthermore, the peak height is calculated based on a frequency waveform obtained by enveloping and frequency-analyzing filtered data.

[0083] Therefore, a more effective logistic regression model can be easily generated, and by using this logistic regression model, abnormalities in the rolling bearing 11 can be detected earlier.

[0084] Furthermore, in Implementation 1, the value of overall value × peak height was not selected as an indicator, so it can also be excluded from the input data of supervised learning. Additionally, in Implementation 1, the value of overall value × peak height was not used as a reference value, so the multiplication part 28 can also be omitted.

[0085] Implementation method 2.

[0086] Next, Figure 9 This is a flowchart illustrating the model generation process performed on the main body 21 of the detection device based on Embodiment 2. Figure 9 In this process, the steps S201 and S202 are handled in the same way as in implementation method 1.

[0087] In Implementation Method 2, after performing binary classification on the data of multiple samples, the detection device body 21 performs logistic regression analysis in step S205. The detection device body 21 of Implementation Method 2 selects three indicators from the binary-classified learning data: the overall value, the peak height, and the value obtained by multiplying the overall value and the peak height.

[0088] Then, in step S206, the detection device body 21 derives the following logic function P as a logistic regression model and ends the processing.

[0089]

[0090] z = A1 × Overall value + A2 × Peak height + A3 × Overall value × Peak height + A4

[0091] In addition, A1, A2, A3, and A4 are constants, derived through logistic regression analysis.

[0092] Thus, among the multiple indicators in Implementation 2 are the overall value, the peak height, and the value of overall value × peak height. Therefore, among the multiple reference values ​​in Implementation 2 are also the overall value, the peak height, and the value of overall value × peak height.

[0093] The other structures and operations in Implementation 2 are the same as in Implementation 1.

[0094] In such anomaly detection device 20 and the method for generating the logistic regression model, a value of overall value × peak height is included among multiple indicators. Therefore, for example, even if only the overall value and the peak height increase due to damage to the rolling bearing 11, anomalies in the rolling bearing 11 can be detected with higher precision.

[0095] Implementation method 3.

[0096] Next, Figure 10 This is a block diagram illustrating the abnormality detection device for the rolling bearing according to Embodiment 3. The detection device body 21, besides being similar to... Figure 2 In addition to the same structure, as a functional module, it also has an operation data acquisition unit 34 and a total revolutions calculation unit 35.

[0097] The operation data acquisition unit 34 acquires operation data related to the rolling bearing 11 from the control device 18. The operation data includes the operation time of the rolling bearing 11, the rotational speed of the rolling bearing 11, and the operation mode of the rolling bearing 11. For example, if the rolling bearing 11 is a bearing of an elevator traction machine, the control device 18 is an elevator control device that controls the elevator traction machine.

[0098] In the international standard ISO 218, such as the basic rated life L10 with a reliability of 90%, the operating data of the rolling bearing 11 is also an important factor in determining the condition of the rolling bearing 11.

[0099] The total rotation count calculation unit 35 calculates the total rotation count of the rolling bearing 11, i.e. the total rotation count of the rotating shaft 16, based on the operation data obtained by the operation data acquisition unit 34.

[0100] In Embodiment 3, the probability calculation unit 31 calculates the probability of anomaly in the rolling bearing 11 based on multiple reference values ​​and a logistic regression model. The multiple reference values ​​are output values ​​from the overall value calculation unit 25, the peak height calculation unit 27, and the operation data acquisition unit 34, respectively.

[0101] Figure 11 It means based on Figure 10 The flowchart describes the model generation process performed by the detection device main body 21. When the model generation process begins, the detection device main body 21 acquires learning data in step S301.

[0102] After obtaining the learning data, in step S302, the detection device body 21 sets the normal data to "0" and the abnormal data to "1", and performs binary classification on the data of multiple samples.

[0103] Figure 12 This is a table representing an example of the learning data used in the model generation process of Implementation Method 3. Figure 12The image shows the learning data after binary classification.

[0104] In the supervised learning of Implementation Method 3, for each sample, the overall value, peak height, the value of overall value × peak height, total number of revolutions, and running time are used as input data, and the judgment result of whether it is normal is used as output data.

