Signal mode decomposition method and system for early fault diagnosis of rotating machinery

Abstract

The invention belongs to the technical field of signal processing. According to the signal mode decomposition method and system for early fault diagnosis of the rotating machinery, vibration signals are decomposed through an MCKD algorithm, a characteristic mode set of the signals is obtained, fusion is carried out according to correlation coefficients of a cross-correlation spectrum of the modes and a mode envelope power spectrum density, and finally a characteristic mode representing a fault is extracted. According to the weak fault feature signal extraction method, the feature modes are extracted by using the fault period, the feature modes are fused on the aspects of time domain and frequency domain features, the sensitivity to complex noise is low, the weak fault feature signal extraction capability is high, the precision is high, and more efficient and accurate weak fault feature signal extraction is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a signal modal decomposition method and system for early fault diagnosis of rotating machinery. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] Rotating machinery uses rotation as its core working mode and is widely used in mechanical processing and manufacturing, energy, metallurgy and other industrial fields. Its key components (gearboxes, bearings, etc.) are prone to wear, breakage and other faults due to long-term high-speed rotation and high-load conditions, which leads to economic losses and affects production safety. Early fault diagnosis technology can detect faults in time and avoid production accidents by analyzing the vibration signals of rotating machinery. Since the vibration signals of early faults of rotating machinery are relatively weak and there is noise interference, it is difficult to extract early fault signals. Therefore, separating the weak early fault signals from the vibration signals containing a lot of noise is the key to early fault diagnosis technology.

[0004] In the field of early fault diagnosis technology, the most widely used methods include empirical mode decomposition (EMD), variational mode decomposition (VMD) and eigenmode decomposition (FMD). EMD is prone to problems such as mode aliasing, endpoint effect and poor noise resistance. The root cause is that it relies on extreme point interpolation to generate envelopes, lacks a frequency band separation mechanism, and has no explicit denoising mechanism. VMD has parameter sensitivity, high computational complexity and the risk of frequency band overlap. Since its optimization model relies on preset parameters and uses iterative solutions, it is difficult to select parameters, has low computational efficiency, and is difficult to completely separate signal components with close frequency bands. FMD faces the problems of characteristic index dependence, severe noise interference and imperfect theoretical framework. The decomposition results are highly dependent on the selection of characteristic indicators and are prone to misjudgment in a strong noise environment. At the same time, there is a lack of unified mathematical convergence proof.

[0005] For example, patent number CN202410252421.7 discloses a wind turbine gearbox fault diagnosis method based on complementary ensemble empirical mode decomposition-ResNet-BiLSTM. By collecting gearbox vibration signals, complementary ensemble empirical mode decomposition (CEEMDAN) is used to extract multi-band empirical mode components, and the Pearson correlation coefficient is combined to screen strongly correlated features to reduce the dimension, and a ResNet-BiLSTM hybrid model is constructed for training and classification, ultimately achieving high-precision fault diagnosis; however, this method still has the following limitations: 1) CEEMDAN decomposition depends on the quality of vibration signals. If there is high-frequency noise or non-stationary interference in the sensor data, it may cause modal aliasing or feature distortion; 2) The ResNet-BiLSTM model has a large number of parameters and complex calculations, making it difficult to deploy on resource-constrained wind power edge devices; 3) Training requires relying on a balanced sample set. The scarcity of fault samples in actual scenarios may lead to a decrease in the generalization ability of the model; 4) The impact of multiple operating conditions of the gearbox (such as variable speed and variable load) on the vibration signal is not considered, which may lead to a decrease in cross-operating condition diagnosis performance.

[0006] Patent No. CN202210756586.9 discloses a rolling bearing early fault diagnosis method based on adaptive variational modal decomposition. By collecting bearing vibration signals, the sparrow algorithm (SSA) is used to adaptively search for the optimal parameters of variational modal decomposition (VMD), with the goal of minimizing the comprehensive fault index to achieve adaptive decomposition of the signal; then the power spectrum kurtosis of each modal component is calculated, and the components below the average value are removed to reconstruct the signal, and finally the fault type is diagnosed through envelope demodulation analysis; however, this method has the following limitations: 1) The global search capability of the sparrow algorithm is limited by the parameter range and the number of iterations, which may lead to local optimal solutions and affect the accuracy of VMD decomposition; 2) The power spectrum kurtosis threshold screening depends on the statistical characteristics of the modal components, and effective components may be mistakenly deleted under strong non-stationary noise or complex working conditions; 3) VMD is sensitive to the non-stationarity of the signal and multi-component cross-interference, and is prone to modal aliasing or under-decomposition problems; 4) Envelope demodulation analysis requires accurate prior knowledge of fault characteristic frequencies. In practical applications, if the bearing speed fluctuates or the load changes, demodulation failure may occur.

