A fault recognition method fusing variance and 1D-LBP

Abstract

This invention designs a fault identification method that integrates variance and 1D-LBP, belonging to the field of fault diagnosis technology. First, the original vibration signal of rotating machinery is acquired, a moving window is constructed, and the difference between the signal within the moving window (based on the original signal) and the local variance of the signal within the window is calculated. Then, the binarized result of the difference is converted to decimal to generate a texture signal. Next, the moving window function acquires the local texture signal of all vibration signal data. Finally, based on spectral analysis, the acquired local texture signal is analyzed, its characteristic frequencies are extracted, and the fault characteristic information of rotating machinery is extracted and the fault type is identified according to the formula for calculating the characteristic frequency of rotating machinery faults. This invention not only effectively suppresses noise components in the vibration signal and highlights fault characteristic information, but also combines spectral analysis with 1D-LBP, facilitating the diagnosis and identification of rotating machinery faults, and has excellent engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, and specifically relates to a fault identification method that integrates variance and 1D-LBP. Background Technology

[0002] Rotating machinery is widely used in aviation, power, energy, and other fields. The health of its key components, such as bearings and gearboxes, directly affects the operation of the equipment. However, due to the harsh working environment and the long-term operation at high temperatures and speeds, rotating machinery is highly susceptible to failure. Furthermore, the performance of these key components gradually degrades with age, significantly impacting the safety and stability of the entire mechanical system. Therefore, accurate identification and diagnosis of faults in the key components of rotating machinery are crucial for reliable operation.

[0003] To achieve effective diagnosis of rotating machinery faults, scholars have conducted extensive research. This includes studies based on signal separation algorithms, such as wavelet transform, variational mode decomposition, and empirical mode decomposition; various denoising algorithms, such as wavelet thresholding and maximum correlation kurtosis deconvolution; and various information fusion algorithms, such as full vector spectrum and PCA. These signal analysis methods each have their own advantages in processing fault signals. However, they also have certain limitations. For example, the selection of wavelet bases in wavelet transform lacks adaptability; empirical mode decomposition suffers from severe endpoint effects and mode aliasing; and in variational mode decomposition, the determination of the number of decomposition levels and the center frequency has a crucial impact on fault identification.

[0004] Texture feature extraction is a crucial component of computer vision and pattern recognition. The Local Binary Pattern (LBMM) method compares neighboring pixels with the center pixel to generate an 8-bit binary pattern, thus extracting texture features and finding wide application in texture analysis. Research has shown that LBMMs can reflect the information of the entire pattern, reducing the number of features to a manageable level. Recently, LBMM texture analysis methods have been introduced into the field of fault diagnosis. However, research has focused on extracting texture features from signals based on the mean or center value. The resulting signals still contain significant noise interference, and fault feature information is not effectively highlighted, making it difficult to accurately extract fault feature information and accurately identify fault types. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a fault identification method based on the combination of variance and 1D-LBP (One-dimensional Local Binary Pattern); a method of converting the collected vibration signal into a binary pattern with variance (the mean or central value of unconventional methods) as the criterion is proposed to enhance the weak fault feature information and apply it to the fault diagnosis of rotating machinery; the present invention not only effectively suppresses the noise components in the vibration signal, but also further highlights the fault feature information, can effectively achieve the accurate extraction of the fault feature information of the key components of rotating machinery and the identification of fault types, and has excellent engineering application value.

[0006] A fault identification method integrating variance and 1D-LBP specifically includes the following steps:

[0007] Step 1: Use a sensor and a data acquisition card to obtain the original vibration signal {p 1, p 2, …p n} of the rotating machinery, construct a moving window with a window size of L (L < n), and calculate the difference b (j,i) between the signal x j intercepted according to the original signal in the moving window and the local variance x (j,i) of the signal in the window, where:

[0008]

[0009] b (j,i) = x (j,i) - x j (2)

[0010] n is the length of the original signal, x (j,i) is the original vibration signal in the moving window, i = 1, 2, … L, j is the number of moving windows, j = 1, 2, … n - L + 1, is the average value of the signal x (j,i) in the moving window;

