Rotary machinery composite fault diagnosis method and system

Abstract

The invention provides a composite fault diagnosis method and system for a rotating machine, and relates to the technical field of mechanical signal processing and equipment fault diagnose.The method comprises the steps that fault signals of an inner ring and an outer ring of a rolling bearing are collected, a resonant frequency band is positioned through a sliding window filter, modal optimization indexes are calculated, and modals with fault characteristics are screened; estimating a fault characteristic frequency by means of an envelope harmonic product spectrum, and removing non-converged sub-signals; separating a composite fault signal based on an improved sparse Bayesian learning (SBL) classification model in combination with variational Bayesian reasoning (VBI); and finally, judging the fault type through envelope demodulation. The problems that under strong noise interference, composite fault features are difficult to separate, and a traditional method depends on prior knowledge are solved, efficient and accurate rotary machine composite fault diagnosis is achieved, and noise immunity is high.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical signal processing and equipment fault diagnosis, and particularly to a method and system for diagnosing compound faults of rotating machinery. Background Art

[0002] In the modern industrial system, rotating machinery, as a key equipment, is widely used in many important fields such as energy power, railway transportation, aerospace, etc. As the core support component of rotating machinery, the operating state of rolling bearings is directly related to the reliability and stability of the entire equipment. Once a rolling bearing fails, it is very likely to cause equipment shutdown, production interruption, and even serious safety accidents, resulting in huge economic losses.

[0003] Under actual working conditions, the operating environment of rotating machinery is extremely complex, and its compound fault vibration signals exhibit non-stationary and non-linear characteristics. The signals generated by different faults are cross-coupled with each other, making the separation and diagnosis of fault characteristics face severe challenges. At the same time, strong noise interference is widespread, further masking the fault signals, making it difficult to effectively extract and identify fault characteristics, and greatly increasing the difficulty of compound fault diagnosis.

[0004] Existing decomposition methods have obvious defects when dealing with the diagnosis of compound faults of rotating machinery. On the one hand, these methods highly rely on accurate theoretical fault characteristics and prior knowledge of decomposition moduli. In actual application scenarios, due to the diversity and uncertainty of equipment operating conditions, it is not easy to accurately obtain this prior information. Once the prior information is biased, the accuracy of the diagnosis results will be greatly reduced. On the other hand, traditional signal processing methods usually adopt frequency band decomposition technology. However, when there are signals with similar frequencies in the same frequency band, they will overlap with each other, making it difficult to achieve effective separation. Moreover, as the noise intensity gradually increases, the anti-interference ability of traditional methods is insufficient, and it cannot guarantee the accuracy and reliability of the diagnosis results.

[0005] In summary, it is of great practical significance to develop a method and system for diagnosing compound faults of rotating machinery that can operate efficiently under strong noise interference. This can not only improve the safety and reliability of the operation of rotating machinery, but also provide a strong guarantee for the stable production of related industrial fields. Summary of the Invention

[0006] Therefore, the present invention provides a method and system for diagnosing compound faults of rotating machinery, which are used to solve the problems in the prior art that the diagnosis of compound faults of rotating machinery relies too much on accurate theoretical fault characteristics and prior decomposition moduli, has poor robustness under strong noise interference, traditional frequency band decomposition methods are difficult to separate signals due to the overlap of signals with similar frequencies in the same frequency band, and the accuracy cannot be guaranteed when the noise increases, resulting in the difficulty of effectively separating and identifying the compound fault characteristics of rolling bearings.

[0007] To solve the above problems, an embodiment of the present invention provides a method for diagnosing compound faults in rotating machinery, the method comprising: S1: Collect the fault signals of the inner and outer rings of the rolling bearing, locate the resonant frequency band of the fault signal based on the sliding window filter, separate the sub-signal set, calculate the modal optimization index values of each sub-signal successively, select the top four sub-signals with the largest modal optimization index values, and eliminate the redundant components by calculating the correlation coefficient between signals, and retain the modes containing fault characteristics; S2: Estimate the fault characteristic frequency of the resonant frequency band based on envelope demodulation, use the envelope harmonic product spectrum to estimate the frequency of the remaining sub-signals, eliminate the non-convergent sub-signals, and use the envelope harmonic product spectrum value of the remaining signals as the fault characteristic frequency; S3: Classify the optimal sub-signals based on the improved sparse Bayesian learning classification model, update and solve the maximum a posteriori probability through variational Bayesian inference, and realize the separation of compound fault signals; S4: Perform envelope demodulation on the separated signals, and judge the specific fault type according to the presence or absence of the bearing fault characteristic frequency.