[0007] Patent No. CN202410382441.6 discloses a rolling bearing fault diagnosis method based on parameter adaptive characteristic mode decomposition. It collects vibration signals, calculates the theoretical fault characteristic frequency, uses the sparrow algorithm (SSA) to optimize the parameters of characteristic mode decomposition (FMD), and uses the envelope frequency domain signal-to-noise ratio (EFDSNR) as an indicator to screen the optimal parameter combination. Then, the signal is decomposed by FMD, and finally the fault characteristic frequency is extracted through Hilbert envelope demodulation to achieve diagnosis. However, this method has the following limitations: 1) The search range of fault characteristic frequency is limited to ±3% of the theoretical value. In actual working conditions, if the actual frequency deviation exceeds this range due to bearing wear, installation error or speed fluctuation, parameter optimization may fail; 2) The theoretical fault frequency calculation depends on accurate bearing geometric parameters. In practical applications, if the parameters are unknown or the measurement error is large, it will directly affect the accuracy of fault feature extraction; 3) Hilbert envelope demodulation depends on the significance and singularity of the fault frequency. When multiple faults coexist or strong modulation interference occurs, spectrum aliasing may occur, leading to misjudgment; 4) The EFDSNR indicator may not be able to effectively distinguish between noise and fault components under extreme non-stationary noise or complex harmonic interference, resulting in the false deletion of effective modes. Summary of the invention

[0008] In order to address the shortcomings of the prior art, the present invention provides a signal modal decomposition method and system for early fault diagnosis of rotating machinery. The vibration signal is decomposed by the MCKD algorithm to obtain the characteristic modal set of the signal, which is then fused according to the correlation coefficient of the modal cross-correlation spectrum and the modal envelope power spectral density, and finally the characteristic mode characterizing the fault is extracted, thereby achieving more efficient and accurate extraction of weak fault characteristic signals.

[0009] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a signal modal decomposition method for early fault diagnosis of rotating machinery.

[0010] A signal modal decomposition method for early fault diagnosis of rotating machinery includes the following processes: The original vibration signal is converted into a column vector and replicated to form multiple sub-signals. The entire frequency band is divided into multiple sub-bands. Each sub-channel corresponds to a sub-signal. The frequency band boundaries of each sub-band are calculated. For each sub-band, the corresponding filter coefficient is generated. Perform iterative decomposition operation on each sub-signal in column vector form to obtain the mode and filter coefficient corresponding to each sub-signal; Calculate the correlation coefficient between each mode and the original vibration signal, sort them in descending order according to the size of the correlation coefficient, start from the first mode, calculate its cross-correlation spectrum with the remaining modes, and perform the first fusion of the modes and the corresponding filter coefficients according to the peak value of the standardized cross-correlation spectrum; Calculate the power spectral density of the normalized envelope of each mode after the first fusion, sort them in descending order according to the size of the power spectral density, start from the normalized envelope power spectral density of the first mode, calculate its correlation coefficient with the normalized envelope power spectral density of the other modes after the first fusion, and perform the second fusion of the mode and the corresponding filter coefficient according to the correlation coefficient; If the number of iterations does not reach the maximum number of iterations, the first fusion operation is continued for the modal set after the second fusion; if the number of iterations reaches the maximum number of iterations, the obtained mode is used as the final feature mode set; The periodic impact fault signal is obtained according to the final characteristic mode set.

[0011] As a further limitation of the first aspect of the present invention, the peak value of the cross-correlation spectrum is found, and whether to fuse the two modes is determined based on the characteristics of the peak value. If the fusion conditions are met, the two modes are fused, and the corresponding filter coefficients are fused at the same time, and the cross-correlation spectrum of each mode is calculated in turn to complete the first fusion.

[0012] As a further limitation of the first aspect of the present invention, the fusion conditions include: the number of peaks is greater than or equal to a first set threshold, the maximum peak is greater than a second set threshold, and the peak intervals are consistent.