[0011] Step 2: Binarize b (j,i) obtained in Step 1 with variance as the criterion, and f(b (j,i) ) is the signal after binarization corresponding to b (j,i) ; when the value x (j,i) in the moving window is greater than the local variance x j , that is, b (j,i) > 0, let it be equal to 1; when the value in the moving window is less than or equal to the local variance, that is, b (j,i) ≤ 0, let it be equal to 0;

[0012]

[0013] Step 3: Binarize the signal f(b) within the window (j,i) ), convert to decimal, and generate a new local texture signal LCS(j);

[0014]

[0015] Step 4: Move the window function by 1 unit, repeat steps 1-3, and calculate the new local texture signal LCS(j) corresponding to the vibration signal within each window function; when the end of the original signal is reached, stop moving the window function and obtain the local texture signal LCS of all data, LCS={LCS(1),LCS(2)…LCS(n-L+1)};

[0016] Step 5: Based on the spectral analysis method, analyze the local texture signal LCS obtained in Step 4, extract the characteristic frequency of the signal LCS, and realize the extraction of the characteristic information of rotating machinery faults and the identification of fault types according to the formula for calculating the characteristic frequency of rotating machinery faults.

[0017] Beneficial technical effects of the present invention:

[0018] This invention first considers that variance, compared to mean or center value, can better highlight weak local fault information in a sequence. Therefore, when binarizing vibration signals based on one-dimensional local binary data, this invention uses variance (replacing local mean or center value) as the criterion for binarization. The obtained binary sequence is then converted to decimal values, resulting in a new signal containing local texture information. Spectral analysis is then used to extract features from the obtained signal containing local texture information, thereby achieving the extraction of characteristic frequencies of rotating machinery faults and accurate identification of fault types. Compared with existing technologies, this invention not only effectively suppresses noise components in vibration signals and highlights fault characteristic information, but also combines spectral analysis with 1D-LBP, facilitating the diagnosis and identification of rotating machinery faults, and has excellent engineering application value. Attached Figure Description

[0019] Figure 1 A flowchart of a fault identification method that integrates variance and 1D-LBP according to an embodiment of the present invention;

[0020] Figure 2 Application of this invention in rolling bearing fault identification—Example 1;

[0021] Figure 3 Application of this invention in rolling bearing fault identification—Example 2;

[0022] Figure 4 Application of the present invention in rolling bearing fault identification—Example 3. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments;

[0024] A fault identification method based on a combination of variance and 1D-LBP is shown in the attached figure. Figure 1 As shown, taking a rolling bearing with a window size L=8 as an example, the specific steps include:

[0025] Step 1: Acquire the original discrete vibration signal of the rolling bearing using an accelerometer and a data acquisition card {p 1, p 2, …p n Construct a moving window with a window size of L=8, and calculate the signal x within the moving window based on the original signal. (j,i) Local variance of the signal within the window x j The difference b (j,i) ,in:

[0026]

[0027] b (j，i) =x (j，i) -x j (2)

[0028] n is the length of the original signal, x (j,i) Let i be the original vibration signal within the moving window, i = 1, 2, ..., 8, and j be the number of moving windows, j = 1, 2, ..., n-7. For the moving window signal x (j,i) The average value;

[0029] Step 2: Calculate b from step 1 using variance as the criterion. (j,i) Binarization, f(b) (j,i) ) for b (j,i) The corresponding binarized signal; when the value x within the moving window... (j,i) Greater than the local variance x j At that time, that is, b (j,i) When the value is greater than 0, set it to 1; when the value within the moving window is less than or equal to the local variance, i.e., b (j,i) When ≤0, set it to 0;

[0030]

[0031] Step 3: Binarize the signal f(b) within the window (j,i) ), convert to decimal, and generate a new local texture signal LCS(j);

[0032]

[0033] Step 4: Move the window function by 1 unit, repeat steps 1-3, and calculate the new local texture signal LCS(j) corresponding to the vibration signal within each window function; when the end of the original signal is reached, stop moving the window function and obtain the local texture signal LCS of all data, LCS = {LCS(1), LCS(2)...LCS(n-7)};

[0034] Step 5: Based on the Teager energy spectrum analysis method, analyze the local texture signal LCS obtained in Step 4, and extract the rolling bearing fault feature information and identify the fault type according to the calculation formula of fault feature frequency in Table 1.

