Reliable rolling bearing early fault real-time detection method

Abstract

The invention discloses a reliable rolling bearing early fault real-time detection method. The bearing fault diagnosis method comprises the following operation steps: setting mounting positions and detection parameters of a vibration sensor and a rotating speed sensor; obtaining one-dimensional synchronous vibration data X; standardizing the one-dimensional synchronous vibration data; expanding the standardized one-dimensional synchronous vibration data into multi-dimensional synchronous data Y; decomposing the multi-dimensional synchronous data Y by adopting a zero-centralization technology to obtain multi-dimensional synchronous data Y1 and Y2; respectively calculating the mean square value of each dimension of synchronous data in the multi-dimensional synchronous data Y1, the cosine similarity of two adjacent dimensions of synchronous data and the DTW distance; respectively calculating a mean square value of each dimension of synchronous data in the multi-dimensional synchronous data Y2; constructing a fault feature matrix; a Hotelling T2 method is used to calculate a T2 value of the fault feature matrix; judging whether the bearing has a fault or not; the method is not affected by rotation speed changes, and can be applied to rolling bearing fault detection at different rotation speeds. In addition, by constructing the fault feature matrix, the defect that the bearing fault cannot be comprehensively evaluated by a single fault index is overcome, the bearing fault depicting capability is improved, the missing report rate and the false report rate of bearing early fault detection are reduced, and the robustness of the fault detection method is enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of fault detection and diagnosis of rotating machinery, and relates to a method for real-time detection of bearing faults, and particularly to a reliable method for real-time detection of early faults of rolling bearings. Background Art

[0002] Rotating machinery is a mechanical device that relies on rotational motion to achieve energy conversion, transmission, or work. At present, rotating machinery has been widely used in many fields such as industry, energy, and transportation. Common rotating machinery includes: electric motors, internal combustion engines, generators, centrifuges, etc. During long-term operation, various faults may occur in rotating machinery. Bearing faults are the most common fault forms in rotating machinery, and bearing faults often cause chain effects, leading to gear wear, shaft fracture, and even damage to the entire machine. Therefore, in order to ensure the stable operation of rotating machinery and improve its safety, it is necessary to perform real-time fault detection and regular maintenance on bearings.

[0003] At present, real-time detection technology for early bearing faults mainly focuses on fault feature extraction. Common early bearing fault features include standard deviation, correlation coefficient, various entropies, etc. For example, a Chinese invention patent (Patent No.: 2023112773207) discloses a method for diagnosing rolling bearing faults, and the extracted fault feature is the correlation coefficient.

[0004] Different fault features have different sensitivities to various bearing faults. The fault types of rolling bearings generally include rolling element faults, outer race faults, inner race faults, cage faults, etc. It is difficult for a single fault feature to sensitively characterize all the above-mentioned fault types of rolling bearings. Therefore, using a single fault feature sometimes cannot reliably identify various early faults of bearings. To solve the above problems, the present invention patent discloses a real-time detection method for early bearing faults based on Hotelling T 2 which integrates the fault information of multiple fault features, so as to be able to more reliably detect bearing faults. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide a reliable method for real-time detection of early faults of rolling bearings. The inventive method extracts multiple fault features of bearings and adopts Hotelling T 2 to integrate the fault information of multiple fault features, thereby improving the reliability of bearing fault detection.

[0006] Technical Solution: The present invention discloses a reliable method for real-time detection of early faults of rolling bearings, which specifically includes the following steps:

[0007] Step S1: Set the installation positions and detection parameters of vibration sensors and speed sensors;

[0008] Step s2: Obtain the one-dimensional synchronous vibration data X and perform normalization processing on it;

[0009] Step s3: Expand the normalized one-dimensional synchronous vibration data into multi-dimensional synchronous data Y;

[0010] Step s4: Use the zero-centering technique to decompose the multi-dimensional synchronous data Y to obtain multi-dimensional synchronous data Y1 and Y2;

[0011] Step s5: Calculate the mean square value of each dimension of synchronous data in the multi-dimensional synchronous data Y1, the cosine similarity and DTW distance between two adjacent dimensions of synchronous data respectively;

[0012] Step s6: Calculate the mean square value of each dimension of synchronous data in the multi-dimensional synchronous data Y2 respectively;

[0013] Step s7: Construct a fault feature matrix;

[0014] Step s8: Use the Hotelling T 2 method to calculate the T 2 value of the fault feature matrix;

[0015] Step s9: Determine whether the bearing has a fault.

