Method for bearing fault recognition based on signal decomposition and one-dimensional local ternary pattern

Abstract

The application provides a bearing fault recognition method based on signal decomposition and one-dimensional local ternary mode, comprising: collecting a vibration signal and pre-processing to obtain a reconstructed signal; determining the optimal parameter combination of the mode number and the filter length according to the complexity of the mode component after feature mode decomposition, and obtaining the mode component under the optimal parameter combination; taking the absolute value of the difference between the average value of the global amplitude and the average value of the signal amplitude in the window as the quantization threshold, carrying out one-dimensional local ternary mode operation on each mode component under the optimal parameter combination to obtain the corresponding texture feature signal; screening the texture feature signal according to the impact feature strength of the texture feature signal to obtain the texture feature signal containing rich fault information; and performing spectral analysis on the screened texture feature signal to realize bearing fault recognition. The bearing fault recognition method can extract bearing fault feature information and accurately recognize the bearing state and the fault type.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, and specifically relates to a bearing fault identification method based on signal decomposition and one-dimensional local ternary mode. Background Technology

[0002] Bearings are widely used in various types of rotating machinery due to their compact structure, low frictional resistance, and high operational precision. However, rotating machinery often operates in harsh environments and needs to withstand significant load fluctuations, making bearings susceptible to damage such as wear, fatigue spalling, and cracks. This damage not only reduces bearing performance but can also lead to equipment failure, resulting in significant economic losses and safety hazards. Therefore, accurately extracting the fault characteristics of bearings is crucial for preventing equipment failures, extending service life, and ensuring stable equipment operation.

[0003] To achieve accurate bearing fault identification, scholars both domestically and internationally have conducted extensive research. Data-driven methods, with their advantages in signal analysis, feature extraction, and historical data inference, have become a research hotspot in the field of bearing fault diagnosis. However, bearing fault vibration signals exhibit strong nonlinearity and non-stationarity, with abundant redundant information in the original signals, making direct analysis difficult to achieve ideal results. Decomposing the fault signal using signal decomposition algorithms and then performing fault identification based on the decomposed component signals can more effectively obtain feature information. Existing decomposition methods still have shortcomings in practical applications: some methods require preset parameters, resulting in limited noise resistance and adaptability; key parameters in traditional eigenmode decomposition methods often rely on empirical settings, and parameter uncertainty directly affects the decomposition results.

[0004] In recent years, texture feature extraction methods from the field of image processing have been introduced into signal processing. Methods such as one-dimensional local binary mode and one-dimensional local ternary mode can represent signal information from the perspective of texture analysis. However, existing one-dimensional local ternary mode methods mostly use the local mean of the signal within the window as the criterion to directly quantize the original signal, without fully considering the overall characteristics of the fault signal. The mining of fault information is not comprehensive enough, and it is easily affected by noise. Therefore, it still has limitations in the accurate identification of bearing faults.

[0005] Therefore, proposing a bearing fault identification method with strong adaptability, good noise resistance, and accurate fault feature extraction to meet the engineering requirements of efficient bearing fault identification has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, the present invention proposes a bearing fault identification method based on signal decomposition and one-dimensional local ternary mode to solve the problems existing in the prior art.

[0007] The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode provided by this invention includes:

[0008] S1: Collect vibration signals during the operation of the bearing equipment. , ,in, The length of the vibration signal;

[0009] S2: Based on the vibration signal Constructing the Hankel matrix The Hankel matrix is ​​calculated through relevant operations. covariance matrix Based on the covariance matrix Constructing reconstructed signals ;

[0010] S3: Within a preset optimization range of mode number and filter length, the reconstructed signal is... Perform eigenmode decomposition and trial calculations, and determine the optimal combination of mode number and filter length based on the complexity of the optimized trial calculation mode components obtained after decomposition. The smaller the complexity of the optimized trial calculation mode components, the better the corresponding combination of mode number and filter length parameters.

[0011] S4: The reconstructed signal is processed using the optimal combination of the mode number and filter length. Perform eigenmode decomposition to obtain the modal components corresponding to the optimal parameter combination;

[0012] S5: Using the absolute value of the difference between the average global amplitude and the average signal amplitude within the window as the quantization threshold, perform one-dimensional local ternary mode operation on each modal component under the optimal parameter combination to obtain the corresponding texture feature signal.

[0013] S6: The texture feature signal is filtered according to the impact feature intensity of the texture feature signal to obtain a texture feature signal containing rich fault information;

[0014] S7: Perform spectral analysis on the texture feature signal containing rich fault information and identify bearing faults based on the relationship between the prominent frequency components in the spectrum and the bearing fault feature frequencies.