[0035] Table 1 shows the parameters represented by each symbol and the calculation formulas for each characteristic frequency of the rolling bearing.

[0036]

[0037] The specific identification method for rolling bearing failure types is as follows:

[0038] If the frequency proposed based on the spectral analysis of the local texture signal is the outer ring fault characteristic frequency f o If the frequency is an integer multiple of the outer ring fault frequency, then the frequency is the characteristic frequency of the outer ring fault.

[0039] If the frequency proposed based on the spectral analysis of the local texture signal is the inner ring fault characteristic frequency f i multiples of an integer, or multiples of f i ±qf r ±pf c Where q = 0, 1, 2, then this frequency is the characteristic frequency of the inner ring fault;

[0040] If the frequency proposed based on the spectral analysis of the local texture signal is the characteristic frequency f of the rolling element fault... b multiples of an integer, or multiples of f b ±qf r ±pf c Where q = 0, 1, 2, then this frequency is the characteristic frequency of rolling element failure.

[0041] Example 1: First, taking the simulation signal of rolling bearing failure as an example, the detailed information of the simulation signal is shown in Table 2: The fault characteristics calculated according to Table 1 are as follows: rotation frequency f r =30Hz, characteristic frequency of rolling element failure f b =52.25Hz, inner ring fault frequency f i =133Hz, outer ring fault frequency f o =77Hz, cage failure frequency f c =11Hz.

[0042] Table 2 shows the parameters represented by each symbol and their physical meanings;

[0043]

[0044] The result obtained after processing according to the present invention is shown in the figure: Figure 2 (a) represents the time domain of the original vibration acceleration signal. Figure 2 (b) is Figure 2 (a) Teager energy spectrum. Figure 2 (c) Based on variance pairs Figure 2 (a) Temporal domain of the local texture signal LCS obtained after binarization. Figure 2 (d) is Figure 2 (c) Teager energy spectrum.

[0045] exist Figure 2 (b) The frequency components in the original signal's Teager energy spectrum are very complex, with a very large noise component. The fault characteristic frequencies are all submerged in noise, easily leading to misdiagnosis. Furthermore, only a prominent 1177Hz ((1177+11*2) / 23=52.13) frequency component can be detected, which corresponds to f b The original signal has a frequency of 23 times the original frequency; however, the simulated signal corresponds to a combined fault involving the inner race, outer race, and rolling element. Therefore, the fault type identification based on the energy spectrum of the original signal is inaccurate (no characteristic frequencies of outer race and inner race faults were found).

[0046] exist Figure 2 (d) Based on the fault identification method combining variance and 1D-LBP proposed in this invention, the Teager energy spectrum of the local texture signal LCS can be found to have the following frequency components and the following characteristics:

[0047] (1) Effectively highlights the frequency of fault characteristics;

[0048] (2) There exists f b The 4th and 34th harmonic components of (52.25Hz) are 237Hz ((237-30) / 4=51.75); 1781Hz (1781 / 34=52.38);

[0049] (3) There exists f i The 9th and 20th harmonics of (133Hz) are: 1190Hz (1190 / 9 = 132.22); 2704Hz ((2704-30) / 20 = 133.7).

[0050] (4) There exists f oThe 10th and 30th harmonic components of (77Hz) are: 775Hz (775 / 10 = 77.5) and 2316Hz (2316 / 30 = 77.2), respectively.

[0051] It can be seen that, according to the fault identification method based on the combination of variance and 1D-LBP proposed in this invention, noise components are effectively suppressed in the obtained Teager energy spectrum of the signal. Simultaneously, the characteristic frequencies completely and accurately corresponding to the rolling bearing fault types can be extracted, achieving accurate identification of composite rolling bearing fault types.

