Bearing fault detection method based on square prediction error
Through a square prediction error-based method, bearing failure detection is performed using a single sensor, which solves the problem of multiple sensors in the prior art that increase cost and expert knowledge dependence, and achieves efficient and reliable bearing failure detection.
Patent Information
- Application Number
- CN202510521300.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-24
- Publication Date
- 2025-08-08
AI Technical Summary
The prior art requires multiple sensors in bearing failure detection, which increases detection cost and failure risk, while also having high dependence on human expert knowledge and requires a large amount of historical data to train models.
The square prediction error is used to determine bearing failure by setting up vibration sensors and speed sensors, and the one-dimensional synchronous vibration data is obtained, which is expanded into multi-dimensional data, pre-processing and principal component analysis and dimensional reduction, and calculating square prediction errors to judge bearing failures, reducing dependence on expert knowledge and historical data requirements.
Reliable bearing failure detection is realized, reducing detection cost and calculation time, suitable for bearing failure detection at different speeds, and does not require multiple sensors.
Smart Images
Figure CN120448796A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of mechanical system fault detection and diagnosis, and relates to bearing fault detection and diagnosis, and in particular to a bearing fault detection method based on squared prediction error (SPE). Background Art
[0002] Bearings, as the core supporting components of rotating machinery systems, have a health that directly determines the reliability of equipment operation and the continuity of industrial production. According to statistics, approximately 30% of rotating machinery failures are due to bearing failure. Sudden bearing failures can lead to cascading equipment damage or even accidents. Therefore, research on bearing fault detection technology is crucial in the field of machinery condition monitoring.
[0003] In recent years, with the rise of big data and artificial intelligence technologies, fault detection methods based on statistical models and machine learning have gradually attracted attention. Principal component analysis (PCA), a classic dimensionality reduction technique, effectively extracts key features from multidimensional data signals acquired by multiple sensors by mining the intrinsic characteristics of high-dimensional data, reducing reliance on prior knowledge.
[0004] Currently, principal component analysis is primarily used to fuse data collected by multiple vibration sensors for fault detection. While using multiple sensors can provide more comprehensive fault information, it increases fault detection costs. Furthermore, using multiple sensors to collect signals increases the risk of sensor failure.
[0005] To extract more fault information from signals collected by a single vibration sensor, this paper discloses a bearing fault detection method based on squared prediction error. This method extracts bearing fault characteristics by quantifying the Euclidean distance between observed data and reconstructed data from principal component analysis. This method requires fewer detection parameters, has low reliance on human expert knowledge, and eliminates the need for extensive historical data for model training. Summary of the Invention
[0006] Purpose of the invention: In order to overcome the shortcomings of the existing methods described in the background technology, the present invention discloses a bearing fault detection method based on squared prediction error, which can detect bearing faults in real time and reliably.
[0007] Technical solution: The bearing fault detection method based on squared prediction error described in the present invention specifically includes the following steps:
[0008] Step (1): setting the installation positions and relevant detection parameters of the vibration sensor and the speed sensor;
[0009] Step (2): obtaining one-dimensional synchronous vibration data;
[0010] Step (3): Expanding the one-dimensional synchronous vibration data into multi-dimensional synchronous data;
[0011] Step (4): preprocessing the multi-dimensional synchronous data;
[0012] Step (5): Use principal component analysis to reduce the dimensionality of the pre-processed multi-dimensional synchronous data, and project the reduced dimensionality data back to the original space for data reconstruction;
[0013] Step (6): Calculate the squared prediction error;
[0014] Step (7): Determine whether a fault occurs.
[0015] Furthermore, in step (1), the vibration sensor is installed near the bearing to be detected to measure the vibration data of the bearing; 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 N per rotation of the rotating shaft, the number of rotating shaft rotations n, the cumulative contribution rate threshold T, and the fault threshold r.
[0016] Furthermore, in step (2), according to the preset number of synchronous sampling points N per shaft revolution and the shaft revolution number n, the shaft speed signal is used as a reference signal to obtain one-dimensional synchronous vibration data. The obtained one-dimensional synchronous vibration data can be expressed as:
[0017] V=[v(1),v(2),v(3),...,v(Nn)].
[0018] Furthermore, the implementation process of step (3) is as follows:
[0019] According to the number of synchronous sampling points N per rotation of the shaft, the one-dimensional synchronous vibration data V is expanded into N-dimensional data V1. The expanded N-dimensional data V1 is as follows:
[0020]
[0021] Wherein, g=1, 2, ..., N.
[0022] Furthermore, the implementation process of step (4) is as follows:
[0023] The multi-dimensional synchronous data V1 is preprocessed, and each dimensional vector in the N-dimensional data V1 is zero-centered using the zero-centering method. The N-dimensional data V2 after zero-centering is as follows:
[0024]
[0025] in, is the average value of the g-th row of data in the N-dimensional data V1.