[0008] Preferably, the modal optimization index (MOI) value is composed of kurtosis (Kurt), envelope spectrum kurtosis (ESK), and mutual information (MI), and its calculation method is: ; Where ; ; ; In the formula, is the modal optimization index value; is kurtosis; ESK is envelope spectrum kurtosis; MI is mutual information; is the time-domain sequence of the signal; is the mean value of the time-domain signal; is the signal length; is the time-series index; represents the amplitude of the envelope spectrum at frequency ; is the mean value of the envelope spectrum; is the signal entropy value; is the signal when known, the conditional entropy of the signal .

[0009] Preferably, the initialization of the sliding window filter includes: Divide the original signal into segments, and the length of each segment is , and determine the upper and lower cut-off frequencies of each segment according to the sampling frequency and construct finite impulse response filters, where the upper and lower cut-off frequencies and and satisfy: ; In the formula, represents the filter index, is the total number of filters.

[0010] Preferably, the updating method of the sliding window filter includes: Using the correlation kurtosis as the objective function to update the filter bank, and approximating the filtered signal corresponding to the maximum value of the correlation kurtosis by iteratively solving the generalized eigenvalue problem, where the update formula is: ; ; In the formula, is the correlation kurtosis value, is the shift number; is the signal component output by the th filter; is the sampling period; signal length; is the th filter coefficient vector; is the filter length; , , , all represent indices; represents the constraint condition.

[0011] Preferably, the method for estimating the frequency of the remaining sub-signal using the envelope harmonic product spectrum is: ; In the formula, is the envelope harmonic product spectrum value, , , , , are the amplitudes of the Fourier transform of the signal envelope; is the number of harmonics; represents the index; is the angular frequency.

[0012] Preferably, the improved sparse Bayesian learning classification model adopts a block sparse structure, and its mathematical expression is: ; Joint probability density The formula is: ; In the formula, is the conditional probability of the block sparse structure; is the block structure matrix; is the fault feature component, is the control hyperparameter for sparsity, and the subscript represents the in-block position; is the block structure weight coefficient, defined as the degree of dependence on adjacent sparse parameters; is a Gaussian distribution with a mean of 0 and a variance of ; is the fault signal; is the observed signal; is the set of latent variables; is the likelihood function; is the noise precision of the prior distribution, is the prior distribution of the sparse parameter ; is the conditional probability of the block sparse structure; is the prior distribution of the block structure matrix; is the signal length; , both represent indices.

[0013] Preferably, the improved sparse Bayesian learning classification model classifies the optimal sub-signals, and updates and solves the maximum a posteriori probability through variational Bayesian inference, specifically including: Design a relaxed evidence lower bound formula, and iteratively update the latent variables through variational Bayesian inference to solve the approximate solution of the maximum a posteriori probability, where the relaxed evidence lower bound calculation formula is: ; where

[0014] In the formula, is the relaxed evidence lower bound, is the variational distribution, used to approximate the true posterior distribution that is difficult to calculate ; is the generalized likelihood function; is the observed likelihood function; is the conditional probability; is the set of latent variables; is the variational parameter.

[0015] Preferably, the relaxed evidence lower bound minimizes the variational distribution by introducing the KL divergence from the true posterior distribution . The specific optimization formula is: ; And by decomposing the latent variable update rule, the computational complexity is reduced.

[0016] Preferably, the method for envelope demodulation of the separated signal is as follows: First, perform Hilbert transform on the separated signal to construct an analytic signal; then, extract the amplitude of the analytic signal; finally, perform Fourier transform on the amplitude to obtain the envelope spectrum, and observe the characteristic frequency through the envelope spectrum.