[0052] Example 2: Taking a single fault (rolling element) in a rolling bearing as an example, the sensor is installed in the vertical direction, and the rotational speed is 1813.19 r / min. The fault characteristics calculated according to Table 1 are as follows: rotational frequency f... r =30.21Hz, characteristic frequency of rolling element failure f b =52.62Hz, inner ring fault frequency f i =133.94Hz, outer ring fault frequency f o =77.54Hz, cage failure frequency f c =11.07Hz. The result obtained after processing according to the present invention is as follows: Figure 3 As shown: Figure 3 (a) represents the time domain of the original vibration acceleration signal. Figure 3 (b) is Figure 3 (a) Teager energy spectrum. Figure 3 (c) Based on variance pairs Figure 3 (a) Temporal domain of the local texture signal LCS obtained after binarization. Figure 3 (d) is Figure 3 (c) Teager energy spectrum.

[0053] exist Figure 3 (b) The following frequency components can be found in the Teager energy spectrum of the original signal, and they have the following characteristics:

[0054] (1) There exists f r The 20th harmonic of (33Hz) is the 608Hz (608 / 20=30.4) frequency component;

[0055] (2) There exists f b The 30th harmonic of (52.62Hz) is 1598Hz ((1598-30.21+11.07) / 30=52.63) and the 52nd harmonic component is 2721Hz (2721 / 52=52.33);

[0056] (3) Although the characteristic frequency components that match the fault type can be extracted, the frequency components of the signal Teager energy spectrum are very complex and the noise components are large. The fault characteristic frequencies are all submerged in the noise, which can easily lead to misdiagnosis. This is not conducive to the accurate extraction of fault characteristic frequencies and the effective identification of fault types.

[0057] exist Figure 3 (d) Based on the fault identification method proposed in this invention, which combines variance and 1D-LBP, the Teager energy spectrum of the signal can be analyzed to identify the following frequency components, which have the following characteristics:

[0058] (1) It reduces the interference of noise components and effectively highlights the fault characteristic frequency;

[0059] (2) There exists f b The 4th harmonic of 52.25Hz is 198Hz ((198+11.07) / 4=52.27), the 19th harmonic is 1012Hz (1012 / 19=53.26), the 37th harmonic is 1931Hz (1931 / 37=52.19), the 44th harmonic is 2332Hz (2332 / 44=53), and the 52nd harmonic is 2750Hz (2750 / 52=52.88).

[0060] It can be seen that, according to the fault identification method based on the combination of variance and 1D-LBP proposed in this invention, the noise component is effectively suppressed in the Teager energy spectrum of the obtained signal, and the fault characteristic frequency is effectively highlighted. The characteristic frequency that completely corresponds to the rolling bearing fault type can be extracted completely and accurately, thus realizing the accurate identification of the rolling bearing fault type.

[0061] Example 3: Taking a compound fault (inner ring + outer ring + rolling element) in a rolling bearing as an example, the sensor is installed horizontally, and the rotational speed is 1534.52 r / min. The fault characteristics calculated according to Table 1 are as follows: rotational frequency f... r =25.57Hz, characteristic frequency of rolling element failure f b =44.54Hz, inner ring fault frequency f i =113.38Hz, outer ring fault frequency f o =65.64Hz, cage failure frequency f c = 9.38Hz. The result obtained after processing according to the present invention is as follows: Figure 4 As shown: Figure 4 (a) represents the time domain of the original vibration acceleration signal. Figure 4 (b) is Figure 4 (a) Teager energy spectrum. Figure 4 (c) Based on variance pairs Figure 4(a) Temporal domain of the local texture signal LCS obtained after binarization. Figure 4 (d) is Figure 4 (c) Teager energy spectrum.

[0062] exist Figure 4 (b) Based on the fault identification method combining variance and 1D-LBP proposed in this invention, the Teager energy spectrum of the signal can be found to have the following frequency components and the following characteristics:

[0063] (1) There exists f r The 8th harmonic of (25.57Hz) is the 209Hz (209 / 8=26.12) frequency component;

[0064] (2) There exists f b The 30th harmonic component of (44.54Hz) is the 1326Hz (1326 / 30=44.2) frequency component;

[0065] (3) Although some fault characteristic frequency components can be extracted, the frequency components are very complex and the noise components are very large. The fault characteristic frequencies are easily submerged in the noise, which can easily lead to misdiagnosis. At the same time, the identification of fault types is inaccurate (fault characteristic frequencies of the outer and inner rings are not found).

