Bearing multi-fault feature extraction method based on principal component analysis

Abstract

The invention discloses a bearing multi-fault feature extraction method based on principal component analysis. The bearing fault detection and diagnosis method based on artificial intelligence comprises the following operation steps: selecting the type of a sensor and the installation position of the sensor, and setting related parameters; collecting one-dimensional synchronous vibration data; standardizing the one-dimensional synchronous vibration data; expanding the standardized one-dimensional synchronous vibration data into multi-dimensional synchronous data; performing zero-centralization processing on the multi-dimensional synchronous data; calculating a plurality of principal component contribution rates and a cumulative contribution rate; taking the obtained multiple principal component contribution rates and the cumulative contribution rate as bearing fault features; according to the method, a plurality of fault features can be effectively extracted from the synchronous vibration data, so that the fault can be depicted in more detail; the method is not influenced by rotating speed change, and can be applied to fault feature extraction work of bearings with different rotating speeds; in addition, the method provided by the invention has relatively low calculation complexity and relatively low required calculation cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault detection and diagnosis, and relates to a bearing fault feature extraction method based on artificial intelligence, and specifically to a bearing multi-fault feature extraction method based on principal component analysis. Background Art

[0002] Bearing fault detection and diagnosis are crucial for ensuring the proper operation of mechanical equipment. This importance is particularly prominent in major rotating machinery, such as major chemical and nuclear power plants, and high-speed trains. Bearings, as core components in rotating machinery, have a direct impact on the reliability and lifespan of the entire system. According to statistics, bearing failures account for approximately 30% of rotating machinery failures, and in motor applications, this proportion can reach as high as 50%. Bearing failures not only degrade the performance of rotating machinery systems but can also cause downtime and potentially trigger chain reactions, resulting in significant economic losses and safety incidents. Therefore, timely and accurate detection and diagnosis of bearing failures are crucial.

[0003] Fault feature extraction is one of the most important aspects of bearing fault detection and diagnosis. Traditional bearing fault feature extraction methods primarily rely on signal processing techniques in the time domain, frequency domain, or time-frequency domain. With the development of artificial intelligence (AI) technology, AI-based bearing fault feature extraction methods have gradually become a research hotspot. These methods utilize machine learning or deep learning algorithms to automatically learn and extract features from large amounts of data, thereby improving the accuracy and efficiency of fault detection. However, AI-based methods typically require a large amount of historical data to train the model, while bearing fault data collected in the field is often dispersed and limited, limiting the effectiveness of model training and the quality of fault feature extraction.

[0004] Principal Component Analysis (PCA) is an efficient data dimensionality reduction technology that can convert high-dimensional and potentially redundant data into low-dimensional data while retaining the key features of the original data set as much as possible. In the field of fault detection and diagnosis, the application of PCA is mainly focused on data dimensionality reduction. By extracting the principal components that contain most of the information of the original data, the data structure is simplified and the complexity of subsequent analysis is reduced. A Chinese patent (application number: 2019107441756) discloses a bearing fault diagnosis method based on principal component analysis, which uses principal component analysis to reduce the dimensionality of the original data, and then selects the first few principal components with a certain cumulative contribution rate for fault diagnosis. While reducing the dimensionality of the original data, principal component analysis also retains the main information of the original data. Through the study of PCA, it was found that in addition to being used for data dimensionality reduction, PCA can also be directly used to extract bearing fault features, but there are relatively few studies in this area.

[0005] To address the above issues, the present invention discloses a method for extracting multiple bearing fault features based on principal component analysis (PCA), in which PCA is used to extract fault features. Currently, existing literature uses cumulative contribution rates to select the number of principal components after dimensionality reduction, but lacks research that uses both principal component contribution rates and cumulative contribution rates as multiple fault features to characterize faults. The present method innovatively uses both principal component contribution rates and cumulative contribution rates as multiple fault features to characterize bearing faults.

[0006] The principal component analysis-based bearing multi-fault feature extraction method disclosed in the present invention not only effectively extracts multiple fault features from bearing vibration data, but also requires relatively low computational overhead. Furthermore, the principal component analysis-based bearing multi-fault feature extraction method disclosed in the present invention requires fewer parameters to set, has low reliance on expert knowledge, and does not require large amounts of historical data for model training. The principal component analysis-based bearing multi-fault feature extraction method disclosed in the present invention enables efficient and accurate extraction of bearing fault features, providing strong technical support for predictive maintenance of industrial equipment. Summary of the Invention

[0007] Purpose of the invention: In order to overcome the deficiencies of the existing methods mentioned in the background technology, the present invention requests to disclose a bearing multi-fault feature extraction method based on principal component analysis, which can effectively extract multiple fault features from bearing vibration data and improve the ability to characterize bearing faults.