[0016] Furthermore, in step s1, the vibration sensor is installed near the bearing to be detected to measure the bearing vibration data; the rotational speed sensor is installed near the rotating shaft to measure the rotational speed data of the rotating shaft; the detection parameters include: the number of synchronous sampling points a per revolution of the rotating shaft, the number of revolutions b of the rotating shaft, and the fault threshold T.

[0017] Furthermore, in step s2, according to the preset number of synchronous sampling points a per revolution of the rotating shaft and the number of revolutions b of the rotating shaft, using the rotational speed signal of the rotating shaft as a reference signal, obtain the synchronous vibration data during the operation of the bearing. The obtained one-dimensional synchronous vibration data can be expressed as X = {x(1), x(2), x(3)... x(ab)};

[0018] Furthermore, the normalization processing process of step s2 is as follows:

[0019] Use the Z-score normalization processing method to normalize the one-dimensional synchronous vibration data X. The normalized one-dimensional synchronous vibration data X' is as follows:

[0020]

[0021] where, represents the mean value of the one-dimensional synchronous vibration data X, and σ X represents the standard deviation of the one-dimensional synchronous vibration data X.

[0022] Further, the implementation process of step s3 is as follows:

[0023] According to the number of synchronous sampling points a per revolution of the rotating shaft and the number of revolutions b of the rotating shaft, the standardized synchronous vibration data X′ is extended to a-dimensional synchronous data Y. The extended a-dimensional synchronous data Y is as follows:

[0024]

[0025] Let Then the a-dimensional synchronous data Y can be expressed as follows:

[0026]

[0027] Further, the implementation process of step s4 is as follows: The multi-dimensional synchronous data Y is decomposed using the zero-centering technique to obtain multi-dimensional synchronous data Y1 and Y2, as follows:

[0028]

[0029] Where represents the mean value of the i-th dimension data in data Y,

[0030]

[0031] Let Then the a-dimensional synchronous data Y1 can be expressed as follows:

[0032]

[0033] Further, in step s5, the mean square values S1 = [s1(1) s1(2) s1(3)…s1(i)…s1(a)] of each dimension of synchronous data in the multi-dimensional synchronous data Y1 are calculated respectively T , where the mean square value of the i-th dimension of synchronous data is:

[0034]

[0035] Further, in step s5, the cosine similarities between adjacent dimensions of synchronous data in the multi-dimensional synchronous data Y1 are calculated respectively to obtain cosine similarity data R:

[0036] R = [R(1) R(2) R(3)…R(i)…R(a - 1)] T ,

[0037] where R(i) represents the cosine similarity between the i-th and the (i + 1)-th dimensions of synchronous data in the multi-dimensional synchronous data Y1,

[0038]

[0039] Further, in step S5, the DTW algorithm is used to calculate the DTW distances between adjacent two-dimensional synchronous data respectively, obtaining the DTW distance data D:

[0040] D = [d(1) d(2) d(3) … d(i) … d(a - 1)] T ,

[0041] where d(i) represents the DTW distance between the i-th and the (i + 1)-th dimensional synchronous data in the multi-dimensional synchronous data Y1.

[0042] Further, the implementation process of step S6 is as follows. Calculate the mean square values of each dimensional synchronous data in the multi-dimensional synchronous data Y2 respectively, S2 = [s2(1) s2(2) s2(3) … s2(i) … s2(a)] T , where the mean square value of the i-th dimensional synchronous data in the multi-dimensional synchronous data Y2 is:

[0043]

[0044] Further, the implementation process of step S7 is as follows. Take the first a - 1 mean square values of the two mean square value data S1 and S2, the data R and the data D respectively to construct the fault feature matrix C:

[0045]

[0046] Further, step S8 calculates the T 2 value of the fault feature matrix C as follows:

[0047] First, calculate the mean vector of the fault feature matrix C where; are the means of the data S1, R, D, and S2 respectively; then calculate the T 2 value:

[0048] T 2 = (a - 1)·(Z - μ0) T ∑ -1 (Z - μ0)

[0049] where a - 1 represents the number of samples of each fault feature in the fault feature matrix C,

[0050] μ0 represents the mean vector of the fault features in the healthy state of the bearing,

[0051] ∑ -1 is the inverse matrix of the covariance matrix of the fault feature matrix C.

