A rotating component fault diagnosis method based on multi-scale multi-modal decomposition

Abstract

The application relates to the technical field of high-end equipment intelligent operation and maintenance and signal processing, and discloses a rotating component fault diagnosis method based on multi-scale multi-modal decomposition, which comprises the following steps: collecting original vibration signals of a rotating component and inputting a preset scale parameter set; each scale parameter in the scale parameter set is selected in sequence, the original signal spectrum is equally divided into frequency bands, a corresponding number of window filter coefficients are initialized to construct a filter bank; the filter coefficients of the filter bank are iteratively updated by taking correlation kurtosis maximization as an optimization criterion, and multiple modal components under the current scale are obtained; the square envelope spectrum energy ratios of the modal components are calculated, a multi-scale square envelope spectrum energy ratio indication diagram is constructed, and the optimal modal component is selected according to the maximum value criterion; the optimal modal component is subjected to square envelope spectrum demodulation analysis, fault characteristic frequencies are extracted, and a rotating component fault diagnosis result is obtained.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance and signal processing technology for high-end equipment, specifically to a fault diagnosis method for rotating components based on multi-scale and multi-modal decomposition. Background Technology

[0002] Rotating components (such as bearings, gears, rotors, and plunger pump distribution plates) are core parts of high-end equipment like aero-engines, gas turbines, wind turbine generators, and hydraulic systems. Their operating condition directly affects the safety, reliability, and service life of the entire machine. Statistics show that over 30% of failures in rotating machinery are caused by bearing damage, and bearing failure in aero-engines is a major contributing factor to flight safety accidents. In hydraulic systems, plunger pumps, as core power components, have distribution plates that endure alternating loads and impacts under high-speed, high-pressure conditions for extended periods. This makes them prone to wear, cavitation, uneven wear, and fatigue spalling, which can lead to abnormal system pressure pulsations, reduced volumetric efficiency, and even complete machine failure. Therefore, conducting research on early fault diagnosis of rotating components is crucial for ensuring the safe operation of high-end equipment, preventing major accidents, and reducing maintenance costs.

[0003] Modal decomposition is an important signal processing method widely used in the field of rotating component fault diagnosis. Its core idea is to decompose the complex original vibration signal into several components with physical meaning, thereby facilitating the extraction of fault-related feature information. Existing mode decomposition methods mainly fall into the following categories: First, recursive decomposition methods, represented by Empirical Mode Decomposition (EMD), decompose signals into intrinsic mode functions through iterative sieving. However, they suffer from inherent problems such as mode aliasing and endpoint effects. Although subsequent improved algorithms, such as Ensemble Empirical Mode Decomposition (EEMD) and Adaptive Noise Complete Ensemble Empirical Mode Decomposition (CEEMDAN), have made some improvements, they have not fundamentally overcome the theoretical defects of recursive sieving. Second, non-recursive decomposition methods, represented by Variational Mode Decomposition (VMD), transform the decomposition problem into a variational constrained optimization problem. They have a solid mathematical theoretical foundation, but require pre-setting key parameters such as the number of modes and penalty factors. Parameter selection depends on human experience, resulting in insufficient adaptability. Third, envelope analysis methods, represented by Local Mean Decomposition (LMD), obtain local mean functions and envelope estimation functions through moving averages. They are sensitive to noise interference, and the decomposition accuracy decreases significantly under strong background noise.

[0004] In summary, existing modal decomposition methods still have many shortcomings in the application of fault diagnosis of rotating components in high-end equipment: recursive decomposition methods are prone to modal aliasing, non-recursive decomposition methods rely on manual preset of key parameters such as the number of modes and lack adaptive ability, most decomposition methods are sensitive to strong background noise and generally lack a mechanism for the collaborative utilization of multi-scale features, resulting in insufficient extraction accuracy of weak fault features under complex working conditions, and the diagnostic accuracy and robustness need to be further improved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a fault diagnosis method for rotating components based on multi-scale and multi-modal decomposition. This method can adaptively extract weak fault features under complex working conditions, effectively improving the accuracy and robustness of diagnosis.

[0006] A fault diagnosis method for rotating components based on multi-scale multi-modal decomposition includes: Collect the raw vibration signals of the rotating component and input the preset set of scale parameters; Each scale parameter in the set of scale parameters is selected in turn, the original signal spectrum is divided into frequency bands, and the corresponding number of window filter coefficients are initialized to construct a filter bank. Using the maximization of correlation kurtosis as the optimization criterion, the filter coefficients of the filter bank are iteratively updated to obtain multiple modal components at the current scale; Calculate the energy ratio of the squared envelope spectrum of each modal component, construct a multi-scale squared envelope spectrum energy ratio indicator map, and select the optimal modal component based on the maximum value criterion; The optimal modal components are subjected to square envelope spectrum demodulation analysis to extract fault characteristic frequencies and obtain fault diagnosis results for rotating components.

[0007] Furthermore, by sequentially selecting each scale parameter from the set of scale parameters, the original signal spectrum is divided into equal frequency bands, and a corresponding number of window filter coefficients are initialized to construct a filter bank, including: Select the scale parameters from the set of scale parameters in sequence, and determine the total number of frequency band modal components based on the current scale parameters; The original signal spectrum is evenly divided according to the total number of modal components in the frequency band to obtain a corresponding number of segmentation boundaries. The upper and lower cutoff frequencies of each frequency band are determined by each segmentation boundary. By combining the preset filter length, multiple Hamming window filter coefficients are initialized according to the upper and lower cutoff frequencies of each frequency band to construct a finite-length unit impulse response filter bank, and the iteration counter is initialized.