[0008] Technical solution: The method for extracting bearing multi-fault features based on principal component analysis described in the present invention specifically includes the following steps:

[0009] Step (1): Select the sensor type and its installation location, and set relevant parameters;

[0010] Step (2): collecting one-dimensional synchronous vibration data;

[0011] Step (3): standardize the one-dimensional synchronous vibration data;

[0012] Step (4): expanding the standardized one-dimensional synchronous vibration data into multi-dimensional synchronous data;

[0013] Step (5): Zero-center the multi-dimensional synchronous data separately;

[0014] Step (6): Calculate the contribution rates and cumulative contribution rates of multiple principal components;

[0015] Step (7): The multiple principal component contribution rates and cumulative contribution rates obtained in step (6) are used as multiple fault features of the bearing.

[0016] Furthermore, in step (1), the sensor types are vibration sensors and laser speed sensors; the laser speed sensor is installed near the rotating shaft to measure the rotating shaft speed data; the vibration sensor is installed near the bearing to be detected to measure the bearing vibration data; the relevant parameters include: the number of synchronous sampling points M per rotation of the rotating shaft, the number of rotating shaft revolutions m, the cumulative contribution rate threshold L T .

[0017] Furthermore, in step (2), based on the preset number of synchronous sampling points M per rotation of the rotating shaft and the number of rotations m of the rotating shaft, the synchronous vibration data of the bearing during operation is synchronously collected with reference to the speed signal of the rotating shaft. The collected one-dimensional synchronous vibration data can be expressed as Y = [y(1), y(2), y(3), ..., y(Mm)].

[0018] Furthermore, the implementation process of step (3) is as follows:

[0019] The one-dimensional synchronous vibration data Y is standardized using the Z-score standardization method. The standardized one-dimensional synchronous vibration data Y′ is shown below:

[0020]

[0021] in, represents the mean value of one-dimensional synchronous vibration data Y, σ Y Represents the standard deviation of the one-dimensional synchronous vibration data Y.

[0022] Furthermore, the implementation process of step (4) is as follows:

[0023] According to the number of synchronous sampling points M per shaft revolution and the shaft revolution m, the standardized synchronous vibration data Y′ is expanded into M-dimensional synchronous data X. The expanded M-dimensional data X is shown as follows:

[0024]

[0025] Where, h = 1, 2, ..., M;

[0026] To simplify writing, Then the M-dimensional synchronous data X can be expressed as follows:

[0027]

[0028] Furthermore, the implementation process of step (5) is as follows:

[0029] The zero-centering method is used to perform zero-centering on each m-dimensional vector in the M-dimensional synchronous data X. The M-dimensional synchronous data X' after zero-centering is as follows:

[0030]

[0031] in, is the average value of the h-th row of data in the M-dimensional synchronized data X.

[0032] Furthermore, the implementation process of step (6) is as follows:

[0033] Calculate the covariance matrix C of M-dimensional synchronized data as follows:

[0034]

[0035] According to the calculated covariance matrix C, calculate the eigenvalue λ i , the process is as follows:

[0036] det(C-λI)=0,

[0037] Where I is the identity matrix;

[0038] According to the eigenvalue λ i Arrange the eigenvalue array Φ in descending order and calculate the eigenvalue λ in the eigenvalue array Φ. j ′(j=1,2,3,…,a) corresponds to the contribution rate R of the principal component j and cumulative contribution rate L j , the calculation process is as follows:

[0039]

[0040] Furthermore, the implementation process of step (7) is as follows:

[0041] The cumulative contribution rate L j Greater than or equal to the cumulative contribution rate threshold L T The first b contribution rates and the first b cumulative contribution rates are regarded as the 2b fault characteristics of the bearing.

[0042] Beneficial effects

[0043] Compared with the prior art, the method of the present invention mainly includes the following beneficial effects:

[0044] (1) The bearing multi-fault feature extraction method based on principal component analysis disclosed in the patent of this invention considers the principal component contribution rate and the cumulative contribution rate as fault features, which can effectively extract multiple fault features from synchronous vibration data, thereby characterizing the bearing fault in more detail.

[0045] (2) The bearing multi-fault feature extraction method based on principal component analysis disclosed in the patent of this invention has low computational complexity and low computational cost.

[0046] (3) The bearing multi-fault feature extraction method based on principal component analysis disclosed in the patent of this invention requires fewer parameters to be set, has low dependence on artificial expert knowledge, and does not require a large amount of historical data for model training.