[0015] Preferably, in S3, the steps for determining the optimal combination of modes and filter length are as follows:

[0016] S31: Set the optimization range for the number of modes and the filter length;

[0017] S32: Under different combinations of mode number and filter length parameters, respectively, the reconstructed signal is processed. Perform eigenmode decomposition to obtain the optimization trial modal components corresponding to each set of parameter combinations;

[0018] S33: Analyze the complexity of the optimization trial modal components corresponding to each parameter combination, and determine the optimal parameter combination as the combination of the number of modes and the filter length parameter corresponding to the optimization trial modal components with the minimum complexity.

[0019] Further optimization involves using fractional permutation entropy as an evaluation index for the complexity of the optimization trial modal component in S3, and taking the number of modes and filter length corresponding to the optimization trial modal component with the minimum fractional permutation entropy as the optimal parameter combination of the number of modes and filter length.

[0020] Further optimization, in S5, the first parameter combination under optimal parameter combination Modal components Corresponding texture feature signal The generation method is as follows:

[0021] S51: Set the parameters for the sliding window;

[0022] S52: Calculate the modal components The average global magnitude ;

[0023] S53: Utilize the sliding window in the modal components The signal sequence is traversed point by point with a step size of 1, and the local texture feature value of the signal within each window is calculated. The calculation method of the local texture feature value of the signal within the window is as follows:

[0024] S531: Calculate the average value of the signal amplitude within the window;

[0025] S532: Calculate the difference between the amplitude of each sampling point in the window signal and the average amplitude of the window signal;

[0026] S533: Using the absolute value of the difference between the average global amplitude and the average amplitude of the signal within the window as the quantization threshold, the difference between the amplitude of each sampling point in the signal within the window and the average amplitude of the signal within the window is encoded to generate an encoded sequence;

[0027] S534: Convert the encoded sequence into decimal feature values, which are used as local texture feature values ​​of the signal within the window;

[0028] S54: Arrange the decimal feature values ​​corresponding to all signals within the window in the order in which the sliding window slides to obtain the texture feature signal. .

[0029] Further optimization, in S533, the first The first signal in the window Encoding the difference between the amplitude of each sampling point and the average amplitude of the signal within the window as follows:

[0030] ;

[0031] in, Represents the first under the optimal parameter combination Modal components The average global amplitude, Indicates the first The average value of the signal amplitude within each window , Indicates the length of the sliding window. Represents the modal components The signal length; Indicates the first The first signal in the window sampling points The amplitude and the average value of the signal amplitude within the window The difference, .

[0032] Further optimization involves using the squared envelope Gini index as an evaluation index for the impact characteristic intensity of the texture feature signal in S6.

[0033] Further optimization, the specific steps of S6 are as follows:

[0034] S61: Perform Hilbert transform on each texture feature signal to extract the envelope signal corresponding to each signal;

[0035] S62: For the envelope signal of each texture feature signal, calculate its squared envelope Gini index;

[0036] S63: Calculate the squared envelope Gini index of all texture feature signals, exclude the signal corresponding to the maximum value, and select the texture feature signal corresponding to the second largest squared envelope Gini index as the final texture feature signal containing rich fault information for fault identification.

[0037] This invention provides a bearing fault identification method based on signal decomposition and a one-dimensional local ternary mode. It constructs a Hankel matrix from vibration signals and reconstructs the signal by calculating the covariance matrix through correlation operations, thereby enhancing the periodicity of fault characteristics and suppressing random noise. By adaptively determining the optimal parameter combination for feature mode decomposition, it eliminates dependence on empirical parameters and improves decomposition stability and signal adaptability. By using an improved one-dimensional local ternary mode that takes the absolute value of the difference between the global and window-internal signal amplitude averages as a threshold to process modal components, it can accurately capture fault texture features in noisy environments. By filtering texture feature signals containing rich fault information through impact feature intensity, it can comprehensively extract composite fault information. By performing spectral analysis on texture feature signals containing rich fault information and matching fault feature frequencies, it can achieve accurate bearing fault identification. Furthermore, the entire process has low engineering implementation costs and strong practicality, providing reliable support for the safe operation of rotating machinery. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 The flowchart shows the bearing fault identification method based on signal decomposition and one-dimensional local ternary mode provided by the present invention.