[0017] The embodiment of the present invention also provides a rotating machinery compound fault diagnosis system, which is used to implement the above-mentioned rotating machinery compound fault diagnosis method, and specifically includes: A frequency band adaptive positioning module, which is used to collect the fault signals of the inner and outer rings of the rolling bearing, locate the resonant frequency band of the fault signal based on the sliding window filter, separate the sub-signal set, calculate the modal optimization index values of each sub-signal successively, select the top four sub-signals with the largest modal optimization index values, and eliminate redundant components by calculating the correlation coefficient between signals, and retain the mode containing fault characteristics; A characteristic frequency estimation module, which is used to estimate the fault characteristic frequency of the resonant frequency band based on envelope demodulation, estimate the frequency of the remaining sub-signals by using the envelope harmonic product spectrum, eliminate the unconverged sub-signals, and use the envelope harmonic product spectrum value of the remaining signals as the fault characteristic frequency; A fault separation module, which is used to classify the optimal sub-signals based on the improved sparse Bayesian learning classification model, update and solve the maximum a posteriori probability through variational Bayesian inference, and realize the separation of the compound fault signal; An envelope demodulation module, which is used to perform envelope demodulation on the separated signal, and judge the specific fault type according to whether there is a bearing fault characteristic frequency.

[0018] From the above technical solutions, it can be seen that the present invention application has the following beneficial effects: The embodiment of the present invention provides a rotating machinery compound fault diagnosis method and system, which innovatively proposes an adaptive modal decomposition and an improved sparse Bayesian method. The adaptive modal decomposition can realize the adaptive selection of the decomposition mode with the help of the filter bank, and at the same time estimate the fault characteristic frequency, which can solve the problem that the existing methods rely too much on the accurate theoretical fault characteristic frequency and the prior knowledge of the decomposition modulus. The improved sparse Bayesian learning method has strong anti-noise interference, and the proposed relaxed evidence lower bound can effectively reduce the computational complexity of the model, and finally realize more accurate and rapid rotating machinery compound fault diagnosis. Description of the Drawings

[0019] In order to more clearly illustrate the implementation cases of the present invention or the technical solutions in the prior art, the following is a brief description of the drawings required for use in the embodiments. By referring to the drawings, the features and advantages of the present invention will be more clearly understood. The drawings are schematic and should not be understood as limiting the present invention in any way. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 A flow chart of a rotating machinery compound fault diagnosis method provided by the present invention; Figure 2 It is a waveform diagram of the original rolling bearing inner and outer ring fault signals in the present invention; Figure 3 is an envelope diagram of a rolling bearing separation fault signal in the present invention; Figure 4 The present invention provides a block diagram of a rotating machinery composite fault diagnosis system. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Embodiment 1:

[0022] In order to solve the problem that the diagnosis of rotating machinery composite faults in the prior art is too dependent on accurate theoretical fault features and decomposition modulus priors, and has poor robustness under strong noise interference, the traditional frequency band decomposition method is difficult to separate signals due to the overlap of similar frequency signals in the same frequency band, and the accuracy cannot be guaranteed when the noise increases, resulting in the difficulty in effectively separating and identifying the composite fault features of rolling bearings. Figure 1 As shown, the present invention proposes a rotating machinery composite fault diagnosis method, the method comprising: S1: Collect the fault signals of the inner and outer rings of the rolling bearing, locate the resonant frequency band of the fault signal based on the sliding window filter, separate the sub-signal set, calculate the modal optimization index value of each sub-signal one by one, select the first four sub-signals with the largest modal optimization index value, and eliminate the redundant components by calculating the correlation coefficient between the signals, and retain the mode containing the fault characteristics; S2: Estimate the fault characteristic frequency of the resonant frequency band based on envelope demodulation, use the envelope harmonic product spectrum to estimate the frequency of the remaining sub-signals, remove the unconverged sub-signals, and use the envelope harmonic product spectrum value of the remaining signal as the fault characteristic frequency; S3: Classify the optimal sub-signals based on the improved sparse Bayesian learning classification model, update and solve the maximum a posteriori probability through variational Bayesian inference, and achieve the separation of composite fault signals; S4: Perform envelope demodulation on the separated signals, and judge the specific fault type according to whether there is the bearing fault characteristic frequency.