[0008] Furthermore, the elements in the scale parameter set are arranged in ascending order, and the total number of frequency band modal components is an integer obtained by exponentiation of the scale parameter with base 2 and then rounding down.

[0009] Furthermore, the window function used for the window filtering coefficients is any one of the following: Hamming window, Hanning window, Blackman window, rectangular window, triangular window, flat-top window, Bartlett window, and Taylor window.

[0010] Furthermore, using the maximization of relevant kurtosis as the optimization criterion, the filter coefficients of the filter bank are iteratively updated to obtain multiple modal components at the current scale, including: The original vibration signal is convolved with the current filtering coefficients to obtain the filtered modal component signal. Estimate the fault period, combine the original vibration signal, modal component signal and fault period to update the filter coefficients, complete a single iteration and update the iteration count; Determine if the preset number of iterations has been reached. If not, repeat the convolution and coefficient update process. If the preset number of iterations has been reached, save all final modal components at the current scale.

[0011] Furthermore, the fault period is estimated by the moment when the autocorrelation spectrum of the original vibration signal reaches a local maximum after crossing zero.

[0012] Furthermore, during the process of updating the filter coefficients, an objective function is constructed using the correlation kurtosis as an indicator, transforming the problem of maximizing the correlation kurtosis into the problem of maximizing the generalized eigenvalue. The generalized eigenvalue problem is solved using the Lagrange multiplier method, and the eigenvector corresponding to the largest eigenvalue is used as the optimal filter coefficient.

[0013] Furthermore, the energy ratio of the squared envelope spectrum of each modal component is calculated, a multi-scale squared envelope spectrum energy ratio indicator map is constructed, and the optimal modal component is selected according to the maximum value criterion, including: Calculate and store the squared envelope spectrum energy ratio of each final modal component at the current scale; Determine if the current scale parameter is the maximum value in the scale parameter set. If the traversal is not complete, repeat the scale selection and decomposition process. After traversing all scale parameters, a multi-scale multi-mode indicator map is constructed based on the frequency band boundaries corresponding to each scale, and the modal component corresponding to the global maximum value of the square envelope spectrum energy ratio is selected as the optimal modal component.

[0014] Furthermore, the definition of the squared envelope spectrum energy is as follows: (8) In the formula: f fault For the characteristic frequency of rotating component failure, Here, H represents the frequency deviation, and E represents the harmonic number. sq Let be the squared envelope spectrum of the modal components.

[0015] Furthermore, when performing square envelope spectrum demodulation analysis on the optimal modal components, the fundamental frequency component and its harmonic components of the fault characteristic frequency are extracted, and the fault state of the rotating component is determined accordingly.

[0016] The invention employing the above technical solution has the following advantages: This invention combines multi-scale spectrum equalization with multi-band MCKD filter bank optimization to achieve adaptive segmentation of signal frequency bands and joint optimization of filter coefficients under different scale parameters. This allows for the extraction of fault impact components in different frequency bands, effectively alleviating the mode aliasing problem inherent in traditional recursive decomposition methods. Simultaneously, by automatically determining the filter coefficients for each frequency band using the maximization of correlation kurtosis as the optimization criterion, it eliminates the need for manually presetting key parameters such as the number of modes and penalty factors, reducing the dependence of decomposition parameters on human experience. Furthermore, by automatically selecting the optimal component from the multi-scale, multi-modal results using the SESR index for squared envelope spectrum demodulation, it enhances the fault feature extraction capability under strong background noise environments, thereby effectively improving the accuracy and robustness of fault diagnosis.

[0017] The SESR of this invention introduces frequency deviations near the fault characteristic frequency and its harmonics. That is, in , , Energy integration is performed within a certain range, which fully considers the possible drift and fluctuation of fault characteristic frequencies in actual working conditions. At the same time, the energy of the fundamental frequency and its multiple harmonics are also included in the evaluation, which can more comprehensively reflect the spectral structure of fault characteristics and has stronger engineering adaptability. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 This invention provides a flowchart of a fault diagnosis method for rotating components based on multi-scale multi-modal decomposition. Figure 1 ; Figure 2 This is a time-domain waveform diagram of a fault measurement signal of a plunger pump distribution plate in Embodiment 1 of the present invention; Figure 3 This is a spectrum diagram of the measurement signal of a faulty bearing in the distributor plate of a plunger pump in Embodiment 1 of the present invention; Figure 4 This is the envelope spectrum of the measurement signal of a plunger pump distribution plate in Embodiment 1 of the present invention; Figure 5This is a graph showing the SESR index values ​​of each modal component under each selected scale parameter in Embodiment 1 of the present invention. Figure 6 This is a SESR index indication diagram for each modal component under each selected scale parameter in Embodiment 1 of the present invention; Figure 7 This is a graph showing the maximum SESR index under each scale parameter selected in Embodiment 1 of the present invention; Figure 8 This is the time-domain signal waveform diagram of the optimal modal component finally selected in Embodiment 1 of the present invention; Figure 9 This is the spectrum of the time-domain signal waveform of the optimal modal component finally selected in Embodiment 1 of the present invention; Figure 10 This is the square envelope spectrum of the time-domain signal waveform of the optimal modal component finally selected in Embodiment 1 of the present invention; Figure 11 The time-domain signal waveforms and squared envelope spectra of each modal component in EMD decomposition according to Embodiment 1 of the present invention are shown. Figure 12 The time-domain signal waveforms and squared envelope spectra of each modal component decomposed by EEMD in Embodiment 1 of the present invention are shown. Figure 13 The time-domain signal waveforms and squared envelope spectra of each modal component decomposed by CEEMDAN in Embodiment 1 of the present invention are shown. Figure 14 The time-domain signal waveforms and squared envelope spectra of each modal component in LMD decomposition according to Embodiment 1 of the present invention are shown. Figure 15 The image shows the time-domain signal waveforms and squared envelope spectra of each modal component in the VMD decomposition of Embodiment 1 of the present invention.

