Impact characteristic mode decomposition method for mechanical signal analysis
By adopting the impact feature mode decomposition method in mechanical fault signal processing, using the kraft index and adaptive filtering framework, the problem of difficulty in extracting fault information under low signal-to-noise ratio conditions in the existing technology is solved, and the fault component extraction robustness and diagnostic effect are achieved.
Patent Information
- Application Number
- CN202510018823.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-07
- Publication Date
- 2025-04-11
AI Technical Summary
The existing mechanical fault signal decomposition method is difficult to effectively extract fault information when dealing with rolling bearing composite faults, especially under low signal-to-noise ratio conditions, and the feature extraction capability is poor.
The impact feature mode decomposition method is adopted, and the kraft index is set as the decomposition target, and the adaptive filtering framework is used to achieve synchronous acquisition of the fault mode, and the envelope spectrum kraft is introduced as the mode filtering criterion.
It improves the robustness of the extraction of fault components, overcomes the diagnosis problems under low signal-to-noise ratio and complex interference, and significantly improves the effectiveness of mechanical fault signal processing.
Smart Images
Figure CN119803934A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of mechanical equipment fault diagnosis and signal analysis and processing, and in particular relates to an impulse feature mode decomposition method (IFMD) for mechanical signal analysis. Background Art
[0002] The rapid development of mechanical equipment towards large-scale, complex and integrated has also brought great challenges to the safe service status monitoring of equipment. The decomposition method can separate the fault feature information in complex mechanical signals and is an effective means to solve mechanical fault diagnosis. Therefore, the invention and research of the decomposition method is of great significance to the safety of mechanical equipment.
[0003] Taking rolling bearings, which are the most vulnerable to damage in mechanical equipment, as an example, vibration detection technology is the most widely used in its monitoring and diagnosis, especially wavelet analysis, deconvolution technology, spectral kurtosis and other feature extraction technologies, which have achieved remarkable results in rolling bearing fault diagnosis in many industrial fields. However, these technologies often focus on screening the optimal fault frequency band, but may ignore the non-optimal frequency bands that also contain fault information. For the composite faults of rolling bearings, due to the multiplicity and discreteness of its fault frequency bands, the above methods face great challenges in successfully extracting fault information.
[0004] Decomposition methods can decompose different frequency band information in the test signal into different modes by setting decomposition strategies. They are important tools for solving the multi-component and multi-interference problems of complex signals and have been widely used in the field of mechanical fault diagnosis. As a classic decomposition method, empirical mode decomposition (EMD) has played an important role in mechanical signal analysis since it was proposed. However, the EMD method has deficiencies in mathematical theoretical basis and physical significance, and the mode aliasing phenomenon is serious, making it difficult to cope with the increasingly complex mechanical signal decomposition needs. Subsequently, new methods such as wavelet packet decomposition, empirical wavelet transform (EWT), and local mean decomposition (LMD) have been introduced. Although they have overcome the defects of EMD to a certain extent, they still have their own problems, such as the wavelet basis design dependence of EWT and the low processing efficiency of LMD, which limit their application in mechanical fault signal processing.
[0005] The introduction of variational mode decomposition (VMD) plays a key role in solving the above decomposition method problems. As a new non-recursive decomposition method, VMD transforms the decomposition problem into a variational optimization problem, takes narrowband bandwidth minimization as the optimization goal, and realizes signal decomposition through Wiener filter banks. It has been widely used in the field of mechanical fault diagnosis.
[0006] Although VMD has significant advantages over traditional EMD methods in solving problems such as mode aliasing, VMD still has some problems in rolling bearing fault diagnosis:
[0007] (1) VMD is not designed specifically for mechanical fault signal analysis, and its decomposition target is not typical mechanical fault pulses. This makes it susceptible to interference when processing mechanical faults, which affects the extraction of fault features.
[0008] (2) VMD is based on the filtering principle of the Wiener filter bank. However, it is limited by the center frequency and fixed filter shape, which makes its feature extraction ability poor under low signal-to-noise ratio conditions. Summary of the invention
[0009] The purpose of the present invention is to provide an impact characteristic pattern decomposition method for mechanical signal analysis. The kurtosis index that can quantitatively evaluate the fault impact is used as the decomposition target, and the fault mode is synchronously acquired with the help of an adaptive filtering framework, so as to realize the extraction and separation of the mechanical fault impact component in the test signal.