[0023] As can be seen from the above technical solutions, the present invention proposes a method for diagnosing composite faults of rotating machinery, which uses a filter bank to accurately locate the fault frequency band, and at the same time estimates the fault characteristic frequency through envelope harmonic product spectrum, realizing the adaptive selection of the fault characteristic frequency and the decomposition modulus; in the process of separating faults, a sparse Bayesian learning classification model is established, a relaxed evidence lower bound is established, and the maximum a posteriori probability is solved through variational Bayesian, accurately separating faults and enhancing the noise resistance of the method; improving the frequency clear characterization and anti-noise interference ability of different fault bearing signals, and finally realizing more accurate rolling bearing composite fault diagnosis.

[0024] In step S1, the present invention collects the fault signals of the inner and outer rings of the rolling bearing. As Figure 2 shown, based on the sliding window filter, the resonance frequency band of the fault signal is located, and the sub-signal set is separated. The modal optimization index values of each sub-signal are calculated successively, the first four sub-signals with the largest modal optimization index values are selected, and the redundant components are removed by calculating the correlation coefficient between signals, and the modes containing fault characteristics are retained.

[0025] Specifically, the modal optimization index (MOI) value of the present invention consists of kurtosis (Kurt), envelope spectrum kurtosis (ESK) and mutual information (MI).

[0026] Among them, the calculation formula of the kurtosis (Kurt) of the signal is: ; In the formula, is the time-domain sequence of the signal; is the mean value of the time-domain signal; is the signal length; represents the index.

[0027] The calculation formula of the envelope spectrum kurtosis (ESK) is: ; In the formula, represents the amplitude of the envelope spectrum at the frequency ; is the mean value of the envelope spectrum.

[0028] In order to eliminate the interference components that also have a high kurtosis and envelope spectrum kurtosis but have a low correlation with the original signal and do not contain fault information, the present invention introduces mutual information, and the calculation formula is: ; In the formula, is the entropy value of signal ; is the conditional entropy of signal when is known, for signal

[0029] Finally, the modal optimization index (MOI) value is defined as follows: .

[0030] Furthermore, in order to initialize the sliding window filter, the present invention divides the original signal into segments, each segment having a length of , and determines the upper and lower cut-off frequencies and of each segment according to the sampling frequency , constructs finite impulse response (FIR) filters (for example, selects a Hann window as the filter, with a length of ), where the upper and lower cut-off frequencies and satisfy: ; In the formula, represents the index, is the total number of filters.

[0031] Furthermore, the present invention uses the correlation kurtosis as the objective function to update the filter bank, and approximates the filtered signal corresponding to the maximum correlation kurtosis by iteratively solving the generalized eigenvalue problem, where the update formula is: ; ; In the formula, is the correlation kurtosis value, is the shift number; is the signal component output by the th filter; is the sampling period; is the signal length; is the coefficient vector of the th filter; is the filter length; , , , all represent indexes; Represents a constraint condition.

[0032] The present invention proposes a method for solving the above formula by iterative eigenvalue decomposition as follows: ; Wherein, , represents the matrix form of the signal component output by the k-th filter; , represents the matrix form of the signal; , represents the matrix form of the k-th FIR filter.

[0033] Correlation kurtosis Is defined as: ; In the formula, Is the conjugate transpose operation; Is the weighted correlation matrix. For Further derivation: ; In the formula, Is the weighted correlation matrix; Is the correlation matrix.

[0034] Furthermore, the present invention transforms the Maximization problem into a generalized eigenvalue problem, calculates the eigenvector associated with the maximum eigenvalue , and the expression is as follows: .

[0035] Through continuous iteration, the coefficients of the -th filter will be continuously updated, so as to continuously approximate the filtered signal corresponding to the maximum value of the correlation kurtosis.

[0036] In step S2, the present invention estimates the fault characteristic frequency of the resonant frequency band based on envelope demodulation, uses the envelope harmonic product spectrum to estimate the frequency of the remaining sub-signals, eliminates the unconverged sub-signals, and takes the envelope harmonic product spectrum value of the remaining signals as the fault characteristic frequency.

[0037] Specifically, the present invention uses the envelope harmonic product spectrum (EHPS) to estimate the frequency of the remaining sub-signals, and its mathematical expression is: ; In the formula, Is the envelope harmonic product spectrum value, , , , , is the amplitude of the Fourier transform of the signal envelope; is the number of harmonics; represents the index; is the angular frequency.

[0038] In step S3, the present invention classifies the optimal sub-signals based on an improved sparse Bayesian learning classification model, updates and solves the maximum a posteriori probability through variational Bayesian inference, and realizes the separation of composite fault signals.