[0105] Subsequently, the detection device body 21 performs logistic regression analysis in step S303. The detection device body 21 of Embodiment 3 selects three indicators—overall value, peak height, and operating time—from the learning data after binary classification.

[0106] Then, in step S304, the detection device body 21 derives the following logic function P as a logistic regression model and ends the processing.

[0107]

[0108] z = A1 × Overall value + A2 × Peak height + A3 × Operating time + A4

[0109] In addition, A1, A2, A3, and A4 are constants, derived through logistic regression analysis.

[0110] Thus, the multiple indicators in Embodiment 3 include the overall value, peak height, and operating time. Therefore, the multiple reference values ​​in Embodiment 3 also include the overall value, peak height, and operating time of the rolling bearing 11.

[0111] The other structures and operations in Implementation 3 are the same as in Implementation 1.

[0112] In this anomaly detection device 20, operating time is included among multiple indicators and multiple reference values. Therefore, anomalies in the rolling bearing 11 can be detected with higher precision.

[0113] Furthermore, the logistic regression model generation method includes operating time among multiple indicators. Therefore, by using the generated logistic regression model, anomalies in the rolling bearing 11 can be detected with higher accuracy.

[0114] Furthermore, in Embodiment 3, the total value × peak height and the total number of revolutions are not selected as indicators, so they can be excluded from the input data of supervised learning. Additionally, in Embodiment 3, the total value × peak height and the total number of revolutions are not used as reference values, so the multiplication unit 28 and the total number of revolutions calculation unit 35 can be omitted respectively.

[0115] Implementation method 4.

[0116] Next, Figure 13This is a flowchart illustrating the model generation process performed on the main body 21 of the detection device according to Embodiment 4. Figure 13 In this process, the processing of steps S301 and S302 is the same as in implementation method 3.

[0117] In embodiment 4, after performing binary classification on the data of multiple samples, the detection device body 21 performs logistic regression analysis in step S305. The detection device body 21 of embodiment 4 selects three indicators—overall value, peak height, and total revolutions—from the binary-classified learning data.

[0118] Then, in step S306, the detection device body 21 derives the following logic function P as a logistic regression model and ends the processing.

[0119]

[0120] z = A1 × Overall value + A2 × Peak height + A3 × Total revolutions + A4

[0121] In addition, A1, A2, A3, and A4 are constants, derived through logistic regression analysis.

[0122] Thus, among the various indicators in Embodiment 4 are the overall value, peak height, and total revolutions. Therefore, among the various reference values ​​in Embodiment 4 are also the overall value, peak height, and the total revolutions of the rolling bearing 11.

[0123] The other structures and operations in Implementation 4 are the same as in Implementation 3.

[0124] In this anomaly detection device 20, the total number of revolutions is included among multiple indicators and multiple reference values. Therefore, anomalies in the rolling bearing 11 can be detected with higher precision.

[0125] Furthermore, the logistic regression model generation method includes total revolutions among multiple indicators. Therefore, by using the generated logistic regression model, anomalies in the rolling bearing 11 can be detected with higher accuracy.

[0126] Furthermore, even if the rolling bearing 11 is not operated continuously, but rather in a programmed operation that includes acceleration, deceleration, and stopping, abnormalities in the rolling bearing 11 can be detected more accurately.

[0127] Furthermore, in Embodiment 4, the total value × peak height and the operating time were not selected as indicators, so they can be excluded from the input data of supervised learning. Additionally, in Embodiment 4, the total value × peak height was not used as a reference value, so the multiplication part 28 can be omitted.

[0128] Implementation method 5.

[0129] Next, Figure 14 This is a flowchart illustrating the model generation process performed on the main body 21 of the detection device based on Embodiment 5. Figure 14 In this process, the processing of steps S301 and S302 is the same as in implementation method 3.

[0130] In embodiment 5, after performing binary classification on the data of multiple samples, the detection device body 21 performs logistic regression analysis in step S307. The detection device body 21 in embodiment 5 selects four indicators from the binary-classified learning data: the overall value, the peak height, the value obtained by multiplying the overall value and the peak height, and the operating time.

[0131] Then, in step S308, the detection device body 21 derives the following logic function P as a logistic regression model and ends the processing.