[0010] To achieve the above object, the present invention discloses the following technical solution:
[0011] The impact characteristic mode decomposition method for mechanical signal analysis comprises the following steps:
[0012] Step 1: Obtain a vibration signal x=[x(1) x(2) ... x(N)] of the rolling bearing to be tested, where the signal length is N.
[0013] Step 2: Set the input parameters including filter length L, number of initialized filters K, number of pre-iterations n and number of target decomposition patterns m.
[0014] Step 3: Divide the signal analysis frequency band into K segments equally, and then construct an initialized finite impulse response filter bank based on the Hanning window. The upper cutoff frequency f of each filter is l and the lower cutoff frequency f u The design is carried out according to the following formula (1):
[0015]
[0016] Where: f srepresents the sampling frequency of the signal. Then, the iteration number i=1 is marked.
[0017] Step 4: Use the filter bank to filter and decompose the test signal x to obtain K decomposition patterns. This filtering process can be rewritten as the matrix form of formula (2):
[0018] u k =Xf k , (2)
[0019] Where: u k It represents the kth decomposition mode, f k Then it means to obtain the filter coefficient of the kth mode, and X represents the Toeplitz matrix constructed based on the test signal x. The expressions are as follows:
[0020]
[0021] Step 5: Update the filter coefficients using the iterative eigenvalue decomposition algorithm. Specifically, the kurtosis expression of the decomposition pattern can be rewritten as the matrix form of the following formula (3):
[0022]
[0023] Where: R represents the weighted correlation matrix: R XX =X H X. The symbol H denotes conjugate, and diag denotes a diagonal matrix.
[0024] According to the maximum eigenvalue principle, solving the maximum kurtosis value problem of the decomposition mode is transformed into solving the maximum generalized eigenvalue problem of the following formula. The eigenvector corresponding to the maximum eigenvalue λ is the optimal filter coefficient to be solved: R XWX f k =R XX f k λ.
[0025] Then based on u k =Xf k Get a new decomposition mode and set the number of iterations i=i+1.
[0026] Step 6: Determine whether the number of iterations i reaches the number of pre-iterations n. If not, return to step 4; if reached, proceed to step 7.
[0027] Step 7: Based on the correlation coefficient formula (4), calculate the correlation coefficients between the K decomposition patterns:
[0028]
[0029] Where: and They are the decomposition modes u p and u q The average value of .
[0030] Step 8: Lock the two modes with the largest correlation coefficients, calculate their envelope spectrum kurtosis (ESK) values according to the following formula (5), and terminate the iteration of the mode with the smaller value. Then, set K=K-1.
[0031]
[0032] Where: U k Represented as mode u k Envelope spectrum.
[0033] Step 9: Determine whether K=m is true. If not, return to step 4; if true, proceed to step 10.
[0034] Step 10: Finally, m output decomposition modes are obtained, and the corresponding envelope spectrum is constructed, and the envelope spectrum is used to perform fault diagnosis on the bearing to be tested.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] The present invention uses the kurtosis index that can quantitatively evaluate the mechanical impact characteristics as the decomposition target. Compared with traditional decomposition methods such as VMD and EMD, which are not derived from the field of mechanical fault diagnosis, whose decomposition targets are not typical mechanical fault characteristics and which are easily affected by other interfering components when processing mechanical faults, the present invention is more robust when extracting fault components.