[0047] (4) The bearing multi-fault feature extraction method based on principal component analysis disclosed in the patent of this invention is not affected by the change of speed and can be applied to the multi-fault feature extraction of bearings with different speeds. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the method for extracting multiple bearing fault features based on principal component analysis disclosed in the present invention;

[0049] Figure 2 This is a synchronous vibration data diagram of a healthy rolling bearing in an example of the present invention;

[0050] Figure 3 This is a synchronous vibration data diagram when a rolling bearing fails in an example of the present invention;

[0051] Figure 4 This is the contribution rate extraction result when the rolling bearing is healthy in the example of the present invention;

[0052] Figure 5 The cumulative contribution rate extraction result when the rolling bearing is healthy in the example of the present invention;

[0053] Figure 6 This is the contribution rate extraction result when the rolling bearing fails in the example of the present invention;

[0054] Figure 7 This is the cumulative contribution rate extraction result when the rolling bearing fails in the example of the present invention. DETAILED DESCRIPTION

[0055] The present invention will be described in further detail below with reference to the accompanying drawings.

[0056] The present invention provides a method for extracting bearing multi-fault features based on principal component analysis, and its calculation process is as follows: Figure 1 As shown, the specific steps include:

[0057] Step (1): Select the sensor type and its installation location, and set relevant parameters;

[0058] The sensor types are vibration sensors and laser speed sensors; the laser speed sensor is installed near the shaft to measure the shaft speed data; the vibration sensor is installed near the bearing to be tested to measure the bearing vibration data; the relevant parameters include: the number of synchronous sampling points per shaft revolution M, the shaft revolution m, the cumulative contribution rate threshold L T .

[0059] In this embodiment, the number of synchronous sampling points M per shaft revolution is set to 100, the number of shaft revolutions m is set to 30, and the cumulative contribution rate threshold L is set to 100. T =80%.

[0060] Step (2): collecting one-dimensional synchronous vibration data;

[0061] According to the preset number of synchronous sampling points M per shaft revolution and the shaft revolution number m, the synchronous vibration data of the rolling bearing during operation is synchronously collected with the speed signal of the rotating shaft as a reference. The collected one-dimensional synchronous vibration data can be expressed as Y = [y(1), y(2), y(3), ..., y(Mm)].

[0062] This embodiment uses a rolling bearing simulation signal constructed in existing literature for further explanation. Based on the set number of synchronous sampling points per shaft revolution M = 100 and the shaft revolution number m = 30, the one-dimensional synchronous vibration data Y can be expressed as Y = [y(1), y(2), y(3), ..., y(3000)]. Figure 2 and 3 The synchronous vibration signals of the rolling bearing when it is healthy and faulty are shown respectively. It is difficult to identify the healthy and faulty states of the rolling bearing from the synchronous vibration signals.

[0063] Step (3): standardize the one-dimensional synchronous vibration data;

[0064] The one-dimensional synchronous vibration data Y is standardized using the Z-score standardization method. The standardized one-dimensional synchronous vibration data Y′ is shown below:

[0065]

[0066] in, represents the mean value of one-dimensional synchronous vibration data Y, σ Y Represents the standard deviation of the one-dimensional synchronous vibration data Y.

[0067] Step (4): Expanding the one-dimensional normalized synchronous vibration data into multi-dimensional synchronous data;

[0068] According to the number of synchronous sampling points per shaft revolution M = 100 and the shaft revolution number m = 30, the standardized synchronous vibration data Y′ is expanded into M-dimensional synchronous data X. The expanded M-dimensional data X is shown as follows:

[0069]

[0070] Where, h = 1, 2, ..., M;

[0071] To simplify writing, Then the M-dimensional synchronous data X can be expressed as follows:

[0072]

[0073] Step (5): Zero-center the multi-dimensional synchronous data separately;

[0074] Zero-centering processing technology is used to perform zero-centering processing on multi-dimensional synchronous data. The data after zero-centering processing can be expressed as:

[0075]

[0076] in, is the average value of the h-th row of data in the M-dimensional synchronized data X.

[0077] Step (6): Calculate the contribution rates and cumulative contribution rates of multiple principal components;

[0078] Calculate the covariance matrix C of M-dimensional synchronized data as follows:

[0079]

[0080] According to the calculated covariance matrix C, calculate the eigenvalue λ i , the process is as follows:

[0081] det(C-λI)=0,

[0082] Where I is the identity matrix;

[0083] According to the eigenvalue λ i Arrange the eigenvalue array Φ in descending order and calculate the eigenvalue λ in the eigenvalue array Φ. j ′(j=1,2,3,…,a) corresponds to the contribution rate R of the principal component j and cumulative contribution rate L j , the calculation process is as follows:

[0084]

[0085] Step (7): The multiple principal component contribution rates and cumulative contribution rates obtained in step (6) are used as bearing fault features.