[0020] Figure 16 This is a schematic diagram of an aero-engine test bench and the sensor locations according to Embodiment 2 of the present invention; Figure 17 This is a time-domain waveform diagram of the measurement signal of a faulty bearing in an aero-engine according to Embodiment 2 of the present invention; Figure 18 This is a spectrum diagram of the measurement signal of a faulty bearing in an aero-engine in Embodiment 2 of the present invention; Figure 19 This is the envelope spectrum of the measurement signal of a faulty bearing in an aero-engine in Embodiment 2 of the present invention; Figure 20 This is a graph showing the SESR index values ​​of each modal component under each selected scale parameter in Embodiment 2 of the present invention; Figure 21 This is a SESR index indication diagram for each modal component under each selected scale parameter in Embodiment 2 of the present invention; Figure 22This is a graph showing the maximum SESR index under each scale parameter selected in Embodiment 2 of the present invention; Figure 23 This is the time-domain signal waveform diagram of the optimal modal component finally selected in Embodiment 2 of the present invention; Figure 24 This is the spectrum of the time-domain signal waveform of the optimal modal component finally selected in Embodiment 2 of the present invention; Figure 25 This is the square envelope spectrum of the time-domain signal waveform of the optimal modal component finally selected in Embodiment 2 of the present invention; Figure 26 The time-domain signal waveforms and squared envelope spectra of each modal component in EMD decomposition according to Embodiment 2 of the present invention are shown. Figure 27 The time-domain signal waveforms and squared envelope spectra of each modal component decomposed by EEMD in Embodiment 2 of the present invention are shown. Figure 28 The time-domain signal waveforms and squared envelope spectra of each modal component decomposed by CEEMDAN in Embodiment 2 of the present invention are shown. Figure 29 The time-domain signal waveforms and squared envelope spectra of each modal component in LMD decomposition in Embodiment 2 of the present invention are shown. Figure 30 The time-domain signal waveforms and squared envelope spectra of each modal component in VMD decomposition in Embodiment 2 of the present invention are shown. Figure 31 In this invention, the frequency deviation The variation of the SESR index of the optimal modal component under different combinations of harmonic number H; Figure 32 This invention provides a flowchart of a fault diagnosis method for rotating components based on multi-scale multi-modal decomposition. Figure 2 . Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0023] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0024] like Figures 1 to 32 As shown, the present invention provides a method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition, comprising: Collect the original vibration signal of the rotating component and input the preset set of scale parameters.

[0025] Specifically, the original vibration signal x(g) is collected by accelerometers placed at the measuring points of the rotating parts of high-end equipment, where g is the sequence number of the signal sampling point; at the same time, the preset parameters required for the algorithm to run are input, including the scale parameter set S and the filter length L, where the elements in the scale parameter set S are arranged in ascending order to facilitate subsequent sequential traversal of different decomposition scales.

[0026] Each scale parameter in the scale parameter set is selected sequentially, the original signal spectrum is divided into frequency bands, and the corresponding number of window filter coefficients are initialized to construct a filter bank.

[0027] In this embodiment, each scale parameter in the scale parameter set is selected sequentially to divide the original signal spectrum into equal frequency bands, and a corresponding number of window filter coefficients are initialized to construct a filter bank, including: Select scale parameters from the scale parameter set in sequence, and determine the total number of frequency band modal components based on the current scale parameter; The original signal spectrum is evenly divided according to the total number of modal components in the frequency band to obtain a corresponding number of segmentation boundaries. The upper and lower cutoff frequencies of each frequency band are determined by each segmentation boundary. By combining the preset filter length, multiple Hamming window filter coefficients are initialized according to the upper and lower cutoff frequencies of each frequency band to construct a finite-length unit impulse response filter bank, and the iteration counter is initialized.

[0028] In this embodiment, the elements in the scale parameter set are arranged in ascending order, and the total number of frequency band modal components is an integer obtained by exponentiation of the scale parameter with base 2 and then rounding down.

[0029] Specifically, the correspondence between the two is a piecewise function, in the following form: (1) In the formula: is the floor function, s is the scaling parameter, and N is the total number of modal components in the frequency band.

[0030] In this embodiment, the window function used for the window filtering coefficients is any one of the following: Hamming window, Hanning window, Blackman window, rectangular window, triangular window, flat-top window, Bartlett window, and Taylor window.

