Acoustic emission fault diagnosis method for mechanical rotating part based on bidirectional weighted cyclostationarity

Abstract

The invention provides a mechanical rotating member acoustic emission fault diagnosis method based on bidirectional weighted cyclostationary, and the method comprises the steps: collecting a to-be-diagnosed acoustic emission signal, obtaining a cyclic spectrum coherence function, and obtaining baseline discrete spectrum coherence mapping and to-be-diagnosed discrete spectrum coherence mapping; obtaining a spectrum frequency weight redistribution vector based on the baseline discrete spectrum coherent mapping and the discrete spectrum coherent mapping to be diagnosed; obtaining a differential spectrum coherence matrix based on the cyclic spectrum coherence function; and obtaining a bidirectional reweighted differential spectrum coherence matrix based on the spectrum frequency weight redistribution vector and the differential spectrum coherence matrix. According to the method, health baseline data is used as prior information to be fused into a cyclostationary analysis framework, difference and weighting thoughts are creatively fused, unique advantages and potential of an acoustic emission technology in the field of fault diagnosis are deeply excavated, complex and changeable industrial application scene requirements are fully met, and the method has practical application value.

Description

Technical Field

[0001] The invention belongs to the technical field of rotating machinery fault diagnosis, in particular to a mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationarity. Background Art

[0002] As an effective tool for monitoring the damage process of rotating machinery components, acoustic emission technology has demonstrated high sensitivity in detecting early weak faults, providing an indispensable supplement to the current condition monitoring technology system.

[0003] Although the combination of acoustic emission technology and health indicators can achieve real-time tracking of damage status, its analysis focus tends to process time domain information, and is still insufficient in the in-depth mining and utilization of frequency domain information. When a rotating component fails, its acoustic emission response usually shows significant impact and cyclostationarity characteristics. Therefore, extracting signal components that match these characteristics has become the core goal of the research. In the existing acoustic emission signal spectrum analysis and diagnosis technology, most methods draw on the algorithm framework based on vibration acceleration, among which the resonance demodulation technology is particularly prominent, which covers a series of blind filtering methods such as 1 / 3 binary tree spectrum analysis, deconvolution processing and signal decomposition. The core of these methods is to accurately locate and extract the target narrowband signal containing the maximum amount of fault information through optimization strategies, and then perform envelope analysis to further highlight the fault characteristics.

[0004] In practical applications, when the signal-to-noise ratio of the target frequency band is low or mixed with other non-fault related information, the accuracy and reliability of the diagnostic results are often seriously affected. In addition, the selection of filter parameters needs to be flexibly adjusted according to the specific application scenario, and this process is highly dependent on the knowledge accumulation and practical experience of experts in the field. Another important method is to use cyclostationary analysis tools, which can effectively detect the statistical periodic characteristics of the signal in the full carrier frequency domain, thus avoiding the dilemma of frequency band selection. In the field of mechanical condition monitoring, studies have successfully introduced cyclostationary technology into the detection of unknown targets, and its application examples include the monitoring of helicopters and underwater propellers. The cyclostationary algorithm system, including spectral correlation analysis and spectral coherence analysis, provides a new perspective for signal processing by converting time domain signals into a dual spectrum mapping form of cyclic frequency and carrier frequency.

[0005] However, the current cyclostationary analysis method integrating acoustic emission technology still faces several challenges: the first problem is that it is difficult to effectively extract fault feature information distributed in multiple narrow frequency bands at the same time within the wider range of the acoustic emission information frequency band; secondly, the prior information on which feature extraction relies is limited by the fault characteristics derived from the dynamic model, and the noise suppression problem within the preferred frequency band has not been properly solved; finally, acoustic emission monitoring technology based on cyclostationary analysis has mainly focused on single components of rolling bearings in the past, and there is a lack of reports on monitoring applications of other complex mechanical rotating parts.