[0039] Specifically, according to the traditional sparse Bayesian model, the fault signal is expressed as obeying a Gaussian distribution with a mean of 0 and a variance of , and the process is as follows: ; where affects the sparsity of . Similarly, is modeled as an independent and identically distributed Gaussian distribution, and the process is as follows: .

[0040] Since the main elements of the signal mainly appear in the form of blocks, that is, a block-sparse structure. is not only related to , but also related to and . Therefore, the improved sparse Bayesian learning (SBL) classification model designed by the present invention adopts a block-sparse structure, and its mathematical expression is: ; The joint probability density formula is: ; In the formula, is the conditional probability of the block-sparse structure; is the block structure matrix, , ; is the fault feature component, is the hyperparameter that controls the sparsity, and the subscript represents the position within the block; is the block structure weight coefficient, which defines the degree of dependence on adjacent sparse parameters; is a Gaussian distribution with a mean of 0 and a variance of ; is the fault signal; is the observed signal; is the set of latent variables; is the likelihood function; is the noise precision of the prior distribution, is the sparse parameter of the prior distribution; is the conditional probability of the block sparse structure; is the prior distribution of the block structure matrix; is the signal length; and both represent indices.

[0041] Since in order to extract the fault pulse, the true posterior distribution must be estimated , however, the traditional SBL algorithm has a high computational complexity when updating the latent variables. The present invention reduces the computational complexity by designing the relaxed evidence lower bound and introduces variational Bayesian inference (VBI) to approximately solve the maximum a posteriori probability: .

[0042] Since the variational distribution is an approximation, the present invention introduces KL divergence to minimize the error between the variational distribution and the true posterior distribution . The specific optimization formula is: .

[0043] Expanding the above formula, we get: ; Let: ; is the evidence lower bound, and the relaxed evidence lower bound proposed by the present invention is: ; where ; In the formula, is the relaxed evidence lower bound, is the variational distribution used to approximate the intractable true posterior distribution ; is the generalized likelihood function; is the observation likelihood function; is the conditional probability; is the set of latent variables; is the variational parameter.

[0044] The improved sparse Bayesian learning (SBL) model of the present invention can reduce the computational complexity from 𝒪(N³) to 𝒪(FN) by introducing the relaxed evidence lower bound.

[0045] According to the alternating update algorithm, the optimal approximate solution can be updated as follows: ; ; ; ; In the formula, is the -th iteration; is an additive scalar.

[0046] Substitute into to get: ; Since can be separated for each , so: ; It can be seen from this that follows a Gaussian distribution .

[0047] Similarly, substitute into several other formulas to get: ; Obviously, follows a gamma distribution.

[0048] ; Therefore, can be separated for each : ;

[0049] follows a gamma distribution .

[0050] ; Since is discrete, only one element is 1, and can also be separated for each , so we only need to calculate , by exhaustive method, which is expressed as follows: ; Finally, the optimal approximate solution is updated as follows: ; ; ; ; wherein, is the probability density of the signal after iterations; is the noise precision after iteration; is the sparsity after iteration; is the nth element of the signal after iterations; is the probability density of the block structure matrix after iterations; is the gamma distribution; and are usually set to a very small constant; is a constant; and are smaller constants; .

[0051] In step S4, the present invention performs envelope demodulation on the separated signal, and determines the specific fault type according to the presence or absence of the bearing fault characteristic frequency.

[0052] Specifically, first, perform Hilbert transform on the separated signal to construct an analytic signal : ; wherein, is the signal after Hilbert transform; is the imaginary unit.

[0053] Then, extract the amplitude of the analytic signal: .

[0054] Finally, perform Fourier transform on the amplitude to obtain the envelope spectrum , and observe the characteristic frequency through the envelope spectrum.

[0055] Figure 3 is the envelope diagram of the separated fault signal. By observing the inner and outer ring fault frequencies of the bearing and their harmonics, it can be determined whether there are faults in the inner and outer rings of the rolling bearing.