[0013] As a further limitation of the first aspect of the present invention, when the correlation coefficient is greater than a third set threshold, the two modes are fused, and the corresponding filter coefficients are fused at the same time to complete the second fusion.

[0014] As a further limitation of the first aspect of the present invention, during the first fusion and the second fusion, the process of fusing the two modes and fusing the corresponding filter coefficients is the same, including: The two modes are directly added together as a new mode, and the filter coefficients corresponding to the two modes are added together to obtain the filter coefficients corresponding to the new mode.

[0015] As a further limitation of the first aspect of the present invention, for each sub-band, generating a corresponding filter coefficient includes: ; in, For the The filter coefficients corresponding to the sub-bands; It is the Hanning window; is the filter length; is the number of sub-bands; is a constant; Representative The frequency band boundaries corresponding to the sub-bands are The function is used to generate the filter coefficients.

[0016] As a further limitation of the first aspect of the present invention, the entire frequency band is divided into subbands, the set of band boundaries for: .

[0017] As a further limitation of the first aspect of the present invention, for any sub-signal Perform iterative decomposition operations, including: If no subsignal is provided Fault period, the fault period is estimated by Hilbert transform and autocorrelation function; Construct a multi-channel delay matrix; For the first iteration, the multi-channel delay matrix is ​​used to calculate the vector , according to the calculation vector Constructing the delay matrix , according to the delay matrix Calculate the weights, update the filter coefficients according to the weights and normalize them, and output the eigenmodes and the corresponding filter coefficients corresponding to this iteration round; Perform multiple rounds of iterations until the characteristic mode corresponding to the maximum iteration round and the corresponding filter coefficients are output.

[0018] As a further limitation of the first aspect of the present invention, the fault period is estimated by Hilbert transform and autocorrelation function ,include: , where is the position of the first zero crossing point of the autocorrelation spectrum, Sub-signal The autocorrelation spectrum of .

[0019] In a second aspect, the present invention provides a signal modal decomposition system for early fault diagnosis of rotating machinery.

[0020] A signal modal decomposition system for early fault diagnosis of rotating machinery, comprising: The filter bank generating unit is configured to: convert the original vibration signal into a column vector and replicate it into multiple sub-signals, divide the entire frequency band into multiple sub-bands, each sub-channel corresponds to a sub-signal, and calculate the frequency band boundary of each sub-band, and generate a corresponding filter coefficient for each sub-band; The iterative decomposition operation unit is configured to: perform iterative decomposition operation on each sub-signal in the form of a column vector to obtain a mode and a corresponding filter coefficient corresponding to each sub-signal; The first fusion unit is configured to: calculate the correlation coefficient between each mode and the original vibration signal, sort them in descending order according to the size of the correlation coefficients, start from the first mode, calculate its cross-correlation spectrum with the remaining modes, and perform the first fusion of the modes and the corresponding filter coefficients according to the peak value of the standardized cross-correlation spectrum; The second fusion unit is configured to: calculate the power spectral density of the normalized envelope of each mode after the first fusion, sort them in descending order according to the size of the power spectral density, start from the normalized envelope power spectral density of the first mode, calculate its correlation coefficient with the normalized envelope power spectral density of the remaining modes after the first fusion, and perform a second fusion of the mode and the corresponding filter coefficient according to the correlation coefficient; The characteristic modal set generating unit is configured to: if the number of iterations does not reach the maximum number of iterations, continue the first fusion operation on the modal set after the second fusion; if the number of iterations reaches the maximum number of iterations, use the obtained modality as the final characteristic modal set; The fault signal extraction unit is configured to obtain a periodic impact fault signal according to a final characteristic mode set.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention innovatively proposes a signal modal decomposition strategy for early fault diagnosis of rotating machinery. The vibration signal is decomposed by the MCKD algorithm to obtain the characteristic mode set of the signal, which is then fused according to the correlation coefficient of the modal cross-correlation spectrum and the modal envelope power spectral density. Finally, the characteristic mode characterizing the fault is extracted, achieving more efficient and accurate extraction of weak fault characteristic signals.

[0022] 2. The present invention innovatively proposes a signal modal decomposition strategy for early fault diagnosis of rotating machinery. It uses the fault cycle to extract characteristic modes and fuses the characteristic modes in the time domain and frequency domain characteristics respectively. It has low sensitivity to complex noise, strong ability to extract weak fault characteristic signals, and high accuracy.