[0031] Specifically, the appropriate option can be flexibly selected based on the actual signal characteristics and diagnostic needs. Taking the Hamming window as an example, to initialize the Hamming window, the effective frequency band of the original signal is uniformly divided into N frequency band mode components. The formulas for calculating the upper and lower cutoff frequencies corresponding to the nth frequency band mode component are as follows: (2) In the formula: Let be the lower cutoff frequency of the nth frequency band. Let be the upper cutoff frequency of the nth frequency band. This represents the signal sampling frequency.

[0032] Based on the upper and lower cutoff frequencies of each frequency band and the preset filter length L, N Hamming window FIR filter coefficients are generated, and finally the FIR filter bank is constructed as follows: eq F\S( i,s(N)) =[ eq f\S( i,s(1)) , eq f\S( i,s(2)) eq f\S( i,s(n)) , eq f\S( i,s(n+1)) eq f\S( i,s(N)) ], where n=1, 2, N, initial iteration counter i=1.

[0033] Specifically, in the multi-scale spectrum equal division strategy adopted in this invention, the segmentation boundary is uniquely determined by the scale parameter s through the aforementioned piecewise function, resulting in the total number N of frequency band modal components. This uniformly divides the original signal spectrum into N equal-width frequency bands. That is, the segmentation boundary is completely independent of the local extrema, flat-top distribution, or other morphological features of the original signal spectrum. Regardless of where the signal spectrum exhibits peaks or flat-tops, the segmentation boundary position remains fixed under the same scale parameter s. This parameter-driven segmentation method has a fundamentally different technical approach from EWT's data-driven adaptive segmentation: EWT's adaptive segmentation aims to adapt to the personalized spectral characteristics of different signals, while the uniform equal division of this invention aims to provide a regularized and comparable frequency band division basis for subsequent multi-band MCKD optimization, making the decomposition results under different scale parameters comparable on the same coordinate reference, thereby providing a unified frequency band reference system for the multi-scale evaluation of the SESR indicator map.

[0034] Using the maximization of correlation kurtosis as the optimization criterion, the filter coefficients of the filter bank are iteratively updated to obtain multiple modal components at the current scale.

[0035] In this embodiment, the filter coefficients of the filter bank are iteratively updated using the maximization of correlation kurtosis as the optimization criterion to obtain multiple modal components at the current scale, including: The original vibration signal is convolved with the current filter coefficients to obtain the filtered modal component signal. Estimate the fault period, combine the original vibration signal, modal component signal and fault period to update the filter coefficients, complete a single iteration and update the iteration count; Determine if the preset number of iterations has been reached. If not, repeat the convolution and coefficient update process. If the preset number of iterations has been reached, save all final modal components at the current scale.

[0036] In this embodiment, the fault period is estimated by the moment when the autocorrelation spectrum of the original vibration signal reaches a local maximum after crossing zero.

[0037] Specifically, the number of sampling points corresponds to the fault period of the nth filter. In a single iteration, the original vibration signal is first convolved with the current filter coefficients to obtain the filtered nth modal component signal. The formula for discrete convolution is as follows: eq u\S(i,s(n)) =x(g)* eq f\S( i,s(n)) In the formula, "*" represents the convolution operation.

[0038] After completing the filtering calculation, the filter coefficients are updated by combining the original signal x(g), the decomposed modal components eq u\S(i,s(n)), and the estimated fault period eq T\S(i,s(n)).

[0039] Here, eq T\S( i,s(n)) is estimated by the time when the autocorrelation spectrum eq R\S( i,s(n)) reaches a local maximum after crossing zero. That is, the autocorrelation function of the original signal is first calculated, and within the range after the autocorrelation function crosses zero, the delay time corresponding to the first local maximum of its amplitude is searched, and this time is used as the estimated value of the fault period. After completing one iteration, the iteration counter is updated to i=i+1.

[0040] In this embodiment, during the process of updating the filter coefficients, an objective function is constructed using the correlation kurtosis as an indicator, transforming the problem of maximizing the correlation kurtosis into the problem of maximizing the generalized eigenvalue. The generalized eigenvalue problem is solved using the Lagrange multiplier method, and the eigenvector corresponding to the largest eigenvalue is used as the optimal filter coefficient.

[0041] Specifically, the derivation process is as follows: Under the scale parameter s, the optimization problem with the objective of maximizing the correlation kurtosis is defined as follows: (3) In the formula: eq u\S( ,s(n)) represents the nth decomposition mode under the scale parameter s; f s(n) This represents the nth FIR filter with length L under the scaling parameter s; T s(n) is the input period (measured by the number of samples); M is the shift order; G is the total number of samples of the original signal x(g).

[0042] Substituting the convolution relationship between the modal components and the filter into the above equation, the CK of the decomposed modes can be approximately transformed into matrix form, defined as follows: (4) In the formula: R1=X H W s(M) X; R2=X H X; the expressions for the other parameters are as follows: (5) Formula (4) can be rewritten as follows: (6) Therefore, the problem of finding the maximum correlation kurtosis CK value of the decomposition mode is transformed into the problem of finding the maximum generalized eigenvalue as shown in formula (6). By solving the problem using the Lagrange multiplier method, it can be finally summarized as follows: (7) In the formula, the eigenvector corresponding to the eigenvalue λ is the optimal filter coefficient to be solved.

[0043] According to formula (6), with the denominator fixed, we find a direction f. s(n) This makes the numerator reach its maximum relative to the denominator.

[0044] Here let eq f\S( H,s(n)) R2f s(n) =1, then maximizing λ is equivalent to maximizing the numerator eq f\S( H,s(n)) R1f under the above constraints. s(n) This is a standard constrained optimization problem, which can be solved using the Lagrange multiplier method.