[0040] Figure 2 This is a time-domain diagram of the original vibration signal of the rolling bearing device in Example 1;

[0041] Figure 3 The spectrum of the original vibration signal of the rolling bearing equipment in Example 1;

[0042] Figure 4 The power spectrum of the original vibration signal of the rolling bearing equipment in Example 1;

[0043] Figure 5 The reconstructed signal in Example 1 The time-domain plot;

[0044] Figure 6 The reconstructed signal in Example 1 The power spectrum;

[0045] Figure 7 Processing the reconstructed signal in Example 1 At that time, the entropy values ​​of the fractional order arrangement of each modal component under different modal numbers and filter lengths are plotted.

[0046] Figure 8 For the filter length 60, number of modes When the value is 5, the reconstructed signal in Example 1 The five modal component maps obtained by performing eigenmode decomposition;

[0047] Figure 9 To Figure 8 The five modal components in the image are implemented in a one-dimensional local ternary mode to obtain the texture feature signal maps;

[0048] Figure 10 for Figure 9 The power spectrum of each texture feature signal in the middle;

[0049] Figure 11 for Figure 9 The values ​​of the squared envelope Gini index of each texture feature signal in the text.

[0050] Figure 12 for Figure 8 Mid-modal components A magnified view of a portion of the image;

[0051] Figure 13 for Figure 9 Medium texture feature signal A magnified view of a portion of the image;

[0052] Figure 14 for Figure 9 Medium texture feature signal The power spectrum. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, to avoid obscuring the invention with unnecessary details, only processing steps closely related to the solution of this invention are shown in the drawings, while other details not closely related to this invention are omitted.

[0054] like Figure 1 As shown, this invention provides a bearing fault identification method based on signal decomposition and one-dimensional local ternary mode, comprising the following steps:

[0055] S1: Collect vibration signals during the operation of the bearing equipment. , ,in, The length of the vibration signal;

[0056] S2: Based on the vibration signal Constructing the Hankel matrix The Hankel matrix is ​​calculated through relevant operations. covariance matrix Based on the covariance matrix Constructing reconstructed signals ;

[0057] Among them, the vibration signal Constructing the Hankel matrix as follows:

[0058] ;

[0059] In the formula, , (i=1, 2, ..., j=1,2,… );

[0060] In practical applications, an n-order Hankel matrix is ​​typically constructed with a delay step size of 1. :

[0061] ;

[0062] Where, when the Hankel matrix for When, the covariance matrix as follows:

[0063] ;

[0064] Preferably, based on the covariance matrix The first row and last column construct the reconstructed signal , Specifically as follows:

[0065] ;

[0066] S3: Within a preset optimization range of mode number and filter length, the reconstructed signal is... Perform eigenmode decomposition and trial calculations, and determine the optimal combination of mode number and filter length based on the complexity of the optimized trial calculation mode components obtained after decomposition. The smaller the complexity of the optimized trial calculation mode components, the better the corresponding combination of mode number and filter length parameters.

[0067] The specific steps are as follows:

[0068] S31: Set the optimization range for the number of modes and the filter length, where [2,9] is the preferred optimization range for the number of modes, and the preferred optimization step size is 1; [10,120] is the preferred optimization range for the filter length, and the preferred optimization step size is 10.

[0069] S32: Under different combinations of mode number and filter length parameters, respectively, the reconstructed signal is processed. Perform eigenmode decomposition to obtain the optimization trial modal components corresponding to each parameter combination. Among them, the optimization trial calculation modal components corresponding to each set of parameter combinations The number of modes is equal to the number of modes under the corresponding parameter combination;

[0070] S33: Analyze the complexity of the optimization trial modal components corresponding to each parameter combination and select the optimization trial modal components with the minimum complexity. The corresponding combination of mode number and filter length parameter is determined to be the optimal parameter combination;

[0071] Preferably, fractional permutation entropy is used as the optimization trial calculation modal component. The evaluation metric for complexity is to find the optimal trial-and-error modal components that minimize the entropy of the fractional order permutation. The corresponding number of modes and filter length are used as the optimal parameter combination of the number of modes and filter length;

[0072] Among them, the optimization trial calculation of modal components The entropy of a fractional permutation is calculated as follows:

[0073] S331: For the optimization trial calculation modal components Phase space reconstruction is performed to obtain The formula is as follows:

[0074] ;

[0075] in, Satisfy the permutation pattern ,have Possible arrangements;

[0076] S332: Calculate the optimization trial modal components. The entropy of a fractional permutation is given by the following formula:

[0077] ;

[0078] The fractional derivative is expressed as follows:

[0079] ;