[0132]

[0133] z = A1 × Overall value + A2 × Peak height + A3 × Overall value × Peak height + A4 × Operating time + A5

[0134] In addition, A1, A2, A3, A4, and A5 are constants derived through logistic regression analysis.

[0135] Thus, among the multiple indicators in Embodiment 5 are the overall value, peak height, the value of overall value × peak height, and operating time. Therefore, among the multiple reference values ​​in Embodiment 4 are also the overall value, peak height, the value of overall value × peak height, and the operating time of the rolling bearing 11.

[0136] The other structures and operations in Implementation 5 are the same as in Implementation 3.

[0137] In this anomaly detection device 20, multiple indicators and multiple reference values ​​include the overall value × peak height and the operating time. Therefore, it is possible to detect anomalies in the rolling bearing 11 with higher precision.

[0138] Furthermore, the logistic regression model generation method includes the overall value × peak height and operating time among multiple indicators. Therefore, by using the generated logistic regression model, anomalies in the rolling bearing 11 can be detected with even higher precision.

[0139] Furthermore, in Embodiment 5, the total number of revolutions is not selected as an indicator, so it can be excluded from the input data of supervised learning. Additionally, in Embodiment 5, the total number of revolutions is not used as a reference value, so the total number of revolutions calculation unit 35 can be omitted.

[0140] Implementation method 6.

[0141] Next, Figure 15 This is a flowchart illustrating the model generation process performed on the main body 21 of the detection device according to Embodiment 6. Figure 15 In this process, the processing of steps S301 and S302 is the same as in implementation method 3.

[0142] In embodiment 6, after performing binary classification on the data of multiple samples, the detection device body 21 performs logistic regression analysis in step S309. The detection device body 21 of embodiment 6 selects four indicators from the binary-classified learning data: the overall value, the peak height, the value obtained by multiplying the overall value and the peak height, and the total number of revolutions.

[0143] Then, in step S310, the detection device body 21 derives the following logic function P as a logistic regression model and ends the processing.

[0144]

[0145] z = A1 × Overall value + A2 × Peak height + A3 × Overall value × Peak height + A4 × Total revolutions + A5

[0146] In addition, A1, A2, A3, A4, and A5 are constants derived through logistic regression analysis.

[0147] Thus, among the multiple indicators in Embodiment 6 are the overall value, peak height, the value of overall value × peak height, and total number of revolutions. Therefore, among the multiple reference values ​​in Embodiment 4 are also the overall value, peak height, the value of overall value × peak height, and the total number of revolutions of the rolling bearing 11.

[0148] The other structures and operations in Implementation 6 are the same as in Implementation 3.

[0149] In this anomaly detection device 20, the total value × peak height and the total number of revolutions are included in multiple indicators and multiple reference values. Therefore, anomalies in the rolling bearing 11 can be detected with higher precision.

[0150] Furthermore, the logistic regression model generation method includes the total value multiplied by the peak height and the total number of revolutions among multiple indicators. Therefore, by using the generated logistic regression model, anomalies in the rolling bearing 11 can be detected with even higher precision.

[0151] Furthermore, in embodiments 1 to 6, the detection device body 21 can also output the abnormality probability, and the user can determine whether there is an abnormality based on the output abnormality probability.

[0152] Implementation method 7.

[0153] Next, Figure 16 This is a block diagram showing the main parts of the rolling bearing anomaly detection device according to Embodiment 7. In Embodiment 7, the model generation unit 29 generates two or more logistic regression models. Then, the two or more logistic regression models are stored in the detection device body 21. In this example, the six logistic regression models shown in Embodiments 1 to 6 are stored in the detection device body 21.

[0154] The probability calculation unit 31 uses two or more logistic regression models to calculate two or more abnormal probabilities for the same rolling bearing 11.

[0155] The determination unit 32 compares multiple abnormal probabilities with a threshold, for example, 50%, and determines whether the rolling bearing 11 is abnormal based on the number of abnormal probabilities that are above the threshold.

[0156] For example, if the number of abnormal probabilities that exceed the threshold is greater than the number of abnormal probabilities that are less than the threshold, the determination unit 32 determines that there is an abnormality, that is, it determines that damage has occurred in the rolling bearing 11.