[0023] 3. The present invention innovatively proposes a signal modal decomposition strategy for early fault diagnosis of rotating machinery, adopts a fusion algorithm to reduce the number of modes, and fuses multiple modes in each iteration, which has higher operating efficiency than the traditional FMD method.

[0024] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 A schematic flow chart of a signal modal decomposition method for early fault diagnosis of rotating machinery provided in Example 1 of the present invention; Figure 2 The original signal diagram provided by Embodiment 1 of the present invention; Figure 3 A schematic diagram of the MFMD decomposition result (iteration 5 times) provided in Example 1 of the present invention; wherein (a) is a time domain diagram of mode M1, (b) is a frequency domain diagram of mode M1, (c) is a time domain diagram of mode M2, and (d) is a frequency domain diagram of mode M2; Figure 4 Schematic diagram of the FMD decomposition result (iteration 5 times) provided in Example 1 of the present invention; wherein, (a) is a time domain diagram of mode M1, (b) is a frequency domain diagram of mode M1, (c) is a time domain diagram of mode M2, (d) is a frequency domain diagram of mode M2, (e) is a time domain diagram of mode M3, (f) is a frequency domain diagram of mode M3, (g) is a time domain diagram of mode M4, (h) is a frequency domain diagram of mode M4, (i) is a time domain diagram of mode M5, (j) is a frequency domain diagram of mode M5, (k) is a time domain diagram of mode M6, (l) is a frequency domain diagram of mode M6, (m) is a time domain diagram of mode M7, and (n) is a frequency domain diagram of mode M7; Figure 5 Schematic diagram of the FMD decomposition result (iteration 13 times) provided in Example 1 of the present invention; wherein (a) is a time domain diagram of mode M1, (b) is a frequency domain diagram of mode M1, (c) is a time domain diagram of mode M2, and (d) is a frequency domain diagram of mode M2; Figure 6 A schematic diagram of a signal modal decomposition system for early fault diagnosis of rotating machinery provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0029] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0030] Embodiment 1: As described in the background technology, although the existing characteristic mode decomposition method can adaptively and accurately separate the early fault components of the vibration signal and is widely used in the field of early fault diagnosis technology, the method has low operating efficiency and is not conducive to real-time monitoring and rapid diagnosis. In view of this, the present implementation method provides a signal mode decomposition method for early fault diagnosis of rotating machinery, which decomposes the vibration signal through the MCKD algorithm to obtain the characteristic mode set of the signal, and then fuses it according to the correlation coefficient of the cross-correlation spectrum of the mode and the power spectrum density of the modal envelope, and finally extracts the characteristic mode that characterizes the fault. The experiment compared the computational efficiency of MFMD and FMD, and the results showed that the computational efficiency of MFMD proposed in the present invention is significantly improved compared with FMD.

[0031] Specifically, Figure 1 As shown, the following process is included: Step S1: Initialization process of original vibration signal and filter, initializing the original signal, dividing the frequency band and designing the FIR filter. ( is the signal length) is initialized as a column vector, the signal decomposition range is constrained by frequency band segmentation, and the initial filter bank is designed , is the number of sub-bands; Step S2: Decomposition process, also known as iterative mode extraction process, extracts characteristic modes band by band through the maximum correlation kurtosis deconvolution (MCKD) algorithm , dynamically update the filter , is the total number of eigenmodes in this iteration; Step S3: modal fusion post-processing process, the characteristic modes decomposed in step S2 are subjected to two similarity fusion operations according to the modal cross-correlation sequence and the power spectrum density similarity number, and the redundant modes are fused. When the maximum number of iterations is reached (i.e. ), output the final eigenmode set .

[0032] In step S1 of this implementation, specifically, it includes: S101: The original vibration signal Convert to a column vector: (1).

[0033] in, is the signal length.

[0034] S102: Divide the entire frequency band into sub-bands, band boundaries : (2).

[0035] S103: For each sub-band, call The function generates the filter coefficients: (3); In the formula, is the Hanning window, according to the filter length , call Functions and generation; is a small constant to avoid frequency band overlap; represents the band boundary corresponding to the ith sub-band, Representative The filter coefficients corresponding to the sub-bands.

[0036] S104: The original vibration signal copy Get the signal set , the signal set As the input signal set of the MCKD decomposition algorithm in S2.

[0037] In step S2 of this implementation, specifically, it includes: S201: Enter the MCKD decomposition algorithm, for each ( ) to perform iterative decomposition operations.