[0006] Therefore, considering the wide frequency response range and rich information content of acoustic emission signals, it is urgent and necessary to seek an acoustic emission fault diagnosis method for mechanical rotating parts based on bidirectional weighted cyclostationarity to achieve effective extraction and identification of early weak fault features. Summary of the invention

[0007] In view of the defects in the above-mentioned prior art, the present invention proposes a method for diagnosing acoustic emission faults of mechanical rotating parts based on bidirectional weighted cyclostationarity. The method includes collecting the acoustic emission signal to be diagnosed, obtaining the cyclic spectrum coherence function and obtaining the baseline discrete spectrum coherence mapping and the discrete spectrum coherence mapping to be diagnosed; obtaining the spectrum frequency weight redistribution vector based on the baseline discrete spectrum coherence mapping and the discrete spectrum coherence mapping to be diagnosed; obtaining the differential spectrum coherence matrix based on the cyclic spectrum coherence function; and obtaining the bidirectional reweighted differentiated spectrum coherence matrix based on the spectrum frequency weight redistribution vector and the differential spectrum coherence matrix. The present invention integrates the healthy baseline data as prior information into the cyclostationary analysis framework, innovatively integrates the ideas of difference and weighting, deeply explores the unique advantages and potential of acoustic emission technology in the field of fault diagnosis, fully meets the needs of complex and changeable industrial application scenarios, and has practical application value.

[0008] The present invention provides a mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationarity, which comprises the following steps:

[0009] S1, collecting the acoustic emission signal to be diagnosed, obtaining the cyclic spectrum coherence function and obtaining the baseline discrete spectrum coherence mapping and the discrete spectrum coherence mapping to be diagnosed;

[0010] S2. Based on the baseline discrete spectrum coherence map and the to-be-diagnosed discrete spectrum coherence map, a spectrum frequency weight redistribution vector is obtained; based on the maximum mean difference MMD metric, the to-be-diagnosed-baseline difference of each spectrum frequency component in the cyclic spectrum coherence function is evaluated, and the spectrum frequency weight redistribution vector R is calculated:

[0011]

[0012] in, represents the baseline discrete spectral coherence map, represents the discrete spectrum coherence map to be diagnosed, f n Indicates the resolution is F s / N w The nth spectral component in the spectral frequency of s Indicates the sampling frequency; N w The window length set in the cyclic spectrum coherence spectrum estimation;

[0013] S3, obtaining a differential spectral coherence matrix based on the cyclic spectral coherence function;

[0014] S31. Calculate the differential demodulation spectrum: the narrowband demodulation spectrum γ based on the baseline signal H (α) and the narrowband demodulation spectrum γ of the signal to be diagnosed F (α), calculate the differential demodulation spectrum γ diff (α):

[0015]

[0016] Among them, ||·|| L1 represents the L1 norm of the matrix; c represents the separation factor;

[0017] S32, judging the spectral ratio, if the spectral ratio is greater than 1, executing c=0.9c and returning to re-execute step S31; if the spectral ratio is less than 1, executing c=1.1c and returning to re-execute step S31; if the spectral ratio is equal to 1, executing step S33;

[0018] S33, based on the differential demodulation spectrum γ diff (α), separate the cyclic spectrum coherence function γ of the acoustic emission signal x(t) to be diagnosed x The fault inducing factors of each spectral frequency component in (α, f) are used to obtain the differential spectral coherence matrix:

[0019]

[0020] Among them, Max(·,·) means taking the maximum value of the corresponding elements of two vectors;

[0021] S4, based on the spectral frequency weight redistribution vector and the differential spectral coherence matrix, obtain a bidirectionally reweighted differentiated spectral coherence matrix: based on the spectral frequency weight redistribution vector R as the differential spectral coherence matrix CSCoh diff The reintegration strategy is used to obtain the bidirectional reweighted differential spectral coherence matrix DRD:

[0022] DRD=(R×(CSCoh diff ) T ) T (9)

[0023] in,(·) TRepresents the transpose of a matrix.