[0052] Further, the implementation process of step S9 for judging whether there is a fault in the bearing is as follows:

[0053] Take the T calculated in step s8 2 as a fault indicator, and combine it with the fault threshold T to perform bearing fault detection; if T 2 is greater than or equal to the threshold T, it is considered that the bearing has a fault; otherwise, it is considered that the bearing is in a healthy state; at this time, return to step s2 to continue detecting the bearing health condition in the next time window.

[0054] Beneficial effects:

[0055] (1) A reliable real-time detection method for early faults of rolling bearings disclosed by the present invention overcomes the shortcoming of low reliability of existing methods. By using the statistical method Hotelling T 2 to fuse the fault information of multiple fault characteristics, various faults of the bearing can be comprehensively characterized, and the early faults of the bearing can be reliably detected.

[0056] (2) A reliable real-time detection method for early faults of rolling bearings disclosed by the present invention calculates four fault characteristics, namely, the mean square value of the multi-dimensional synchronous data Y1, the cosine similarity and DTW distance between two adjacent-dimensional synchronous data, and the mean square value of the multi-dimensional synchronous data Y2, overcomes the shortcoming that a single fault characteristic cannot comprehensively evaluate various faults of the bearing, improves the ability to characterize bearing faults, reduces the missed alarm rate and false alarm rate of early bearing fault detection, and enhances the robustness of the fault detection method.

[0057] (3) A reliable real-time detection method for early faults of rolling bearings disclosed by the present invention has low calculation cost and can detect the early faults of the bearing in real time.

[0058] (4) A reliable real-time detection method for early faults of rolling bearings disclosed by the present invention is not affected by speed changes and can be applied to the fault detection of rolling bearings at different speeds. Description of the drawings

[0059] Figure 1 is a flow chart of a reliable real-time detection method for early faults of rolling bearings disclosed by the present invention;

[0060] Figure 2 is a synchronous vibration signal diagram when the rolling bearing is healthy;

[0061] Figure 3 is a synchronous vibration signal diagram when the rolling bearing has a fault;

[0062] Figure 4 is the detection result of the method of the present invention; Detailed implementation manners

[0063] The technical solutions of the present invention will be further described in detail below with reference to the drawings.

[0064] The calculation process of a reliable real-time detection method for early faults of rolling bearings disclosed in the present invention is as follows Figure 1 shown, and specifically includes the following steps:

[0065] Step s1: Set the installation positions and detection parameters of vibration sensors and speed sensors;

[0066] The vibration sensor is installed near the bearing to be detected to measure the bearing vibration data; the speed sensor is installed at the shaft to measure the shaft speed data, such as the installation positions and methods disclosed in the invention patent (application number: CN 110186510 B); the detection parameters include: the number of synchronous sampling points a per revolution of the shaft, the number of revolutions b of the shaft, and the fault threshold T. In this example, the number of synchronous sampling points a per revolution of the shaft = 50, the number of revolutions b of the shaft = 10, and the fault threshold T = 10.

[0067] Step s2: Obtain one-dimensional synchronous vibration data X and perform normalization processing on it;

[0068] According to the preset number of synchronous sampling points a per revolution of the shaft and the number of revolutions b of the shaft, using the speed signal of the shaft as a reference signal, obtain the synchronous vibration data during the operation of the bearing. The obtained one-dimensional synchronous vibration data can be expressed as X = {x(1), x(2), x(3)... x(ab)}.