[0038] S202: If no input signal is provided Failure cycle , estimate the fault period by Hilbert transform and autocorrelation function, call The function calculates the envelope of the input signal: (4); In the formula, To take the average value, |hilbert(x)| is the envelope (a series), and taking the average value is to calculate the DC component of the envelope. To obtain The model (signal After Hilbert transform, it is converted into a complex signal), calling Function to calculate signal envelope The autocorrelation spectrum of .

[0039] Step S203: Determine Failure cycle : (5); In the formula, is the position of the first zero crossing of the autocorrelation spectrum.

[0040] Step S204: Construct a multi-channel delay matrix , and calculate the inverse of the delay matrix : (6); In the formula, Represents the row index, (Right now ) is the filter length; Represents the column index, is the input signal length; Represents the third dimension index, is the polynomial order, the default value is 3, means a three-dimensional matrix The two-dimensional matrix of the first layer is a signal page delayed by T points.

[0041] S205: Enter the iterative decomposition process in the MCKD algorithm, using the delay matrix Calculate the vector: (7); S206: Update the filter, first construct the delay matrix : (8); In the formula, is the input signal length, , and then calculate the weight: (9); In the formula, , , , , , The calculation is to calculate Each element of the matrix, Represents a column vector used with Do matrix multiplication, It means that the two-dimensional matrix of the first layer of the three-dimensional matrix XmT is the signal page delayed by mT points.

[0042] Finally update the filter and normalize: (10).

[0043] S207: Vector of application formula (7) Update cycle failure according to S203 , update the delay matrix according to S204 and the inverse matrix .

[0044] S208: Calculate the output signal of this iteration and update the number of iterations: (11); In the formula, is the function for filtering the signal, Represents the number of iterations.

[0045] S209: Return to S205 until the maximum number of iterations is reached to obtain the filter Eigenmode of the frequency band (denoted as ) and its corresponding filter (denoted as ),at this time is the maximum number of iterations.

[0046] In step S3 of this implementation, specifically, it includes: S301: Perform the first fusion operation. The function calculates the correlation coefficient between each mode and the original vibration signal: (12).

[0047] S302: Sort the modes in descending order according to the magnitude of the correlation coefficients between the modes and the original vibration signals, with the purpose of preferentially retaining the modes with strong correlation in the subsequent modal fusion, and merging the modes with weak correlation into the modes with strong correlation.

[0048] S303: According to the order of the modes in step S302, call Function, starting from the first mode, calculates its cross-correlation spectrum with the remaining modes ,Pick ( For time delay, is 80% of the original signal length), and To standardize: (13); In the formula, and represents the mean value of the signal, and Represents the standard deviation of the signal.

[0049] S304: Call Function search The peak value (the minimum peak prominence is set to 0.2), according to The characteristics of the peaks determine whether to fuse the two modes. The specific conditions for fusing the two modes are: 1) the number of peaks ≥ 3 (i.e., the first set threshold); 2) the maximum peak value > 0.25 (i.e., the second set threshold); 3) the peak interval is consistent (if the peak is greater than 15, only the first 15 peaks are tested, using 3 The two modes that meet the fusion conditions ( and ) and its corresponding filter coefficients ( and ) to perform the fusion operation and obtain the result of the first fusion operation and : (14); The operations of S303 and S304 are repeated until all modalities are involved in the fusion.

[0050] S305: Perform the second fusion operation. Use each mode after the first fusion as the input signal of the second fusion operation, calculate the power spectral density (PSD) of each mode envelope, and normalize it.

[0051] Compute the envelope signal of the mode: (15); In the formula, For the mode after the first fusion, call The function calculates the PSD of the envelope signal: (16); In the formula, represents the sampling frequency of the signal, It means using a 256-point Hamming window; 256*3 / 4 means the number of overlapping points is 192; 256 means the FFT length is 256 points;' ' indicates calculation of the bilateral spectrum, is the output parameter, representing the frequency vector; It is the output parameter, which represents the power spectral density (PSD).

[0052] The power spectral density Convert to decibel units and normalize: (17).

[0053] in, represents the power spectral density, represent The minimum value of represent The maximum value of .

[0054] S306: Sort the input modes, which is the same as the operations in S301 and S302.

[0055] S307: According to the order of S306 mode, call Function, starting from the PSD of the first modal normalized envelope, calculates the correlation coefficient with the PSD of the remaining modal normalized envelopes .