[0080] in, The order is the fractional order, and its value range is... , As an embedding dimension, in this invention, the following is selected: , For length is There are a total of arrangement patterns. Possible arrangements; Arrangement mode The probability of occurrence is calculated using the following formula: ,in, Represented as Corresponding arrangement pattern , Representing each information unit, Indicates information unit conduct Fractional derivative operations Represents the gamma function. Represents the digamma function;

[0081] S4: The reconstructed signal is processed using the optimal combination of the mode number and filter length. Perform eigenmode decomposition to obtain the optimal parameter combination. Modal components ,in, It equals the number of modes in the optimal parameter combination;

[0082] S5: Using the absolute value of the difference between the average global amplitude and the average signal amplitude within the window as the quantization threshold, for each modal component under the optimal parameter combination... Perform one-dimensional local ternary mode operations to obtain the corresponding texture feature signals. ;

[0083] Among them, the first under the optimal parameter combination Modal components Corresponding texture feature signal The generation method is as follows:

[0084] S51: Set the parameters for the sliding window, including the length of the sliding window. Preferred The preferred sliding step size is 1;

[0085] S52: Calculate the modal components The average global magnitude The formula is as follows:

[0086] ;

[0087] in, Represents the modal components The signal length; Represents the modal components In the sequence The values ​​of each sampling point;

[0088] S53: Utilize the sliding window in the modal components The signal sequence is traversed point by point with a step size of 1, and the local texture feature value of the signal within each window is calculated. The calculation method of the local texture feature value of the signal within the window is as follows:

[0089] S531: Calculate the average value of the signal amplitude within the window, where the first... The average signal amplitude within each window The calculation formula is as follows:

[0090] ;

[0091] S532: Calculate the difference between the amplitude of each sampling point in the window signal and the average amplitude of the window signal, where the first... The first signal in the window sampling points amplitude ( The average value of the signal amplitude within the window The difference The calculation formula is as follows:

[0092] ;

[0093] S533: Using the absolute value of the difference between the average global amplitude and the average amplitude of the signal within the window as the quantization threshold, the difference between the amplitude of each sampling point in the signal within the window and the average amplitude of the signal within the window is encoded to generate an encoded sequence, wherein the first... The first signal in the window Encoding the difference between the amplitude of each sampling point and the average amplitude of the signal within the window as follows:

[0094] ;

[0095] S534: Convert the encoded sequence into decimal feature values, which are used as local texture feature values ​​of the in-window signal, wherein the first... Local texture feature values ​​of the signal within a window Represented as:

[0096] ;

[0097] S54: Arrange the decimal feature values ​​corresponding to all signals within the window in the order in which the sliding window slides to obtain the texture feature signal. ;

[0098] S6: The texture feature signal is filtered according to the impact feature intensity of the texture feature signal to obtain a texture feature signal containing rich fault information;

[0099] Preferably, the squared envelope Gini index is used as an evaluation index for the impact feature intensity of the texture feature signal.

[0100] The specific screening steps are as follows:

[0101] S61: Perform Hilbert transform on each texture feature signal to extract the envelope signal corresponding to each signal;

[0102] S62: For the envelope signal of each texture feature signal, calculate its squared envelope Gini index. This index is used to quantify the impact of the signal. The larger the index value, the more significant the fault impact feature in the signal.

[0103] The formula for calculating the squared envelope Gini index is as follows:

[0104] ;

[0105] in, Representation norm Representation matrix row vectors The square envelope, Indicates to The sequence after the sequence is sorted in ascending order;

[0106] S63: Calculate the squared envelope Gini index of all texture feature signals, exclude the signal corresponding to the maximum value (to avoid the maximum value being affected by noise or irrelevant interference components), and select the texture feature signal corresponding to the second largest squared envelope Gini index as the final texture feature signal containing rich fault information for fault identification.

[0107] The larger the square envelope Gini index, the more impactful features the texture feature signal contains. However, the texture feature signal with the maximum value may contain noise or other irrelevant components (such as frequency conversion or natural frequency) interference. The texture feature signal with the second largest value can reflect the significant features of the signal to a certain extent, and is relatively more stable and reliable. It can effectively avoid selecting texture feature signals that are extremely abnormal but may not be representative.

[0108] S7: Perform spectral analysis on the texture feature signal containing rich fault information and identify bearing faults based on the relationship between the prominent frequency components in the spectrum and the bearing fault feature frequencies.