[0026] Furthermore, the implementation process of step (5) is as follows:
[0027] First, principal component analysis is used to reduce the dimension of the preprocessed N-dimensional synchronous data V2, and the first a principal components whose cumulative contribution rate is greater than or equal to the cumulative contribution rate threshold T are selected, which is expressed as follows:
[0028] X=[x1,x2,x3,...,xa];
[0029] The reduced-dimensional data X is then projected back to the original space for data reconstruction. The calculation process is as follows:
[0030] X1=V2XX T .
[0031] Furthermore, the implementation process of calculating the squared prediction error in step (6) is as follows:
[0032] SPE=||V2-X1|| 2 .
[0033] Furthermore, the implementation process of step (7) is as follows:
[0034] The calculated square prediction error SPE is regarded as a fault feature and combined with the set threshold r to determine whether a bearing fault occurs; if the square prediction error SPE is greater than or equal to the set threshold r, the bearing is considered to be faulty; otherwise, the bearing is considered to be in a healthy state and the process returns to step (2) to continue fault detection.
[0035] Beneficial effects
[0036] Compared with the prior art, the method of the present invention mainly includes the following beneficial effects: (1) The bearing fault detection method based on square prediction error disclosed in the patent of the present invention can reliably detect bearing faults; (2) The bearing fault detection method based on square prediction error disclosed in the patent of the present invention requires fewer detection parameters to be set, less manual expert knowledge, and no need for a large amount of historical data to train the model; (3) The bearing fault detection method based on square prediction error disclosed in the patent of the present invention uses SPE as a fault feature, with low computing time cost, and can detect whether a bearing fault occurs in a shorter time; (4) The bearing fault detection method based on square prediction error disclosed in the patent of the present invention is not affected by changes in speed and can be applied to bearing fault detection at different speeds. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is a flow chart of the bearing fault detection method based on squared prediction error disclosed by the present invention;
[0038] Figure 2 This is a synchronous vibration data diagram of a healthy rolling bearing in an example of the present invention;
[0039] Figure 3 This is a synchronous vibration data diagram when a rolling bearing fails in an example of the present invention;
[0040] Figure 4 It is the detection result of the method of the present invention. DETAILED DESCRIPTION
[0041] The present invention will be described in further detail below with reference to the accompanying drawings.
[0042] The present invention discloses a bearing fault detection method based on squared prediction error, as shown in the following figure: Figure 1 As shown, the specific steps include:
[0043] Step (1): setting the installation positions and relevant detection parameters of the vibration sensor and the speed sensor;
[0044] The vibration sensor is installed near the bearing to be tested to measure the vibration data of the bearing; 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 N per rotation of the rotating shaft, the number of rotating shaft rotations n, the cumulative contribution rate threshold T, and the fault threshold r.
[0045] In this embodiment, the number of synchronous sampling points N per shaft revolution is set to 100, the number of shaft revolutions n is set to 30, the cumulative contribution rate threshold T=90%, and the fault threshold r=105.45.
[0046] Step (2): obtaining one-dimensional synchronous vibration data;
[0047] According to the preset number of synchronous sampling points N per shaft revolution and the shaft revolution number n, the shaft speed signal is used as the reference signal to obtain one-dimensional synchronous vibration data. The obtained one-dimensional synchronous vibration data can be expressed as:
[0048] V=[v(1),v(2),v(3),...,v(Nn)].
[0049] This embodiment uses the rolling bearing simulation signal constructed in the existing literature for further explanation. According to the set number of synchronous sampling points per shaft revolution N = 100 and the shaft revolution number n = 30, the one-dimensional synchronous vibration data can be expressed as:
[0050] V=[v(1),v(2),v(3),...,v(3000)];
[0051] Figure 2 and 3 Synchronous vibration data of rolling bearings when healthy and faulty are shown respectively.
[0052] Step (3): Expanding the one-dimensional synchronous vibration data into multi-dimensional synchronous data;
[0053] According to the number of synchronous sampling points N per rotation of the shaft, the one-dimensional synchronous vibration data V is expanded into N-dimensional data V1. The expanded N-dimensional data V1 is as follows:
[0054]
[0055] Where g = 1, 2, ..., 100.
[0056] Step (4): preprocessing the multi-dimensional synchronous data;
[0057] The multi-dimensional synchronous data V1 is preprocessed, and each dimensional vector in the N-dimensional data V1 is zero-centered using the zero-centering method. The N-dimensional data V2 after zero-centering is as follows:
[0058]
[0059] in, is the average value of the g-th row of data in the N-dimensional data V1.