[0056] S308: If (i.e., the third set threshold), the two modes and their corresponding filters are fused, the same as the fusion operation part in S304, and the PSD of the fused mode envelope is calculated. S307 and S308 are repeated until all modes are involved in the fusion.

[0057] S308: After completing two fusion operations, if the number of iterations has not reached the maximum number of iterations, return to step S2 and continue to perform decomposition and fusion operations on the decomposed and fused mode set; if the number of iterations has reached the maximum number of iterations, the mode obtained in S308 is used as the final feature mode set. .

[0058] Based on specific data, this implementation provides the following detailed exemplary steps: Step 1: Import the original vibration signal (sampling frequency is 25600Hz, duration is 1s), such as Figure 2 As shown, the initialization vibration signal is (n is the signal length) is a column vector, the number of split frequency bands is set to 10, and the frequency band boundaries are generated , set the filter length to 30, call The function generates the Hanning window and then calls Function Design Filter , the input signal is copied into 10 copies for extracting 10 eigenmodes.

[0059] Step 2: Enter the iteration process of the MFMD algorithm and apply the MCKD decomposition algorithm to decompose each input signal ( The maximum value of is the number of signals to be decomposed in the current iteration, and the initial value is 10) to obtain the characteristic mode (The same as the number of input signals), the MCKD decomposition algorithm is described as follows: (1) Call Calculate the envelope of the input signal, call Calculate the signal autocorrelation spectrum, and take the fault period as the moment when the autocorrelation spectrum reaches the maximum value after crossing the zero point; (2) Constructing the delay matrix and the inverse matrix ; (3) Enter the iterative process within the MCKD algorithm and calculate the vector , by vector Constructing the delay matrix ; (4) Calculate weights, update filters and normalize filters, and call The function calculates the output signal of this iteration , is the number of iterations; (5) Use the obtained characteristic mode as input to update the fault cycle and recalculate the delay matrix and the inverse matrix ; (6) Determine whether the maximum number of iterations of the MCKD algorithm has been reached. If not, return to step 2 (2); if reached, This is the characteristic mode decomposed by the MCKD algorithm , the maximum number of iterations of the MCKD algorithm is 2 times less than the maximum number of iterations of the MFMD algorithm.

[0060] Step 3: Apply the fusion algorithm to perform similar fusion on the characteristic modes, as described below: (1) First enter the first fusion process and calculate the correlation coefficient between each input mode and the original vibration signal , sort the input modes in descending order according to the obtained correlation coefficient; (2) Starting from the first mode, determine in order whether the subsequent modes are integrated with it, and call The function calculates the cross-correlation spectrum of the two modes, and determines whether to fuse them according to the characteristics of the peak values ​​of the cross-correlation spectrum of the two modes. The conditions for fusing the two modes are: 1) the number of peaks ≥ 3 (i.e. the first set threshold); 2) the maximum peak value > 0.25 (i.e. the second set threshold); 3) the peak interval is consistent (if the peak value is greater than 15, only the first 15 peaks are tested, using 3 The two modes that meet the conditions and their corresponding filters are merged; (3) The characteristic mode obtained after the first fusion is used as the input mode, the power spectrum density of each mode envelope is calculated and normalized to obtain , sort the input modalities in the same way as step (1) of step 3; (4) Starting from the first mode, determine in order whether the subsequent modes are integrated with it, and call The function calculates the correlation coefficient of the envelope power spectral density after normalization of the two modes ,like (i.e., the second set threshold), the two modes and the corresponding filters are fused, and the fusion operation is completed.

[0061] (5) Determine whether the number of iterations of the MFMD algorithm exceeds the maximum number of iterations. If not, return to step 2 and use the characteristic modes obtained in this iteration as the input signal to perform decomposition and fusion operations again. If it exceeds, use the characteristic modes obtained in this iteration as the final decomposition mode set. , determine the mode that characterizes the early fault according to the characteristics of the fault signal.

[0062] The number of iterations of the MFMD algorithm is 5. The decomposition and fusion results of the wind turbine gearbox vibration signal of the present invention are as follows: Figure 3 As shown. For the FMD signal decomposition algorithm, the number of iterations is 5, and the other parameters and the original input signal are set the same as the MFMD algorithm. The final decomposition results are as follows: Figure 4 , Figure 5 shown.