[0045] Construct the Lagrangian function L: L = eq f\S( H,s(n)) R1 f s(n) - λ( eq f\S( H,s(n)) R2 f s(n) -1) In the formula, λ is a Lagrange multiplier (also a generalized eigenvalue).

[0046] Then take the partial derivative of eq f\S( H,s(n)) (this involves matrix differentiation), and set the derivative to 0:

[0047] Simplifying the above equation, we get:

[0048] After further refinement, we can finally obtain the following formula:

[0049] In the formula, the largest eigenvalue The corresponding eigenvectors are the optimal filter coefficients to be solved.

[0050] Specifically, the filter coefficient optimization of the present invention is embedded in the aforementioned multi-scale spectrum division framework, forming a two-dimensional optimization architecture of multi-scale and multi-band.

[0051] Specifically: On the one hand, this invention performs filter bank optimization under each scale parameter in the scale parameter set S; on the other hand, within each scale parameter, an independent filter coefficient optimization problem is constructed for each of the N frequency bands generated by equally dividing the spectrum. That is, the optimization of the filter coefficients in this invention is performed separately on a two-dimensional grid of the number of scale parameters × the number of frequency bands. Optimization is performed on the N frequency bands under the first scale parameter, then on the N frequency bands under the second scale parameter, and so on, until all scale parameters are traversed.

[0052] This multi-scale-multi-band two-dimensional optimization architecture enables the present invention to extract fault impact components at different decomposition granularities and at different frequency bands. In addition, the present invention uses the Lagrange multiplier method to unify the filter optimization problem of each frequency band into a generalized eigenvalue problem for closed-form solution.

[0053] Calculate the energy ratio of the squared envelope spectrum of each modal component, construct a multi-scale squared envelope spectrum energy ratio indicator map, and select the optimal modal component based on the maximum value criterion.

[0054] In this embodiment, the squared envelope spectrum energy ratio of each modal component is calculated, a multi-scale squared envelope spectrum energy ratio indicator map is constructed, and the optimal modal component is selected according to the maximum value criterion, including: Calculate and store the squared envelope spectrum energy ratio of each final modal component at the current scale; Determine if the current scale parameter is the maximum value in the scale parameter set. If the traversal is not complete, repeat the scale selection and decomposition process. After traversing all scale parameters, a multi-scale multi-mode indicator map is constructed based on the frequency band boundaries corresponding to each scale, and the modal component corresponding to the global maximum value of the square envelope spectrum energy ratio is selected as the optimal modal component.

[0055] In this embodiment, the square envelope spectrum energy ratio is calculated based on the square envelope spectrum of the rotating component fault characteristic frequency, frequency deviation, harmonic order, and modal components.

[0056] Specifically, the definition is as follows: (8) In the formula: f fault For the characteristic frequency of rotating component failure, Here, H represents the frequency deviation, and E represents the harmonic number. sq Let be the squared envelope spectrum of the modal components.

[0057] This index reflects the ratio of the energy in the frequency bands near the fault characteristic frequency and its harmonics to the energy in the non-characteristic frequency bands of the square envelope spectrum. The higher the ratio, the richer the fault characteristic information contained in the modal component.

[0058] In this invention, frequency deviation The value of is not a fixed constant, but is related to the bandwidth Bw under the current scale parameter s: Where α is a preset proportionality coefficient, Bw= ) represents the bandwidth of each frequency band after being uniformly divided under the current scale parameter s.

[0059] Because the granularity of frequency band division differs under different scale parameters s (N increases with s, while Bw decreases with s). It also adapts accordingly, at the coarse-grained scale (s is small, N is small, Bw is large). Larger values ​​are chosen to accommodate possible frequency drift over a wider bandwidth; at a finer granular scale (larger s, larger N, smaller Bw). The value is set relatively small to maintain a fine frequency resolution. This design allows ∆f to form an adaptive coupling with the multi-scale framework. It changes in tandem with the scale parameter s, rather than being a fixed constant independent of the scale parameter. Based on Figure 31 The differences shown The SESR index change curve under the combination of H, when When H=5, the SESR value tends to be stable and has high evaluation sensitivity, therefore it is the preferred option. H=5 is used as a typical implementation parameter.

[0060] according to Figure 31 As shown, firstly, at the same harmonic number H, with frequency deviation... As the deviation increases, the optimal modal component SESR value gradually increases; secondly, at the same frequency deviation... Below, as the harmonic number H increases, the optimal modal component SESR value also gradually increases; while in the frequency deviation... As the harmonic number H increases, the optimal modal component SESR value becomes more stable. To balance both parameters, this invention sets a frequency deviation. The frequency is 2Hz, and the harmonic number H is 5.

[0061] Compared with the square envelope spectral characteristic energy ratio (FER-SES) and energy concentration index ECm in the prior art, the SESR index of this application has the following differences: Existing indicators such as FER-SES, RCE, and characteristic energy factor all employ point-based energy extraction, calculating amplitude energy or amplitude sum by taking points at precise harmonics of the fault characteristic frequency.

[0062] When there is a deviation between the theoretical fault frequency and the actual fault frequency (i.e., frequency drift caused by factors such as speed fluctuations, load changes, and random slippage of rolling elements during actual operation), the accuracy of this point-based extraction method will decrease.