[0069] According to the set number of synchronous sampling points a = 50 per revolution of the shaft and the number of revolutions b = 10 of the shaft, the one-dimensional synchronous vibration data X can be expressed as X = {x(1), x(2), x(3)... x(500)}. Figure 2 Shows the synchronous vibration signal under the healthy state of the rolling bearing. Figure 3 Shows the synchronous vibration signal when the rolling bearing fails.

[0070] Use the Z-score normalization processing method to normalize the one-dimensional synchronous vibration data X. The normalized one-dimensional synchronous vibration data X' is as follows:

[0071]

[0072] Among them, represents the mean value of the one-dimensional synchronous vibration data X, and σ X represents the standard deviation of the one-dimensional synchronous vibration data X.

[0073] Step s3: Expand the normalized one-dimensional synchronous vibration data into multi-dimensional synchronous data Y;

[0074] According to the number of synchronous sampling points a = 50 per revolution of the shaft and the number of revolutions b = 10 of the shaft, expand the normalized synchronous vibration data X' into a-dimensional synchronous data Y. The expanded a-dimensional synchronous data Y is as follows:

[0075]

[0076] Let Then the a - dimensional synchronous data Y can be expressed as follows:

[0077]

[0078] Step s4: Decompose the multi - dimensional synchronous data Y using the zero - centering technique to obtain multi - dimensional synchronous data Y1 and Y2;

[0079] Decompose the multi - dimensional synchronous data Y using the zero - centering technique to obtain multi - dimensional synchronous data Y1 and Y2, as shown below:

[0080]

[0081] where represents the mean value of the i - th dimension data in data Y,

[0082]

[0083] Let Then the a - dimensional synchronous data Y1 can be expressed as follows:

[0084]

[0085] Step s5: Calculate the mean square value of the multi - dimensional synchronous data Y1, the cosine similarity, and the DTW distance between two adjacent - dimensional synchronous data respectively;

[0086] Calculate the mean square value S1 = [s1(1)s1(2)s1(3)…s1(i)…s1(50)] of each - dimensional synchronous data in the multi - dimensional synchronous data Y1 T , where the mean square value of the i - th - dimensional synchronous data is:

[0087]

[0088] Calculate the cosine similarity between two adjacent - dimensional synchronous data in the multi - dimensional synchronous data Y1 to obtain the cosine similarity data R:

[0089] R = [R(1)R(2)R(3)…R(i)…R(49)] T ,

[0090] where R(i) represents the cosine similarity between the i - th and the (i + 1) - th - dimensional synchronous data in the multi - dimensional synchronous data Y1,

[0091]

[0092] The DTW distances between adjacent two - dimensional synchronous data are calculated respectively using the DTW algorithm to obtain the DTW distance data D:

[0093] D = [d(1) d(2) d(3) … d(i) … d(49)] T ,

[0094] where d(i) represents the DTW distance between the i - th and (i + 1)-th synchronous data in the multi - dimensional synchronous data Y1.

[0095] Step s6: Calculate the mean square values of the multi - dimensional synchronous data Y2 respectively;

[0096] Calculate the mean square values of each synchronous data in the multi - dimensional synchronous data Y2 respectively. S2 = [s2(1) s2(2) s2(3) … s2(i) … s2(50)] T , where the mean square value of the i - th synchronous data in the multi - dimensional synchronous data Y2 is:

[0097]

[0098] Step s7: Construct a fault feature matrix;

[0099] Take the first 49 mean square values of the two mean square value data S1 and S2, the data R and the data D respectively to construct a fault feature matrix C:

[0100]

[0101] Step s8: Use the Hotelling T 2 method to calculate the T 2 value of the fault feature matrix;

[0102] The process of calculating the T 2 value of the fault feature matrix C is as follows:

[0103] First, calculate the mean vector of the fault feature matrix C where; are the means of the data S1, R, D, S2 respectively. Then calculate the T 2 value of the fault feature matrix C:

[0104] T 2 =(a - 1)·(Z - μ0) T ∑ -1 (Z - μ0)

[0105] where a - 1 represents the number of samples of each fault feature in the fault feature matrix C,

[0106] μ0 represents the mean vector of fault features in the healthy state of the bearing,

[0107] ∑ -1 is the inverse matrix of the covariance matrix of the fault feature matrix C.