[0109] The bearing fault identification method proposed in this invention not only reduces signal noise but also enhances signal characteristics, making it more effective in extracting characteristic frequencies corresponding to faults. The following analysis verifies the bearing fault identification method using rolling bearings (Example 1) and intermediate bearings (Example 2) as examples.

[0110] Example 1

[0111] The bearing fault identification method proposed in this invention is verified and analyzed using Case 1 (rolling bearing) from Table 1 as an example. Case 1 corresponds to the combined fault data of the outer and inner rings of the rolling bearing. Represents rotational frequency; with Represents rotational speed; with Inner ring fault characteristic frequency; Outer ring fault characteristic frequency; Represents the characteristic frequency of rolling element failure; Cage failure characteristic frequency.

[0112] This set of data shows the rotational speed. =2027 r / min, the characteristic frequencies of bearing failure are shown in Table 1. Specifically: =33.44Hz, =148.27 Hz, =85.84 Hz, =58.25 Hz, =12.26 Hz.

[0113] Table 1. Case 1 Experimental Data Information Table

[0114]

[0115] The time domain of the original vibration signal in Case 1 is as follows: Figure 2 As shown, Figure 3 and Figure 4 They are respectively Figure 2 The normalized spectrum and normalized power spectrum. Figure 5 and 6 These are the signals reconstructed from the covariance matrix of the original signal using the Hankel matrix. The time domain and power spectrum; Figure 7 This demonstrates the application of Eigenmode Decomposition in processing reconstructed signals. At that time, the minimum fractional order permutation entropy value of each component under different combinations of mode number and filter length; it can be found that when the mode number Choose 5, filter length When the value is 60, the minimum fractional permutation entropy is 4.125; the filter length for eigenmode decomposition is set. The number of modes is 60. The value is 5, for the reconstructed signal Decompose to obtain as follows Figure 8 The five modal components are shown. An improved one-dimensional local ternary mode is applied to each modal component, resulting in the texture feature signals shown below. Figure 9 As shown. Figure 10 It shows Figure 9 The power spectrum of each texture signal. Figure 11 The image shows the values ​​of the squared envelope Gini index of each texture feature signal. The squared envelope Gini index is the second largest, therefore... It is a texture feature signal that contains rich fault information. Figure 12 for Figure 8 medium component signal A magnified view of a portion of the image. Figure 13 for Figure 9 Medium texture feature signal A magnified view of a portion of the image. Figure 14 Texture feature signal The power spectrum.

[0116] analyze Figure 4 We can find that:

[0117] Only the harmonic components of the bearing inner ring fault characteristic frequency (148.271 Hz) can be extracted from the power spectrum of the original signal, including:

[0118] ① 714.11 Hz, this frequency corresponds to 5 and The difference is 714.11 Hz ((714.11+f r ) / 5=149.57);

[0119] ② 1006.08 Hz, this frequency corresponds to 7 and The difference is 1006.08 Hz ((1006.08+ f r ) / 7=148.65);

[0120] analyze Figure 14 and Figure 4 By comparison, it can be found that texture feature signals contain rich fault information. The power spectrum has the following characteristics:

[0121] (1) Effective noise suppression was achieved;

[0122] (2) There are prominent harmonic components corresponding to the characteristic frequency of the bearing outer ring fault (85.84 Hz), including:

[0123] ① 335.69 Hz, this frequency corresponds to 4 (335.69 / 4=83.92);

[0124] ② 695.8 Hz, this frequency corresponds to 8 (695.8 / 8=86.98);

[0125] ③ 958.25 Hz, this frequency corresponds to 11 (958.25 / 11=87.11);

[0126] ④ 1218.26 Hz, this frequency corresponds to 14 (1218.26 / 14=87.02);

[0127] (3) There are prominent harmonic components corresponding to the characteristic frequency of the bearing inner ring fault (148.27 Hz), including:

[0128] ① 872.8 Hz, this frequency corresponds to 6 With 2 The difference, ((872.8+2 () / 6=149.55);

[0129] ② 1031.49 Hz, this frequency corresponds to 7 (1031.49 / 7=147.35);

[0130] It can be seen that the method proposed in this invention effectively suppresses noise while extracting characteristic frequencies corresponding to bearing fault types, and each fault characteristic frequency is relatively prominent, thus achieving accurate identification of bearing fault types.

[0131] It should be noted that the purpose of disclosing the embodiments is to help further understand the present invention; however, those skilled in the art will understand that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the scope of protection of the present invention is defined by the scope of the claims.