[0063] The SESR of this invention employs interval-based energy extraction at the fault characteristic frequency f. fault and its harmonics h·f fault Nearby frequency deviation This forms a continuous frequency band, within which the energy of the square envelope spectrum is integrated.

[0064] The former refers to a point, while the latter refers to a segment; they differ fundamentally in their energy extraction methods. Interval-based extraction gives SESR a natural robustness to unavoidable frequency drift in actual operating conditions, something point-based indicators cannot provide. Energy integration within the frequency band near the fault characteristic frequency and its harmonics can adapt to the drift and fluctuations of the fault characteristic frequency in actual operating conditions.

[0065] (2) The SESR of the present invention incorporates the energy of the fundamental frequency and its multiple harmonics into the evaluation by means of the harmonic order H, rather than only considering the fundamental frequency component, and can more comprehensively reflect the spectrum structure of the fault characteristics.

[0066] (3) The SESR of the present invention is built into the multi-scale multimodal decomposition framework and works in conjunction with multi-scale spectrum equal division and multi-band MCKD optimization. The input of SESR is the high-quality modal components after MCKD optimization, and the output is used to guide the automatic selection of the optimal components, forming a complete decomposition, evaluation and selection closed loop, rather than a post-processing index independent of the decomposition process.

[0067] The optimal modal components are subjected to square envelope spectrum demodulation analysis to extract fault characteristic frequencies and obtain fault diagnosis results for rotating components.

[0068] In this embodiment, when performing square envelope spectrum demodulation analysis on the optimal modal components, the fundamental frequency component and its harmonic components of the fault characteristic frequency are extracted, and the fault state of the rotating component is determined accordingly.

[0069] Specifically, the selected optimal mode component is first subjected to a Hilbert transform to obtain the corresponding analytic signal; the modulus of the analytic signal is then taken to obtain the envelope sequence of the signal; a Fourier transform is then performed on the envelope sequence and the square is taken to obtain the square envelope spectrum of the mode component. The prominence of the fundamental frequency and its multiple harmonics of the fault characteristic frequency is observed in the square envelope spectrum. If the fundamental frequency and harmonic spectral lines are clear, significant, and markedly higher than the surrounding background noise, it can be determined that the rotating component has a corresponding type of fault.

[0070] This application provides a specific implementation method, Example 1: To verify the effectiveness of the multi-scale multimodal decomposition method of this invention in diagnosing faults of rotating components in engineering practice, its application in the fault diagnosis of a heavy-duty equipment plunger pump was investigated. During the experiment, the system load pressure was stabilized at 5 MPa by adjusting the pilot-operated relief valve, and the motor speed was set to 1500 r / min. Based on this, the fundamental frequency of shaft vibration was calculated to be 25 Hz. Considering the pump's 7-plunger structure, the fundamental frequency fm of the plunger vibration was obtained as follows: The experiment simulated two operating conditions: normal operation of the plunger pump and average wear (0.67 μm) on the distributor plate end face. Vibration signals were acquired by an accelerometer, collected via LabVIEW data acquisition, and stored in real time on a computer. The signal sampling frequency was set to 20 kHz, and each operating condition was operated stably under load, collecting bidirectional vibration data.

[0071] Figure 2 The time-domain waveform of the wear fault in the distributor plate of the plunger pump is given. Figure 3 and Figure 4 The spectrum and square envelope spectrum of the time-domain waveform are given respectively. It can be seen that the periodic fault impact signal in the time-domain waveform is completely submerged, and the fundamental frequency fm and its harmonics in the square envelope spectrum are severely interfered with, making it difficult to determine that the distribution plate of the large heavy-duty equipment plunger pump has worn, indicating that the distribution plate is in a weak fault state.

[0072] according to Figure 1 The overall process steps of the multi-scale multi-modal decomposition method. Figure 5 The distribution of SESR index values ​​for each modal component under various scale parameters is presented. As can be seen from the figure, under most scale parameters, the frequency band corresponding to the maximum SESR value is concentrated in the range of 2000Hz to 4000Hz.

[0073] Figure 6 Furthermore, SESR index indication diagrams for each modal component under the selected scale parameters are provided. Under the global scale parameter condition, when the scale parameter value is 5, the center frequency fc is 3281Hz, and the bandwidth Bw is 312Hz, the SESR index reaches the global maximum value, i.e., SESR max =0.0425.

[0074] Furthermore, Figure 7 The variation trend of the maximum SESR index under different scale parameters is presented. As can be seen from the figure, as the scale parameter value increases, the SESR generally shows a trend of decreasing and then gradually increasing.

[0075] Under the above optimal parameter conditions, the selected optimal modal component time-domain waveform is as follows: Figure 8 As shown, its spectrum and squared envelope spectrum are respectively as follows: Figure 9 and Figure 10 As shown.

[0076] exist Figure 10 In the square envelope spectrum shown, the fundamental frequency characteristic 1f can be clearly identified. m Its harmonic components (2fm and 3fm), characteristic frequencies, and harmonic spectral lines are all significantly prominent and have not been significantly disturbed. Based on this, it can be determined that the plunger pump has experienced wear on the distribution plate.

[0077] To further verify the superiority of the multi-scale multi-modal decomposition method proposed in this invention in the diagnosis of distributor plate wear faults in plunger pumps, Figures 11 to 15 The time-domain waveforms and their squared envelope spectra of the five modal components obtained by the EMD, EEMD, CEEMDAN, LMD and VMD methods are shown respectively.