[0056] Embodiment 2:

[0057] As Figure 4As shown in the figure, the present invention provides a rotary machinery compound fault diagnosis system, which is used to implement the rotary machinery compound fault diagnosis method in the first embodiment above, and specifically includes: A frequency band adaptive positioning module 100, which is used to collect the fault signals of the inner and outer rings of the rolling bearing, locate the resonant frequency band of the fault signal based on a sliding window filter, separate the sub-signal set, calculate the modal optimization index values of each sub-signal successively, select the top four sub-signals with the largest modal optimization index values, and eliminate redundant components by calculating the correlation coefficient between signals, and retain the mode containing fault characteristics; A characteristic frequency estimation module 200, which is used to estimate the fault characteristic frequency of the resonant frequency band based on envelope demodulation, estimate the frequency of the remaining sub-signals by using the envelope harmonic product spectrum, eliminate the non-convergent sub-signals, and use the envelope harmonic product spectrum value of the remaining signals as the fault characteristic frequency; A fault separation module 300, which is used to classify the optimal sub-signals based on an improved sparse Bayesian learning classification model, update and solve the maximum a posteriori probability through variational Bayesian inference, and realize the separation of compound fault signals; An envelope demodulation module 400, which is used to perform envelope demodulation on the separated signals, and judge the specific fault type according to whether there is a bearing fault characteristic frequency.

[0058] The rotary machinery compound fault diagnosis system in this embodiment is used to implement the aforementioned rotary machinery compound fault diagnosis method. Therefore, the specific implementation manners in the rotary machinery compound fault diagnosis system can be seen in the embodiment part of the previous rotary machinery compound fault diagnosis method. For example, the frequency band adaptive positioning module 100, the characteristic frequency estimation module 200, the fault separation module 300, and the envelope demodulation module 400 are respectively used to implement steps S1, S2, S3, and S4 in the above rotary machinery compound fault diagnosis method. Therefore, the specific implementation manners can refer to the descriptions of the corresponding individual embodiments. To avoid redundancy, they will not be elaborated here.

[0059] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a means for implementing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 means for implementing the functions specified in one block or multiple blocks.

[0061] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means for implementing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 means for implementing the functions specified in one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 means for implementing the functions specified in one block or multiple blocks.

[0062] Obviously, the above embodiments are only examples for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. And the obvious changes or modifications derived therefrom are still within the protection scope of the present invention.

Claims

1. A method for diagnosing compound faults of rotating machinery, characterized in that, Including: S1: Collect the fault signals of the inner and outer rings of the rolling bearing, locate the resonant frequency band of the fault signal based on the sliding window filter, separate the sub-signal set, calculate the modal optimization index values of each sub-signal successively, select the first four sub-signals with the largest modal optimization index values, and eliminate the redundant components by calculating the correlation coefficient between signals, and retain the mode containing the fault characteristics; S2: Estimate the fault characteristic frequency of the resonant frequency band based on envelope demodulation, use the envelope harmonic product spectrum to estimate the frequency of the remaining sub-signals, eliminate the unconverged sub-signals, and use the envelope harmonic product spectrum value of the remaining signals as the fault characteristic frequency; S3: Classify the optimal sub-signals based on the improved sparse Bayesian learning classification model, update and solve the maximum a posteriori probability through variational Bayesian inference, and realize the separation of the composite fault signal; S4: Perform envelope demodulation on the separated signal, and judge the specific fault type according to whether there is a bearing fault characteristic frequency.

2. The rotary machinery compound fault diagnosis method according to claim 1, characterized in that, The modal optimization index value is composed of kurtosis, envelope spectrum kurtosis and mutual information, and its calculation method is: ； Where ； ； ； Wherein, is the modal optimization index value; is kurtosis; ESK is the kurtosis of the envelope spectrum; MI is the mutual information; is the time-domain sequence of the signal; is the mean value of the time-domain signal; is the signal length; denotes the index; denotes the amplitude of the envelope spectrum at the frequency ; is the mean value of the envelope spectrum; is the signal entropy value; is the signal when known, the conditional entropy of the signal .

3. The rotary machinery compound fault diagnosis method according to claim 1, characterized in that The initialization of the sliding window filter includes: Divide the original signal into segments, each segment having a length of , and determine the upper and lower cut-off frequencies of each segment according to the sampling frequency and , and construct finite impulse response filters, where the upper and lower cut-off frequencies and satisfy: ； In the formula, represents the index, is the total number of filters.