[0066] exist Figure 4 (d) Based on the fault identification method proposed in this invention, which combines variance and 1D-LBP, the Teager energy spectrum of the signal can be analyzed to identify the following frequency components, which have the following characteristics:

[0067] (1) Effectively highlights the frequency of fault characteristics;

[0068] (2) There exists f b The 6th harmonic of 44.54Hz (266 / 6=44.33) and the 33rd harmonic of 1489Hz (1489 / 33=45.12) components;

[0069] (3) There exists f i The 12th harmonic of (113.38Hz) is 1310Hz ((1310+25.57*2) / 12=113.43), the 16th harmonic is 1802Hz (1802 / 16=112.62), and the 24th harmonic is 2692Hz ((2692+25.57) / 24=113.23).

[0070] (4) There exists f o The 11th and 30th harmonic components of (65.64Hz) are 717Hz (717 / 11=65.18) and 1961Hz (1961 / 30=65.37), respectively.

[0071] It can be seen that, based on the fault identification method combining variance and 1D-LBP proposed in this invention, noise components are effectively suppressed in the Teager energy spectrum analysis of the obtained signal. Simultaneously, the characteristic frequencies corresponding to the rolling bearing fault types can be extracted completely and accurately, achieving accurate identification of compound rolling bearing fault types.

[0072] In summary, the fault identification method based on the combination of variance and 1D-LBP proposed in this invention is insensitive to fault type and applicable to both single and compound faults; it is insensitive to sensor installation direction and applicable to acceleration signals acquired in both horizontal and vertical directions; it is insensitive to rotational speed; and it can accurately extract rolling bearing fault features and accurately identify fault types under different conditions, thus possessing excellent engineering application value.

Claims

1. A fault identification method integrating variance and 1D-LBP, characterized in that, Specifically, the following steps are included: Step 1: Obtain the original vibration signal {p 1, p 2, …p n} of the rotating machinery, construct a moving window with a window size of L (L < n), and calculate the signal x intercepted according to the original signal within the moving window (j,i) and the difference b between the signal x j and the local variance x (j,i) within the window; Step 2: Calculate b from step 1 using variance as the criterion. (j,i) Binarization, f(b) (j,i) ) for b (j,i) The corresponding binarized signal; Step 3: Binarize the signal f(b) within the window (j,i) ), convert to decimal, and generate a new local texture signal LCS(j); Step 4: Move the window function by 1 unit, repeat steps 1-3, and calculate the new local texture signal LCS(j) corresponding to the vibration signal within each window function; when the end of the original signal is reached, stop moving the window function and obtain the local texture signal LCS of all data, LCS={LCS(1),LCS(2)…LCS(n-L+1)}; Step 5: Based on the spectral analysis method, analyze the local texture signal LCS obtained in Step 4, extract the characteristic frequency of the signal LCS, and realize the extraction of the characteristic information of rotating machinery faults and the identification of fault types according to the formula for calculating the characteristic frequency of rotating machinery faults.

2. The fault identification method fusing variance and 1D-LBP according to claim 1, characterized in that, Step 1 calculates based on the original signal {p} 1, p 2, …p n The captured signal x within the moving window (j,i) Local variance of the signal within the window x j The difference b (j,i) ,in: b (j,i) =x (j,i) -x j (2) n is the length of the original signal, x (j,i) is the original vibration signal within the moving window, i = 1, 2, … L (L < n), j is the number of moving windows, j = 1, 2, … n - L + 1, is the signal x within the moving window (j,i) average value.

3. The fault identification method fusing variance and 1D-LBP according to claim 1, characterized in that, Step 2 will use the b obtained in Step 1 (j,i) Binarization, when the value x in the moving window is... (j,i) Greater than the local variance x j At that time, that is, b (j,i) When the value is greater than 0, set it to 1; when the value within the moving window is less than or equal to the local variance, i.e., b (j,i) When ≤0, set it to 0; 4. The fault identification method fusing variance and 1D-LBP according to claim 1, characterized in that, Step 3, generating the new local texture signal LCS(j), specifically involves:

Images