[0037] The present invention realizes the acquisition of decomposition patterns under the framework of adaptive filtering and introduces envelope spectrum kurtosis as a pattern screening criterion, which not only overcomes the defect of kurtosis index-based method that is sensitive to random impact interference to a certain extent, but also has better extraction ability when facing diagnosis problems under low signal-to-noise ratio and complex interference, including rolling bearing compound faults. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a flow chart of the rolling bearing fault diagnosis method based on impact characteristic mode decomposition of the present invention;
[0039] Figure 2 A schematic diagram of a freight train wheel-rail bearing fault detection platform according to an embodiment of the present invention;
[0040] Figure 3 A schematic diagram of the time domain waveform and envelope spectrum of a test signal obtained by the acceleration sensor according to an embodiment of the present invention;
[0041] Figure 4 It is a schematic diagram of the final mode and its envelope spectrum decomposed by the IFMD method of the present invention;
[0042] Figure 5 The final mode and its envelope spectrum obtained by decomposition using the existing VMD method are shown in the figure.
[0043] Figure 6 Schematic diagram of the final mode and its envelope spectrum decomposed by the existing EMD method. DETAILED DESCRIPTION
[0044] The following is a description of the embodiments of the present invention with reference to the accompanying drawings:
[0045] Figure 1 FIG. 1 is a flow chart of a bearing fault diagnosis method based on impact characteristic mode decomposition according to the present invention. Figure 1 As shown, the present invention provides an impact characteristic mode decomposition method for mechanical signal analysis, which includes:
[0046] Step 1: Attach the vibration acceleration sensor to the bearing seat of the rolling bearing to be tested, and collect the vibration signal x = [x(1) x(2) ... x(N)] of the rolling bearing to be tested, where the sampling frequency is f s =76800Hz, duration is 1s, and signal length is N=76800.
[0047] Step 2: Set the input parameters including filter length L=40, number of initialized filters K=10, number of pre-iterations n=10 and number of target decomposition patterns m=2.
[0048] Step 3: Divide the signal analysis frequency band into K segments equally, and then construct an initialized finite impulse response filter bank based on the Hanning window. The upper cutoff frequency f of each filter is l and the lower cutoff frequency f u The design is performed according to the following formula (1): Then, the iteration number i=1 is marked.
[0049]
[0050] Step 4: Use the filter bank to filter and decompose the test signal x to obtain K decomposition patterns. This filtering process can be rewritten as the matrix form of formula (2):
[0051] u k =Xf k , (2)
[0052] Where: u k It represents the kth decomposition mode, f kThen it means to obtain the filter coefficient of the kth mode, and X represents the Toeplitz matrix constructed based on the test signal x. The expressions are as follows:
[0053]
[0054] Step 5: Update the filter coefficients using the iterative eigenvalue decomposition algorithm. Specifically, the kurtosis expression of the decomposition pattern can be rewritten as the matrix form of the following formula (3):
[0055]
[0056] Where: R represents the weighted correlation matrix: R XX =X H X. The symbol H denotes conjugate, and diag denotes a diagonal matrix.
[0057] According to the maximum eigenvalue principle, solving the maximum kurtosis value problem of the decomposition mode is transformed into solving the maximum generalized eigenvalue problem of the following formula. The eigenvector corresponding to the maximum eigenvalue λ is the optimal filter coefficient to be solved: R XWX f k =R XX f k λ.
[0058] Then based on u k =Xf k Get a new decomposition mode and set the number of iterations i=i+1.
[0059] Step 6: Determine whether the number of iterations i reaches the number of pre-iterations n. If not, return to step 4; if reached, proceed to step 7.
[0060] Step 7: Based on the correlation coefficient formula (4), calculate the correlation coefficients between the K decomposition patterns:
[0061]
[0062] Where: and They are the decomposition modes u p and u q The average value of .
[0063] Step 8: Lock the two modes with the largest correlation coefficients, calculate their envelope spectrum kurtosis (ESK) values according to the following formula (5), and terminate the iteration of the mode with the smaller value. Then, set K=K-1.
[0064]
[0065] Where: U k Represented as mode u k Envelope spectrum.
[0066] Step 9: Determine whether K=m is true. If not, return to step 4; if true, proceed to step 10.
[0067] Step 10: Finally, m output decomposition modes are obtained, and the corresponding envelope spectrum is constructed, and the envelope spectrum is used to perform fault diagnosis on the bearing to be tested.