[0078] In the EMD method, the square envelope spectra of modal components M1 and M2 show significant spectral lines at the 1fm frequency position, but other harmonic components are not obvious, making it difficult to effectively determine whether the plunger pump has a distributor plate failure. Using the EEMD method, the 1fm to 2fm spectral lines in the square envelope spectrum of the M2 component are relatively prominent, but the spectral line at the 3fm position is masked by interference components, only allowing for a preliminary inference that the plunger pump may have distributor plate wear. In the decomposition results of the CEEMDAN method, the 1fm and 2fm spectral lines of the M4 component are relatively prominent, but the 3fm spectral line is also affected by interference, limiting the fault identification capability, and the conclusion is limited to the possibility of distributor plate failure. The square envelope spectra of the M1 and M2 components obtained by the LMD method show strong interference in the 1fm to 3fm spectral lines, making it difficult to provide more accurate fault characteristic information. The results of the VMD method show that the 1fm and 2fm spectral lines are relatively significant in the M3 and M4 components, but the 3fm spectral line is still submerged in noise, and the diagnostic information is also limited to the preliminary judgment of the distribution plate fault.

[0079] In summary, none of the comparative methods could provide a clear and complete fault feature spectrum structure as the method of this invention, further verifying the effectiveness and superiority of the method of this invention.

[0080] This application provides another specific implementation method, Implementation Two: To verify the effectiveness of the multi-scale multimodal decomposition method of this invention in diagnosing faults in rotating components in practical engineering, its application in the fault diagnosis of aero-engine bearing components was investigated. The selected engineering data was vibration data of a bearing fault in an accessory gearbox of an aero-engine. The engine structure and sensor installation locations are shown below. Figure 16 As shown. The bearing failure details are as follows: the outer ring of the bearing on gearbox L1 shaft has severe scratches and damage, and the outer ring of the bearing on L5 shaft has partial peeling. A pulse tachometer (44 pulses per revolution) is installed on L4 shaft, and accelerometers are installed near L1 and L5 shafts. The signal sampling frequency is 44100Hz, and the measured rotational frequency of the tachometer on L4 shaft is 180Hz. During data acquisition, the accelerometer near L1 shaft malfunctioned, and only the measurement signal from the accelerometer near L5 shaft was usable. Based on the number of teeth on gears 62 and 61 on L4 and L5 shafts respectively, the rotational frequency of L5 shaft can be obtained. Then, based on the ratio of the fault characteristic frequency of the L5 shaft bearing to the L5 shaft rotational frequency (Table 1), the fault characteristic frequency of the outer ring of the L5 bearing can be obtained. .

[0081] Table 1. Ratio coefficient of bearing failure characteristic frequency to L5 shaft rotation frequency

[0082] Figure 17 The time-domain waveforms of 16384 accelerometer measurements near the L5 axis are presented. Figure 18 and Figure 19 The spectrum and square envelope spectrum of the time-domain waveform are given respectively. It can be seen that the periodic fault impact signal in the time-domain waveform is completely submerged, and the characteristic frequency 1f of the outer ring fault of the aero-engine bearing is difficult to observe in the square envelope spectrum. m .

[0083] according to Figure 1 The overall process steps of the multi-scale multi-modal decomposition method. Figure 20 The distribution of SESR index values ​​for each modal component under various scale parameters is presented. As can be seen from the figure, under most scale parameters, the frequency band corresponding to the maximum SESR value is concentrated in the range of 5000Hz to 10000Hz.

[0084] Figure 21 Furthermore, SESR index indication diagrams for each modal component under the selected scale parameters are provided. Under global scale parameter conditions, when the scale parameter is 5, the center frequency fc is 6546Hz, and the bandwidth Bw is 689Hz, the SESR index reaches its global maximum value, i.e., SESR max =0.0175.

[0085] Furthermore, Figure 22 The variation trend of the maximum SESR index under different scale parameters is presented. As can be seen from the figure, the SESR generally shows a fluctuating trend as the scale parameter value increases.

[0086] Under the above optimal parameter conditions, the selected optimal modal component time-domain waveform is as follows: Figure 23 As shown, its spectrum and squared envelope spectrum are respectively as follows: Figure 24 and Figure 25 As shown.

[0087] exist Figure 25 In the square envelope spectrum shown, the characteristic frequency 1fm and its harmonic component (2fm) of the outer ring fault of the aero-engine bearing can be clearly identified. The characteristic frequency and its harmonic spectral lines are significantly prominent and not significantly disturbed. Based on this, it can be determined that there is an outer ring fault in the aero-engine accessory gearbox bearing.

[0088] To further verify the superiority of the multi-scale multi-modal decomposition method proposed in this invention in the fault diagnosis of aero-engine accessory gearbox bearings, Figures 26 to 30The time-domain waveforms and their squared envelope spectra of the five modal components obtained by the EMD, EEMD, CEEMDAN, LMD and VMD methods are shown respectively.