[0108] In this example, μ0 is calculated based on a large number of sample data in the healthy state of the bearing, and μ0 = [0.9 - 0.94 2.41 0.09]. T .

[0109] Step s9: Determine whether the bearing has a fault.

[0110] Regard the calculated T 2 as a fault index, and combine it with the fault threshold T to perform the bearing fault detection work; if T 2 is greater than or equal to the threshold T, it is considered that the bearing has a fault; otherwise, it is considered that the bearing is in a healthy state; at this time, return to step s2 to continue detecting the health of the bearing in the next time window.

[0111] Figure 4 Shows the fault detection results of the method of the present invention. In the figure, the circles represent the fault index T 2 values when the bearing is healthy, and the squares represent the fault index T 2 values when the bearing has a fault. From Figure 4 it can be seen that when the fault index T 2 is greater than or equal to T, it is determined that the bearing has a fault, otherwise, it is determined that the bearing is in a healthy state, and at this time, return to step s2 to continue detecting the health of the bearing in the next time window.

[0112] A reliable real-time detection method for early faults of rolling bearings disclosed by the present invention first standardizes the synchronous vibration data, and expands the standardized synchronous vibration data into multi-dimensional data to facilitate extraction of fault features; then performs zero-centralized decomposition on the multi-dimensional data to obtain multi-dimensional synchronous data Y1, Y2; then calculates the mean square value of the multi-dimensional synchronous data Y1, the cosine similarity and DTW distance between two adjacent-dimensional synchronous data, and the mean square value of the multi-dimensional synchronous data Y2, constructs a fault feature matrix, overcomes the shortcoming that a single fault feature cannot comprehensively evaluate the bearing fault, improves the ability to depict faults, reduces the false negative rate and false positive rate of early bearing fault detection, and enhances the robustness of the fault detection method; finally, uses the statistical method Hotelling T 2 to extract the fault index T 2 , thereby realizing fault detection. A reliable real-time detection method for early faults of rolling bearings disclosed by the present invention has low calculation cost and can detect early faults of bearings in real time. In addition, a reliable real-time detection method for early faults of rolling bearings disclosed by the present invention is not affected by the change of rotational speed and can be applied to the fault detection of rolling bearings at different rotational speeds.

[0113] The above embodiments are only for illustrating the technical concept and features of the present invention, and the purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention cannot be limited thereby. Any equivalent transformation or modification made according to the spirit of the present invention shall be covered within the protection scope of the present invention.

Claims

1. The present invention discloses a reliable real-time detection method for early faults of rolling bearings, and the method comprises the following steps: Step s1: Set the installation positions and detection parameters of vibration sensors and speed sensors; Step s2: Obtain one-dimensional synchronous vibration data X and perform normalization processing on it; Step s3: Expand the one-dimensional synchronous vibration data after normalization processing into multi-dimensional synchronous data Y; Step s4: Decompose the multi-dimensional synchronous data Y by using the zero-centralization technique to obtain multi-dimensional synchronous data Y1 and Y2; Step s5: Calculate the mean square value of the multi-dimensional synchronous data Y1, the cosine similarity and the DTW distance between two adjacent-dimensional synchronous data respectively; Step s6: Calculate the mean square value of each-dimensional synchronous data in the multi-dimensional synchronous data Y2 respectively; Step s7: Construct a fault feature matrix; Step s8: Using the Hotelling T 2 method, calculate the T 2 value of the fault feature matrix; Step s9: Judge whether the bearing has a fault.

2. The reliable real-time detection method for early faults of a rolling bearing according to claim 1, characterized in that, In step s1, the vibration sensor is installed near the bearing to be detected to measure the bearing vibration data; the speed sensor is installed near the rotating shaft to measure the rotating shaft speed data; the detection parameters include: the number of synchronous sampling points a per revolution of the rotating shaft, the number of revolutions b of the rotating shaft, and the fault threshold T.