4. The rotating machinery compound fault diagnosis method according to claim 1 or 3, characterized in that, The update method of the sliding window filter includes: Use the correlation kurtosis as the objective function to update the filter bank, and approximate the filtered signal corresponding to the maximum value of the correlation kurtosis by iteratively solving the generalized eigenvalue problem, where the update formula is: ； ； Wherein, is the relevant kurtosis value, is the shift number; is the th signal component output by the filter; is the sampling period; is the signal length; is the th coefficient vector of the filter; is the filter length; , , , all represent indices; represents the constraint condition.

5. The rotational machinery compound fault diagnosis method according to claim 1, characterized in that The method of using the envelope harmonic product spectrum to estimate the frequency of the remaining sub-signals is: ； Wherein, is the envelope harmonic product spectrum value, , , , , are the Fourier transform amplitudes of the signal envelope; is the number of harmonics; represents the index; is the angular frequency.

6. The rotational machinery compound fault diagnosis method according to claim 1, wherein The improved sparse Bayesian learning classification model adopts a block sparse structure, and its mathematical expression is: ； Joint probability density The formula is as follows: ； Wherein, is the conditional probability of the block sparse structure; is the block structure matrix; is the fault feature component, is for controlling the hyperparameter of sparsity, and the subscript represents the position within the block; is the block structure weight coefficient, which defines the dependence degree on adjacent sparse parameters; is a Gaussian distribution with a mean of 0 and a variance of ; is the fault signal; is the observed signal; is the set of latent variables; is the likelihood function; is the prior distribution of the noise precision ; is the prior distribution of the sparse parameter ; is the conditional probability of the block sparse structure; is the prior distribution of the block structure matrix; is the signal length; , both represent indices.

7. The rotating machinery compound fault diagnosis method according to claim 6, wherein Classifying the optimal sub-signals based on the improved sparse Bayesian learning classification model, and updating and solving the maximum a posteriori probability through variational Bayesian inference specifically includes: Design a relaxed evidence lower bound formula, and iteratively update the latent variable through variational Bayesian inference to solve the approximate solution of the maximum a posteriori probability, where the relaxed evidence lower bound calculation formula is: ； Where ； Wherein, is the lower bound of the relaxed evidence; is the variational distribution used to approximate the true posterior distribution that is difficult to calculate ; is the generalized likelihood function; is the observation likelihood function; is the conditional probability; is the set of latent variables; are the variational parameters.

8. The rotary machinery compound fault diagnosis method according to claim 7, wherein The lower bound of the relaxed evidence is obtained by introducing KL divergence to minimize the variational distribution and the error of the true posterior distribution is as follows. The specific optimization formula is ； And reduce the computational complexity by decomposing the latent variable update rule.

9. The rotary machinery compound fault diagnosis method according to claim 1, wherein The method of performing envelope demodulation on the separated signal is: First, perform Hilbert transform on the separated signal to construct an analytic signal; then, extract the amplitude of the analytic signal; finally, perform Fourier transform on the amplitude to obtain the envelope spectrum, and observe the characteristic frequency through the envelope spectrum.

10. A compound fault diagnosis system for rotating machinery, characterized in that, The system is used to implement the rotating machinery composite fault diagnosis method described in any one of claims 1 to 9, specifically including: A frequency band adaptive positioning module, which is used to collect the fault signals of the inner and outer rings of the rolling bearing, locate the resonant frequency band of the fault signal based on the sliding window filter, separate the sub-signal set, calculate the modal optimization index values of each sub-signal successively, select the first four sub-signals with the largest modal optimization index values, and eliminate the redundant components by calculating the correlation coefficient between signals, and retain the mode containing the fault characteristics; A characteristic frequency estimation module, which is used to estimate the fault characteristic frequency of the resonant frequency band based on envelope demodulation, use the envelope harmonic product spectrum to estimate the frequency of the remaining sub-signals, eliminate the unconverged sub-signals, and use the envelope harmonic product spectrum value of the remaining signals as the fault characteristic frequency; A fault separation module, which is used to classify the optimal sub-signals based on an improved sparse Bayesian learning classification model, update and solve the maximum a posteriori probability through variational Bayesian inference, and achieve the separation of composite fault signals; An envelope demodulation module, which is used to perform envelope demodulation on the separated signals and judge the specific fault type according to the presence or absence of the bearing fault characteristic frequency.

Images