[0089] In the EMD method, the square envelope spectra of modal components M1-M5 do not show significant spectral lines at the 1fm and 2fm frequency positions, making it difficult to effectively determine the fault type of the bearing in the aero-engine accessory gearbox. Using the EEMD method, the 1fm spectral line is relatively prominent in the square envelope spectra of M2 and M3 components, but the 2fm spectral line is masked by interference components, only allowing for a preliminary inference that the bearing may have an outer race fault. In the decomposition results of the CEEMDAN method, the 1fm spectral line of M2 and M3 components is relatively clear; however, the 2fm spectral line is also affected by interference, limiting fault identification capabilities and concluding only the possibility of an outer race fault. The LMD method shows that the 1fm spectral line in the square envelope spectrum of the M2 component is relatively clear, but the 2fm spectral line is strongly interfered with, making it difficult to provide more precise fault characteristics. The VMD method results show that only the 1fm spectral line in the M2 component is significant, while the 2fm spectral line is still submerged in noise, and the fault diagnosis information is again limited to a preliminary judgment of an outer race fault.

[0090] In summary, the modal components extracted by the above comparative methods contain less fault information than those proposed in this invention. This further demonstrates that the multi-scale multi-modal decomposition method proposed in this invention has good effectiveness and superiority in fault diagnosis of rotating components of high-end equipment, and provides a feasible technical path for fault diagnosis of rotating components of high-end equipment.

[0091] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0092] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0093] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A fault diagnosis method for rotating components based on multi-scale multi-modal decomposition, characterized in that, include: Collect the original vibration signal of the rotating component and input the preset set of scale parameters; Each scale parameter in the set of scale parameters is selected in turn, the original signal spectrum is divided into frequency bands, and the corresponding number of window filter coefficients are initialized to construct a filter bank. Using the maximization of correlation kurtosis as the optimization criterion, the filter coefficients of the filter bank are iteratively updated to obtain multiple modal components at the current scale; Calculate the energy ratio of the squared envelope spectrum of each modal component, construct a multi-scale squared envelope spectrum energy ratio indicator map, and select the optimal modal component based on the maximum value criterion; The optimal modal components are subjected to square envelope spectrum demodulation analysis to extract fault characteristic frequencies and obtain fault diagnosis results for rotating components.

2. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 1, characterized in that, Each scale parameter in the set of scale parameters is selected sequentially, and the original signal spectrum is divided into equal frequency bands. A corresponding number of window filter coefficients are initialized to construct a filter bank, including: Select the scale parameters from the set of scale parameters in sequence, and determine the total number of frequency band modal components based on the current scale parameters; The original signal spectrum is evenly divided according to the total number of modal components in the frequency band to obtain a corresponding number of segmentation boundaries. The upper and lower cutoff frequencies of each frequency band are determined by each segmentation boundary. By combining the preset filter length, multiple Hamming window filter coefficients are initialized according to the upper and lower cutoff frequencies of each frequency band to construct a finite-length unit impulse response filter bank, and the iteration counter is initialized.

3. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 2, characterized in that, The elements in the scale parameter set are arranged in ascending order, and the total number of frequency band modal components is an integer obtained by taking the power of the scale parameter as the exponent with base two and then rounding it down.

4. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 2, characterized in that, The window filtering coefficients use any one of the following window functions: Hamming window, Hanning window, Blackman window, rectangular window, triangular window, flat-top window, Bartlett window, and Taylor window.

5. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 1, characterized in that, Using the maximization of correlation kurtosis as the optimization criterion, the filter coefficients of the filter bank are iteratively updated to obtain multiple modal components at the current scale, including: The original vibration signal is convolved with the current filtering coefficients to obtain the filtered modal component signal. Estimate the fault period, combine the original vibration signal, modal component signal and fault period to update the filter coefficients, complete a single iteration and update the iteration count; Determine if the preset number of iterations has been reached. If not, repeat the convolution and coefficient update process. If the preset number of iterations has been reached, save all final modal components at the current scale.

6. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 5, characterized in that, The fault period is estimated by the moment when the autocorrelation spectrum of the original vibration signal reaches a local maximum after crossing zero.

7. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 5, characterized in that, During the process of updating the filter coefficients, an objective function is constructed with correlation kurtosis as the index, and the problem of maximizing correlation kurtosis is transformed into the problem of maximizing generalized eigenvalues. The generalized eigenvalue problem is solved using the Lagrange multiplier method, and the eigenvector corresponding to the largest eigenvalue is used as the optimal filter coefficient.

8. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 1, characterized in that, Calculate the energy ratio of the squared envelope spectrum for each modal component, construct a multi-scale squared envelope spectrum energy ratio indicator map, and select the optimal modal component based on the maximum value criterion, including: Calculate and store the squared envelope spectrum energy ratio of each final modal component at the current scale; Determine if the current scale parameter is the maximum value in the scale parameter set. If the traversal is not complete, repeat the scale selection and decomposition process. After traversing all scale parameters, a multi-scale multi-mode indicator map is constructed based on the frequency band boundaries corresponding to each scale, and the modal component corresponding to the global maximum value of the square envelope spectrum energy ratio is selected as the optimal modal component.

9. A method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 8, characterized in that, The definition of the square envelope spectrum energy is as follows: （8） In the formula: f fault For the characteristic frequency of rotating component failure, Here, H represents the frequency deviation, and E represents the harmonic number. sq Let be the squared envelope spectrum of the modal components.

10. The method for fault diagnosis of rotating components based on multi-scale multi-modal decomposition according to claim 1, characterized in that, When performing square envelope spectrum demodulation analysis on the optimal modal components, the fundamental frequency component and its harmonic components of the fault characteristic frequency are extracted, and the fault state of the rotating component is determined accordingly.

Images