[0063] Depend on Figure 3 It can be seen that these are characteristic modes obtained by MFMD decomposition (iteration 5 times). For mode M1, it can be seen from the figure that M1 has periodic high-peak impacts, indicating that the gear teeth are partially damaged, causing some teeth to produce greater impacts when meshing. The time interval between the impacts is consistent with the gear rotation frequency. For mode M2, it can be seen from the figure that M2 shows continuous and intensive vibrations, and M2 is a noise signal.

[0064] In summary, it can be seen that after 5 iterations, the MFMD algorithm can better extract the periodic impact fault signal from the vibration signal of the wind turbine gearbox (the characteristic of the early fault signal of the gearbox is periodic impact, and it is obvious that mode M1 is a periodic impact signal, that is, the early fault signal of the gearbox).

[0065] like Figure 4 As shown in the figure, the characteristic modes obtained by FMD decomposition (5 iterations) are Figure 4 There is no obvious distinction between the modes M1, M2, M3, M4, M5, M6 and M7 in the time domain and frequency domain diagrams, indicating that the characteristic modes representing the early faults have not been extracted. Figure 5 As shown in the figure, the characteristic modes obtained by FMD decomposition (13 iterations) are similar to the characteristic modes obtained by the MFMD method (modes M1 and M2 are obtained). The decomposition obtains the periodic vibration impact characteristics that characterize the early faults, but more iterations are required. In summary, it can be seen that the MFMD algorithm has a great advantage in computational efficiency over the FMD algorithm.

[0066] Embodiment 2: like Figure 6 As shown, this implementation provides a signal modal decomposition system for early fault diagnosis of rotating machinery, including: The filter bank generating unit is configured to: convert the original vibration signal into a column vector and replicate it into multiple sub-signals, divide the entire frequency band into multiple sub-bands, each sub-channel corresponds to a sub-signal, and calculate the frequency band boundary of each sub-band, and generate a corresponding filter coefficient for each sub-band; The iterative decomposition operation unit is configured to: perform iterative decomposition operation on each sub-signal in the form of a column vector to obtain a mode and a corresponding filter coefficient corresponding to each sub-signal; The first fusion unit is configured to: calculate the correlation coefficient between each mode and the original vibration signal, sort them in descending order according to the size of the correlation coefficients, start from the first mode, calculate its cross-correlation spectrum with the remaining modes, and perform the first fusion of the modes and the corresponding filter coefficients according to the peak value of the standardized cross-correlation spectrum; The second fusion unit is configured to: calculate the power spectral density of the normalized envelope of each mode after the first fusion, sort them in descending order according to the size of the power spectral density, start from the normalized envelope power spectral density of the first mode, calculate its correlation coefficient with the normalized envelope power spectral density of the remaining modes after the first fusion, and perform a second fusion of the mode and the corresponding filter coefficient according to the correlation coefficient; The characteristic modal set generating unit is configured to: if the number of iterations does not reach the maximum number of iterations, continue the first fusion operation on the modal set after the second fusion; if the number of iterations reaches the maximum number of iterations, use the obtained modality as the final characteristic modal set; The fault signal extraction unit is configured to obtain a periodic impact fault signal according to a final characteristic mode set.

[0067] The specific working process of each unit is described in Example 1 and will not be repeated here.

[0068] It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0069] According to another embodiment of the present application, the system described in this embodiment can be constructed, and the method of Example 1 of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A signal modal decomposition method for early fault diagnosis of rotating machinery, characterized in that: The process includes: The original vibration signal is converted into a column vector and replicated to form multiple sub-signals. The entire frequency band is divided into multiple sub-bands. Each sub-channel corresponds to a sub-signal. The frequency band boundaries of each sub-band are calculated. For each sub-band, the corresponding filter coefficient is generated. Perform iterative decomposition operation on each sub-signal in column vector form to obtain the mode and filter coefficient corresponding to each sub-signal; Calculate the correlation coefficient between each mode and the original vibration signal, sort them in descending order according to the size of the correlation coefficient, start from the first mode, calculate its cross-correlation spectrum with the remaining modes, and perform the first fusion of the modes and the corresponding filter coefficients according to the peak value of the standardized cross-correlation spectrum; Calculate the power spectral density of the normalized envelope of each mode after the first fusion, sort them in descending order according to the size of the power spectral density, start from the normalized envelope power spectral density of the first mode, calculate its correlation coefficient with the normalized envelope power spectral density of the other modes after the first fusion, and perform the second fusion of the mode and the corresponding filter coefficient according to the correlation coefficient; If the number of iterations does not reach the maximum number of iterations, the first fusion operation is continued for the modal set after the second fusion; if the number of iterations reaches the maximum number of iterations, the obtained mode is used as the final feature mode set; The periodic impact fault signal is obtained according to the final characteristic mode set.

2. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 1, characterized in that: Find the peak of the cross-correlation spectrum and decide whether to fuse the two modes based on the characteristics of the peak. If the fusion conditions are met, the two modes are fused, and the corresponding filter coefficients are fused at the same time. The cross-correlation spectrum of each mode is calculated in turn to complete the first fusion.

3. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 2, characterized in that: The fusion conditions include: the number of peaks is greater than or equal to a first set threshold, the maximum peak is greater than a second set threshold, and the peak intervals are consistent.

4. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 1, characterized in that: When the correlation coefficient is greater than the third set threshold, the two modes are fused, and the corresponding filter coefficients are fused at the same time to complete the second fusion.

5. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 2 or 4, characterized in that: The two modes are fused, and the corresponding filter coefficients are fused at the same time, including: The two modes are directly added together as a new mode, and the filter coefficients corresponding to the two modes are added together to obtain the filter coefficients corresponding to the new mode.

6. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 1, characterized in that: For each sub-band, generate the corresponding filter coefficients, including: ； in, For the The filter coefficients corresponding to the sub-bands; It is the Hanning window; is the filter length; is the number of sub-bands; is a constant; Representative The frequency band boundaries corresponding to the sub-bands are The function is used to generate the filter coefficients.

7. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 6, characterized in that: Divide the entire frequency band into subbands, the set of band boundaries for: 。 8. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 1, characterized in that: For any sub-signal Perform iterative decomposition operations, including: If no subsignal is provided Fault period, the fault period is estimated by Hilbert transform and autocorrelation function; Construct a multi-channel delay matrix; For the first iteration, the multi-channel delay matrix is ​​used to calculate the vector , according to the calculation vector Constructing the delay matrix , according to the delay matrix Calculate the weights, update the filter coefficients according to the weights and normalize them, and output the eigenmodes and the corresponding filter coefficients corresponding to this iteration round; Perform multiple rounds of iterations until the characteristic mode corresponding to the maximum iteration round and the corresponding filter coefficients are output.

9. The signal modal decomposition method for early fault diagnosis of rotating machinery according to claim 8, characterized in that: Estimation of Failure Period by Hilbert Transform and Autocorrelation Function ,include: ； In the formula, is the position of the first zero crossing point of the autocorrelation spectrum, Sub-signal The autocorrelation spectrum of .

10. A signal modal decomposition system for early fault diagnosis of rotating machinery, characterized in that: include: The filter bank generating unit is configured to: convert the original vibration signal into a column vector and replicate it into multiple sub-signals, divide the entire frequency band into multiple sub-bands, each sub-channel corresponds to a sub-signal, and calculate the frequency band boundary of each sub-band, and generate a corresponding filter coefficient for each sub-band; The iterative decomposition operation unit is configured to: perform iterative decomposition operation on each sub-signal in the form of a column vector to obtain a mode and a corresponding filter coefficient corresponding to each sub-signal; The first fusion unit is configured to: calculate the correlation coefficient between each mode and the original vibration signal, sort them in descending order according to the size of the correlation coefficients, start from the first mode, calculate its cross-correlation spectrum with the remaining modes, and perform the first fusion of the modes and the corresponding filter coefficients according to the peak value of the standardized cross-correlation spectrum; The second fusion unit is configured to: calculate the power spectral density of the normalized envelope of each mode after the first fusion, sort them in descending order according to the size of the power spectral density, start from the normalized envelope power spectral density of the first mode, calculate its correlation coefficient with the normalized envelope power spectral density of the remaining modes after the first fusion, and perform a second fusion of the mode and the corresponding filter coefficient according to the correlation coefficient; The characteristic modal set generating unit is configured to: if the number of iterations does not reach the maximum number of iterations, continue the first fusion operation on the modal set after the second fusion; if the number of iterations reaches the maximum number of iterations, use the obtained modality as the final characteristic modal set; The fault signal extraction unit is configured to obtain a periodic impact fault signal according to a final characteristic mode set.

Images