[0060] Step (5): Use principal component analysis to reduce the dimensionality of the pre-processed multi-dimensional synchronous data, and project the reduced dimensionality data back to the original space for data reconstruction;
[0061] First, principal component analysis is used to reduce the dimension of the preprocessed N-dimensional synchronous data V2, and the first a principal components whose cumulative contribution rate is greater than or equal to the cumulative contribution rate threshold T are selected, which is expressed as follows:
[0062] X=[x1,x2,x3,...,xa];
[0063] The reduced-dimensional data X is then projected back to the original space for data reconstruction. The calculation process is as follows:
[0064] X1=V2XX T .
[0065] Furthermore, the implementation process of calculating the squared prediction error in step (6) is as follows:
[0066] SPE=||V2-X1|| 2 .
[0067] Step (7): Determine whether a bearing fault occurs.
[0068] The calculated square prediction error SPE is regarded as a fault feature and combined with the set threshold r to determine whether a bearing fault occurs; if the square prediction error SPE is greater than or equal to the set threshold r, the bearing is considered to be faulty; otherwise, it is considered to be in a healthy state and returns to step (2) to continue fault detection.
[0069] Figure 4 The figure shows the fault detection results of the method of the present invention. The squares in the figure represent the SPE when the bearing is healthy, and the hollow circles represent the SPE when the bearing is faulty. Figure 4 As can be seen from the figure, there is a clear difference in the SPE between the healthy and faulty states of the bearing. Therefore, in this embodiment, the threshold r is set to 105.45 according to the 3σ rule. When the SPE is greater than or equal to 105.45, the bearing is considered to be faulty. Otherwise, the bearing is considered to be healthy, and the process returns to step (2) to continue fault detection.
[0070] The bearing fault detection method based on squared prediction error disclosed in the present invention not only has high detection accuracy and reliability, but also has relatively low computational cost, meeting the requirements for real-time bearing fault detection. In this embodiment, the computational cost of the method of the present invention was tested on a desktop computer using an Intel Core i5-8300H 2.30GHz CPU. The computational cost of the method of the present invention was approximately 0.019 seconds, meeting the requirements for real-time bearing fault detection. Furthermore, the bearing fault detection method based on squared prediction error disclosed in the present invention requires fewer detection parameters, requiring less expert knowledge to implement, and eliminating the need for extensive historical data to train the model.
Claims
1. The bearing fault detection method based on squared prediction error of the present invention specifically comprises the following steps: Step (1): setting the installation positions and relevant detection parameters of the vibration sensor and the speed sensor; Step (2): obtaining one-dimensional synchronous vibration data; Step (3): Expanding the one-dimensional synchronous vibration data into multi-dimensional synchronous data; Step (4): preprocessing the multi-dimensional synchronous data; Step (5): Use principal component analysis to reduce the dimensionality of the pre-processed multi-dimensional synchronous data, and project the reduced dimensionality data back to the original space for data reconstruction; Step (6): Calculate the squared prediction error; Step (7): Determine whether a fault occurs.
2. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: In step (1), the vibration sensor is installed near the bearing to be tested to measure the vibration data of the bearing; the speed sensor is installed near the rotating shaft to measure the speed data of the rotating shaft; the detection parameters include: the number of synchronous sampling points N per rotation of the rotating shaft, the number of rotating shaft rotations n, the cumulative contribution rate threshold T, and the fault threshold r.
3. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: In step (2), according to the preset number of synchronous sampling points N per shaft revolution and the shaft revolution number n, the shaft speed signal is used as a reference signal to obtain one-dimensional synchronous vibration data. The obtained one-dimensional synchronous vibration data can be expressed as: V=[v(1),v(2),v(3),...,v(Nn)].
4. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: In step (3), the one-dimensional synchronous vibration data V is expanded into N-dimensional data V1 according to the number of synchronous sampling points N per rotation of the shaft. The expanded N-dimensional data V1 is as follows: Wherein, g=1, 2, ..., N.
5. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: In step (4), the multi-dimensional synchronous data V1 is preprocessed, and each dimensional vector in the N-dimensional data V1 is zero-centered using the zero-centering method. The N-dimensional data V2 after zero-centering is as follows: in, is the average value of the g-th row of data in the N-dimensional data V1.
6. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: In step (5), the pre-processed N-dimensional synchronous data V2 is first reduced in dimension using principal component analysis, and the first a principal components whose cumulative contribution rate is greater than or equal to the cumulative contribution rate threshold T are selected, which is expressed as follows: X=[x1,x2,x3,...,xa]; The reduced-dimensional data X is then projected back to the original space for data reconstruction. The calculation process is as follows: X1=V2XX T 。 7. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: In step (6), the implementation process of calculating the squared prediction error is as follows: SPE=||V2-X1|| 2 。 8. The bearing fault detection method based on squared prediction error according to claim 1, characterized in that: The calculated square prediction error SPE is regarded as a fault feature and combined with the set threshold r to determine whether a bearing fault occurs; if the square prediction error SPE is greater than or equal to the set threshold r, the bearing is considered to be faulty; otherwise, the bearing is considered to be in a healthy state and the process returns to step (2) to continue fault detection.