[0157] Furthermore, if the number of abnormal probabilities that are above the threshold is less than the number of abnormal probabilities that are below the threshold, the determination unit 32 determines that the rolling bearing 11 is in a normal state.

[0158] In this example, the probability operation unit 31 has a first operation unit 31a, a second operation unit 31b, a third operation unit 31c, a fourth operation unit 31d, a fifth operation unit 31e, and a sixth operation unit 31f.

[0159] The first calculation unit 31a calculates the first anomaly probability using the same logistic regression model as in Embodiment 1. The second calculation unit 31b calculates the second anomaly probability using the same logistic regression model as in Embodiment 2. The third calculation unit 31c calculates the third anomaly probability using the same logistic regression model as in Embodiment 3.

[0160] The fourth calculation unit 31d uses the same logistic regression model as in Embodiment 4 to calculate the fourth anomaly probability. The fifth calculation unit 31e uses the same logistic regression model as in Embodiment 5 to calculate the fifth anomaly probability. The sixth calculation unit 31f uses the same logistic regression model as in Embodiment 6 to calculate the sixth anomaly probability.

[0161] The other structures and operations in Implementation 7 are the same as those in Implementations 1 to 6.

[0162] In this anomaly detection device 20, two or more anomaly probabilities are calculated, and based on the number of anomaly probabilities that exceed a threshold, it is determined whether the rolling bearing is abnormal. Therefore, anomalies in the rolling bearing 11 can be detected with higher precision.

[0163] Furthermore, in Implementation 7, the number of logistic regression models is not limited to 6, but can be 5 or less or 7 or more.

[0164] Furthermore, in Embodiment 7, the criteria for determining the presence or absence of an anomaly in the determination unit 32 are not limited to the examples described above. For instance, even if only one anomaly probability is above the threshold, it can still be determined that an anomaly exists. Alternatively, it can be determined that an anomaly exists only if all anomaly probabilities are above the threshold.

[0165] In addition, in embodiments 1 to 7, the logistic regression model can also be generated by a model generation device different from the anomaly detection device 20.

[0166] In addition, in embodiments 1 to 7, when a single device, such as an elevator traction machine, is equipped with multiple rolling bearings, the monitoring object of the abnormality detection device 20 may not be all the rolling bearings, but only any one of them.

[0167] Furthermore, in embodiments 1 to 7, the rolling bearing 11 is not limited to the rolling bearing of the elevator traction machine.

[0168] In addition, each function of the main body 21 of the detection device in embodiments 1 to 7 is implemented by the processing circuit. Figure 17 This is a structural diagram of a first example of the processing circuit that implements the functions of the main body 21 of the detection device in embodiments 1 to 7. The processing circuit 100 in the first example is dedicated hardware.

[0169] Furthermore, the processing circuit 100 may be equivalent to a single circuit, a composite circuit, a programmable processor, a parallel programmable processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. Alternatively, the various functions of the detection device body 21 may be implemented by a separate processing circuit 100, or all functions may be implemented uniformly by the processing circuit 100.

[0170] in addition, Figure 18 This is a structural diagram of a second example of the processing circuit that implements the functions of the main body 21 of the detection device in embodiments 1 to 7. The processing circuit 200 of the second example includes a processor 201 and a memory 202.

[0171] In the processing circuit 200, the functions of the detection device main body 21 are implemented through software, firmware, or a combination of software and firmware. The software and firmware are programmed and stored in the memory 202. The processor 201 implements the functions by reading and executing the programs stored in the memory 202.

[0172] The program stored in memory 202 can also be described as the process or method that causes the computer to execute the aforementioned parts. Here, memory 202 is, for example, equivalent to non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), and EEPROM (Electrically Erasable and Programmable Read Only Memory). Additionally, disks, floppy disks, optical disks, compressed optical disks, mini-disks, DVDs, etc., also correspond to memory 202.

[0173] Furthermore, some of the functions of the aforementioned components can be implemented using dedicated hardware, while others can be implemented using software or firmware.

[0174] In this way, the processing circuit can implement the functions of the above-mentioned parts through hardware, software, firmware, or a combination thereof.