[0086] The cumulative contribution rate L obtained in step (6) j Greater than or equal to the cumulative contribution rate threshold L T The first b contribution rates and the first b cumulative contribution rates are regarded as the 2b fault characteristics of the bearing.

[0087] Figure 4 、 Figure 5 、 Figure 6 and Figure 7The features extracted when the rolling bearing is healthy and faulty in the present invention are shown respectively. In this embodiment, the cumulative contribution rate threshold L T =80%, 16 principal component contribution rates and 16 cumulative contribution rates can be obtained when the rolling bearing is healthy, while 15 principal component contribution rates and 15 cumulative contribution rates can be obtained when the rolling bearing is faulty.

[0088] from Figure 4 、 Figure 5 、 Figure 6 and Figure 7 As can be seen from the figure, the method of the present invention can effectively extract multiple rolling bearing fault features. Furthermore, the method of the present invention has low computational complexity and low computational cost. We tested the computational cost of the method of the present invention using a desktop computer equipped with an Intel Core i5-8300H 2.30GHz CPU. The computational cost of extracting 30 features with the method of the present invention was approximately 0.104 seconds, meeting the requirements for real-time bearing fault detection.

[0089] Furthermore, the principal component analysis-based bearing multi-fault feature extraction method disclosed in the present invention requires fewer parameters to be set. Implementing the method requires less expert knowledge and eliminates the need for extensive historical data to train the model.

Claims

1. The present invention discloses a method for extracting bearing multi-fault features based on principal component analysis, which includes the following steps: Step (1): Select the sensor type and its installation location, and set the relevant parameters; Step (2): collecting one-dimensional synchronous vibration data; Step (3): standardize the one-dimensional synchronous vibration data; Step (4): expanding the standardized one-dimensional synchronous vibration data into multi-dimensional synchronous data; Step (5): Zero-center the multi-dimensional synchronous data separately; Step (6): Calculate the contribution rates and cumulative contribution rates of multiple principal components; Step (7): The multiple principal component contribution rates and cumulative contribution rates obtained in step (6) are used as multiple fault features of the bearing.

2. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1 is characterized in that: The sensor types described in step (1) are vibration sensors and laser speed sensors; the laser speed sensor is installed near the shaft to measure the shaft speed data; the vibration sensor is installed near the bearing to be tested to measure the bearing vibration data; the relevant parameters include: the number of synchronous sampling points M per shaft revolution, the shaft revolution m, the cumulative contribution rate threshold L T .

3. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1, characterized in that: Step (2) synchronously collects synchronous vibration data during the operation of the bearing based on the preset number of synchronous sampling points M per rotation of the shaft and the number of rotations m of the shaft, with the speed signal of the rotating shaft as a reference. The collected one-dimensional synchronous vibration data can be expressed as Y = [y(1), y(2), y(3), ..., y(Mm)].

4. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1, characterized in that: Step (3) uses the Z-score standardization method to standardize the one-dimensional synchronous vibration data Y. The standardized one-dimensional synchronous vibration data Y′ is as follows: in, Represents the mean value of one-dimensional synchronous vibration data Y, σ Y Represents the standard deviation of the one-dimensional synchronous vibration data Y.

5. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1, characterized in that: Step (4) expands the standardized synchronous vibration data Y′ into M-dimensional synchronous data X according to the number of synchronous sampling points M per shaft revolution and the shaft revolution m. The expanded M-dimensional data X is shown as follows: Where, h = 1, 2, ..., M; To simplify writing, Then the M-dimensional synchronous data X can be expressed as follows:

6. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1, characterized in that: Step (5) uses the zero-centering method to perform zero-centering on each m-dimensional vector in the M-dimensional synchronous data X. The M-dimensional synchronous data X' after zero-centering is as follows: in, is the average value of the h-th row of data in the M-dimensional synchronized data X.

7. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1, characterized in that: In step (6), the covariance matrix C of the M-dimensional synchronized data is calculated as follows: According to the calculated covariance matrix C, calculate the eigenvalue λ i , the process is as follows: det(C-λI)=0, Where I is the identity matrix; According to the eigenvalue λ i Arrange the eigenvalue array Φ in descending order and calculate the eigenvalue λ in the eigenvalue array Φ. j ′(j=1,2,3,...,a) corresponds to the contribution rate R of the principal component j and cumulative contribution rate L j , the calculation process is as follows:

8. The method for extracting bearing multi-fault features based on principal component analysis according to claim 1, characterized in that: Step (7) will be the cumulative contribution rate L j Greater than or equal to the cumulative contribution rate threshold L T The first b contribution rates and the first b cumulative contribution rates are regarded as the 2b fault characteristics of the bearing.

Images