[0068] Specifically, build Figure 2 The bearing fault detection platform shown in the figure. Its working principle is that the drive motor drives the outer ring of the bearing to rotate through the friction between the drive wheel and the wheel pair bearing, while the inner ring of the bearing remains stationary. This test uses the test data of rolling bearings with combined inner and outer ring faults for verification. Figure 3 The data is shown in Figure 1, where (a) is the time domain waveform, and the overall fluctuation trend of the signal can be clearly observed. Combined with (b) the envelope spectrum, it can be considered that the fluctuation trend in the time domain waveform is partly caused by the rotation frequency of the test bearing or its shaft. After careful inspection, it was found that the cause of the signal fluctuation was the misalignment of the shaft system and its connecting parts connected to the bearing when the faulty bearing was installed.
[0069] based on Figure 3 The vibration signal shown in FIG. 1 is processed by the method of the present invention, and the result is shown in FIG. Figure 4 Combining its time domain waveform and envelope spectrum, it can be seen that the inner ring fault characteristic component and outer ring fault characteristic component of the bearing are decomposed into mode 1 and mode 2 respectively.
[0070] based on Figure 3 The vibration signal shown in FIG. 1 is processed by the existing VMD method and two modes are selected to obtain. The decomposition results are as follows: Figure 5 The bearing outer race fault is visible in mode 1, but mode 2 is completely dominated by noise.
[0071] based on Figure 3 The vibration signal shown is processed by the existing EMD method. Figure 6 This is the result of EMD decomposition. The inner race fault of the bearing can be seen in mode 1, but its mode 2 is also occupied by noise. Compared with IFMD and VMD, the proportion of fault information in its results is relatively low.
[0072] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.
Claims
1. A method for decomposing impact characteristic patterns for mechanical signal analysis, characterized in that: The following steps are involved: Step 1: Obtain a vibration signal x=[x(1) x(2) ... x(N)] of the rolling bearing to be tested, where the signal length is N. Step 2: Set the input parameters including filter length L, number of initialized filters K, number of pre-iterations n and number of target decomposition patterns m. Step 3: Divide the signal analysis frequency band into K segments equally, and then construct an initialized finite impulse response filter bank based on the Hanning window. The upper cutoff frequency f of each filter is l and the lower cutoff frequency f u The design is performed according to the following formula (1): Where: f s represents the sampling frequency of the signal. Then, the iteration number i=1 is marked. Step 4: Use the filter bank to filter and decompose the test signal x to obtain K decomposition patterns. This filtering process can be rewritten as the matrix form of formula (2): u k =Xf k , (2) Where: u k It represents the kth decomposition mode, f k Then it means to obtain the filter coefficient of the kth mode, and X represents the Toeplitz matrix constructed based on the test signal x. The expressions are as follows: Step 5: Update the filter coefficients using the iterative eigenvalue decomposition algorithm. Specifically, the kurtosis expression of the decomposition pattern can be rewritten as the matrix form of the following formula (3): Where: R represents the weighted correlation matrix: R XX =X H X. The symbol H denotes conjugate, and diag denotes a diagonal matrix. According to the maximum eigenvalue principle, solving the maximum kurtosis value problem of the decomposition mode is transformed into solving the maximum generalized eigenvalue problem of the following formula. The eigenvector corresponding to the maximum eigenvalue λ is the optimal filter coefficient to be solved: R XWX f k =R XX f k λ. Then based on u k =Xf k Get a new decomposition mode and set the number of iterations i=i+1. Step 6: Determine whether the number of iterations i reaches the number of pre-iterations n. If not, return to step 4; if reached, proceed to step 7. Step 7: Based on the correlation coefficient formula (4), calculate the correlation coefficients between the K decomposition patterns: Where: and They are the decomposition modes u p and u q The average value of . Step 8: Lock the two modes with the largest correlation coefficients, calculate their envelope spectrum kurtosis (ESK) values according to the following formula (5), and terminate the iteration of the mode with the smaller value. Then, set K=K-1. Where: U k Represented as mode u k Envelope spectrum. Step 9: Determine whether K=m is true. If not, return to step 4; if true, proceed to step 10. Step 10: Finally, m output decomposition modes are obtained, and the corresponding envelope spectrum is constructed, and the envelope spectrum is used to perform fault diagnosis on the bearing to be tested.