Claims

1. A bearing fault identification method based on signal decomposition and one-dimensional local ternary mode, characterized in that, include: S1: Collect vibration signals during the operation of the bearing equipment. , ,in, The length of the vibration signal; S2: Based on the vibration signal Constructing the Hankel matrix The Hankel matrix is ​​calculated through relevant operations. covariance matrix Based on the covariance matrix Constructing reconstructed signals ; S3: Within a preset optimization range of mode number and filter length, the reconstructed signal is... Perform eigenmode decomposition and trial calculations, and determine the optimal combination of mode number and filter length based on the complexity of the optimized trial calculation mode components obtained after decomposition. The smaller the complexity of the optimized trial calculation mode components, the better the corresponding combination of mode number and filter length parameters. S4: The reconstructed signal is processed using the optimal combination of the mode number and filter length. Perform eigenmode decomposition to obtain the modal components corresponding to the optimal parameter combination; S5: Using the absolute value of the difference between the average global amplitude and the average signal amplitude within the window as the quantization threshold, perform one-dimensional local ternary mode operation on each modal component under the optimal parameter combination to obtain the corresponding texture feature signal. S6: The texture feature signal is filtered according to the impact feature intensity of the texture feature signal to obtain a texture feature signal containing rich fault information; S7: Perform spectral analysis on the texture feature signal containing rich fault information and identify bearing faults based on the relationship between the prominent frequency components in the spectrum and the bearing fault feature frequencies.

2. The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode according to claim 1, characterized in that, In S3, the steps to determine the optimal combination of mode number and filter length are as follows: S31: Set the optimization range for the number of modes and the filter length; S32: Under different combinations of mode number and filter length parameters, respectively, the reconstructed signal is processed. Perform eigenmode decomposition to obtain the optimization trial modal components corresponding to each set of parameter combinations; S33: Analyze the complexity of the optimization trial modal components corresponding to each parameter combination, and determine the optimal parameter combination as the combination of the number of modes and the filter length parameter corresponding to the optimization trial modal components with the minimum complexity.

3. The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode according to claim 1, characterized in that, In S3, fractional permutation entropy is used as the evaluation index of the complexity of the optimization trial modal component. The number of modes and the filter length corresponding to the optimization trial modal component with the minimum fractional permutation entropy are taken as the optimal parameter combination of the number of modes and the filter length.

4. The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode according to claim 1, characterized in that, In S5, the first under the optimal parameter combination Modal components Corresponding texture feature signal The generation method is as follows: S51: Set the parameters for the sliding window; S52: Calculate the modal components The average global magnitude ; S53: Utilize the sliding window in the modal components The signal sequence is traversed point by point with a step size of 1, and the local texture feature value of the signal within each window is calculated. The calculation method of the local texture feature value of the signal within the window is as follows: S531: Calculate the average value of the signal amplitude within the window; S532: Calculate the difference between the amplitude of each sampling point in the window signal and the average amplitude of the window signal; S533: Using the absolute value of the difference between the average global amplitude and the average amplitude of the signal within the window as the quantization threshold, the difference between the amplitude of each sampling point in the signal within the window and the average amplitude of the signal within the window is encoded to generate an encoded sequence; S534: Convert the encoded sequence into decimal feature values, which are used as local texture feature values ​​of the signal within the window; S54: Arrange the decimal feature values ​​corresponding to all signals within the window in the order in which the sliding window slides to obtain the texture feature signal. .

5. The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode according to claim 4, characterized in that, In S533, the first The first signal in the window Encoding the difference between the amplitude of each sampling point and the average amplitude of the signal within the window as follows: ； in, Represents the first under the optimal parameter combination Modal components The average global amplitude, Indicates the first The average value of the signal amplitude within each window , Indicates the length of the sliding window. Represents the modal components The signal length; Indicates the first The first signal in the window sampling points The amplitude and the average value of the signal amplitude within the window The difference, .

6. The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode according to claim 1, characterized in that, In S6, the squared envelope Gini index is used as an evaluation index for the impact feature intensity of texture feature signals.

7. The bearing fault identification method based on signal decomposition and one-dimensional local ternary mode according to claim 6, characterized in that, The specific steps for S6 are as follows: S61: Perform Hilbert transform on each texture feature signal to extract the envelope signal corresponding to each signal; S62: For the envelope signal of each texture feature signal, calculate its squared envelope Gini index; S63: Calculate the squared envelope Gini index of all texture feature signals, exclude the signal corresponding to the maximum value, and select the texture feature signal corresponding to the second largest squared envelope Gini index as the final texture feature signal containing rich fault information for fault identification.

Images