[0175] Explanation of reference numerals in the attached figures

[0176] 11... Rolling bearing; 17... Accelerometer; 20... Anomaly detection device; 21... Detection device body; 23... Vibration data acquisition unit; 24... Filtering unit; 25... Overall value calculation unit; 26... Envelope analysis unit; 27... Peak height calculation unit.

Claims

1. A device for detecting abnormalities in rolling bearings, characterized in that, The device includes a detection unit that calculates the overall value and the peak height of the frequency caused by damage to the rolling bearing in the frequency waveform based on vibration data obtained from an accelerometer installed on the rolling bearing. Based on multiple reference values ​​including the overall value and the peak height, and a logistic regression model, the device calculates the probability, i.e., the probability of anomaly, that damage occurs inside the rolling bearing. The logistic regression model is a logistic function generated by performing logistic regression analysis on learning data from multiple samples including normal and abnormal data, using multiple indicators including the overall value and the peak height. The normal data is vibration data obtained by rotating multiple rolling bearings in a normal state, and the abnormal data is vibration data obtained by rotating multiple rolling bearings in a damaged state.

2. The abnormality detection device for rolling bearings according to claim 1, characterized in that, The plurality of indicators and the plurality of reference values ​​also include values ​​obtained by multiplying the overall value by the peak height.

3. The abnormality detection device for rolling bearings according to claim 1 or 2, characterized in that, The rolling bearing's operating time is also included in the plurality of indicators and the plurality of reference values.

4. The abnormality detection device for rolling bearings according to claim 1 or 2, characterized in that, The total number of revolutions of the rolling bearing is also included in the plurality of indicators and the plurality of reference values.

5. The abnormality detection device for rolling bearings according to any one of claims 1 to 4, characterized in that, The detection device body is equipped with a threshold value for the probability of the anomaly. The main body of the detection device determines whether the rolling bearing is abnormal by comparing the calculated abnormality probability with the threshold.

6. The abnormality detection device for rolling bearings according to claim 5, characterized in that, The main body of the detection device is composed of: Using two or more of the aforementioned logistic regression models, calculate two or more of the aforementioned anomaly probabilities for the same rolling bearing. Compare each of the two or more anomaly probabilities with the threshold. The rolling bearing is determined to be abnormal based on the number of abnormal probabilities that exceed the threshold.

7. The abnormality detection device for rolling bearings according to any one of claims 1 to 6, characterized in that, The main body of the detection device has: The vibration data acquisition unit acquires the vibration data from the acceleration sensor; The filtering processing unit performs filtering processing on the vibration data acquired by the vibration data acquisition unit, thereby generating noise-removed data, i.e., noise-removed data. The overall value calculation unit calculates the overall value based on the noise removal data; An envelope analysis unit performs envelope processing on the noise-removed data and outputs a frequency waveform after Fourier transform; and The peak height calculation unit calculates the peak height based on the frequency waveform.

8. A method for generating a logistic regression model, characterized in that, Includes the following steps: The step of obtaining learning data from multiple samples including normal data and abnormal data, wherein the normal data is vibration data obtained by rotating multiple rolling bearings in a normal state, and the abnormal data is vibration data obtained by rotating multiple rolling bearings in a damaged state. The step of binary classifying the learning data into normal data and abnormal data; as well as From the learning data after binary classification, multiple indicators are selected for logistic regression analysis, and a logistic function is derived to calculate the probability of damage occurring inside the rolling bearing, i.e., the probability of anomaly, as a step in the logistic regression model.

9. The method for generating a logistic regression model according to claim 8, characterized in that, Among the multiple indicators are: the overall value calculated by filtering the vibration data, and the peak height of the frequency caused by damage to the rolling bearing. The peak height is calculated based on the frequency waveform, which is obtained by performing envelope processing on the data after the filtering process and then performing frequency analysis.

10. The method for generating a logistic regression model according to claim 9, characterized in that, The plurality of indicators also includes the value obtained by multiplying the overall value by the peak height.

11. The method for generating a logistic regression model according to claim 9 or 10, characterized in that, The rolling bearing's operating time is also included among the various indicators.

12. The method for generating a logistic regression model according to claim 9 or 10, characterized in that, The total number of revolutions of the rolling bearing is also included among the multiple indicators.

Images