3. A reliable real-time detection method for early faults of rolling bearings according to claim 1, characterized in that In step s2, according to the preset number of synchronous sampling points a per revolution of the rotating shaft and the number of revolutions b of the rotating shaft, taking the speed signal of the rotating shaft as a reference signal, obtain the synchronous vibration data during the operation of the bearing, and the obtained one-dimensional synchronous vibration data can be expressed as X = {x(1), x(2), x(3)... x(ab)}; the normalization processing is to normalize the one-dimensional synchronous vibration data X by using the Z-score normalization processing method, and the normalized one-dimensional synchronous vibration data X′ is as follows: Among them, represents the mean of the one-dimensional synchronous vibration data X, and σ X represents the standard deviation of the one-dimensional synchronous vibration data X.

4. A reliable real-time detection method for early faults of rolling bearings according to claim 1, characterized in that, In step s3, according to the number of synchronous sampling points a per revolution of the rotating shaft and the number of revolutions b of the rotating shaft, expand the normalized synchronous vibration data X′ into a-dimensional synchronous data Y, and the expanded a-dimensional synchronous data Y is as follows: Let Then the a-dimensional synchronous data Y can be expressed as follows:

5. A reliable real-time early fault detection method for rolling bearings according to claim 1, characterized in that, In step s4, decompose the multi-dimensional synchronous data Y by using the zero-centralization technique to obtain multi-dimensional synchronous data Y1 and Y2, as follows: Among them, represents the mean value of the i-th row data in data Y.

6. A reliable real-time early fault detection method for rolling bearings according to claim 1, characterized in that In step s5, calculate the mean square value S1 = [s1(1) s1(2) s1(3) … s1(i) … s1(a)] of each dimension of the multi-dimensional synchronous data Y1 respectively T , where the mean square value of the i-th dimension of the synchronous data is: Calculate the cosine similarity between two adjacent-dimensional synchronous data in the multi-dimensional synchronous data Y1 respectively to obtain cosine similarity data R: R = [R(1) R(2) R(3) … R(i) … R(a - 1)] T , wherein, R(i) represents the cosine similarity between the i-th and the (i + 1)-th dimensional synchronous data in the multi-dimensional synchronous data Y1, Use the DTW algorithm to calculate the DTW distance between two adjacent-dimensional synchronous data respectively to obtain DTW distance data D: D = [d(1) d(2) d(3) … d(i) … d(a - 1)] T , wherein, d(i) represents the DTW distance between the i-th and the (i + 1)-th dimensional synchronous data in the multi-dimensional synchronous data Y1.

7. A reliable real-time early fault detection method for rolling bearings according to claim 1, characterized in that The implementation process of step s6 is as follows. Calculate the mean square value S2 = [s2(1) s2(2) s2(3) … s2(i) … s2(a)] of each dimension of synchronous data in the multi-dimensional synchronous data Y2 respectively. T , where the mean square value of the i-th dimension of synchronous data in the multi-dimensional synchronous data Y2 is:

8. A reliable real-time detection method for early faults of rolling bearings according to claim 1, characterized in that, The implementation process of step s7 is as follows. Respectively take the first a - 1 mean square value data of two mean square value data S1 and S2, data R and data D to construct a fault feature matrix C:

9. A reliable real-time early fault detection method for rolling bearings according to claim 1, characterized in that, Step s8 calculates the T value of the fault feature matrix C, and the process is as follows: 2 Value, the process is as follows: First, calculate the mean vector of the fault feature matrix C where; They are the means of data S1, R, D, and S2 respectively; then calculate the T 2 value of the fault feature matrix C: T 2 =(a - 1)·(Z - μ0) T ∑ -1 (Z - μ0) wherein, a - 1 represents the number of samples of each fault feature in the fault feature matrix C, μ0 represents the fault feature mean vector in the healthy state of the bearing, ∑ -1 is the inverse matrix of the covariance matrix of the fault feature matrix C.

10. A reliable real-time early fault detection method for rolling bearings according to claim 1, characterized in that, The implementation process of step s9 to judge whether the bearing has a fault is as follows: Take the calculated T 2 as the fault index, and combine it with the fault threshold T to perform bearing fault detection; if T 2 is greater than or equal to the threshold T, it is considered that the bearing has a fault; otherwise, it is considered that the bearing is in a healthy state; at this time, return to step s2 and continue to detect the bearing health condition of the next time window.

Images