[0024] Further, the step S1 specifically includes the following steps:

[0025] S11. In the continuous monitoring scenario, use the acoustic emission monitoring system to obtain the historical health operation baseline data of the machine x H (t), collect the acoustic emission signal x(t) to be diagnosed;

[0026] S12, calculate the time-varying autocorrelation function R of the acoustic emission signal x(t) to be diagnosed xx (t,τ):

[0027]

[0028] Where, t represents time; τ represents delay; represents the ensemble average operator; x * Indicates the conjugate operation on x;

[0029] S13, calculate the cyclic spectrum correlation function S of the acoustic emission signal x(t) to be diagnosed x (α, f), for the time-varying autocorrelation function R described in formula (1) xx The bispectral representation of (t,τ) after two-dimensional Fourier transform is as follows:

[0030]

[0031] Among them, α represents the cyclic frequency; f represents the spectral frequency; j represents the imaginary number; e represents the exponential function;

[0032] S14, calculating the cyclic spectrum coherence function γ of the acoustic emission signal x(t) to be diagnosed x (α, f), for the cyclic spectrum correlation function S described in formula (2) x (α,f) is normalized in energy, specifically expressed as:

[0033]

[0034] S15, based on the cyclic spectrum coherence function γ x (α, f), combined with the spectrum estimation method, obtain the baseline discrete spectrum coherence mapping of the acoustic emission signal x(t) to be diagnosed and the discrete spectrum coherence mapping to be diagnosed

[0035] Preferably, in step S2, a first probability distribution p and a second probability distribution q are set, assuming that the first group of signal data X obeys X~p, and the second group of signal data Y obeys Y~q, and the maximum mean difference MMD metric is It is expressed as:

[0036]

[0037] in, represents the set of functions on the sample space; f(·) represents the mapping to the regenerated Hilbert space; n p and n q Respectively represent the number of samples of the first group of signal data X and the second group of signal data Y; i and j Respectively represent sample data of the first group of signal data X and the second group of signal data Y.

[0038] Preferably, the spectral ratio in step S32 is defined as:

[0039]

[0040] Among them, min(γ diff (α)) and max(γ diff (α)) respectively represent the differential demodulation spectrum γ diff The minimum and maximum values ​​of (α).

[0041] Preferably, in step S13, the cycle frequency α and the spectrum frequency f correspond to the time t and the delay τ before mapping respectively, and the cycle frequency α and the spectrum frequency f respectively indicate the modulation frequency and the carrier frequency of the acoustic emission signal x(t) to be diagnosed.

[0042] Preferably, in the differential decomposition spectrum in step S31, negative values ​​represent basic components, and positive values ​​represent fault-inducing factors.

[0043] Preferably, the acoustic emission signal to be diagnosed in step S11 is subjected to a preprocessing operation, and the preprocessing operation includes high-frequency sampling, data truncation and de-averaging.

[0044] Compared with the prior art, the technical effects of the present invention are:

[0045] 1. The acoustic emission fault diagnosis method for mechanical rotating parts based on bidirectional weighted cyclostationarity designed in the present invention, on the basis of analyzing the characteristics of acoustic emission monitoring signals, integrates the healthy baseline data as prior information into the cyclostationary analysis framework, effectively extracts the fault characteristic components, and improves the accuracy of fault diagnosis; innovatively integrates the ideas of difference and weighting, and performs double optimization on the cyclic spectrum coherence feature plane. On the one hand, the maximum mean difference is used to measure the distribution difference between the healthy reference and the signal to be diagnosed, so as to achieve accurate reweighting in the spectral frequency direction; on the other hand, through the differential processing on the cyclic frequency, the diagnostic capability of the one-dimensional mapping after integration along the spectral frequency is significantly enhanced, and the sensitivity of fault identification is further improved.

[0046] 2. The mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationarity designed in the present invention can be verified by using a variety of mechanical rotating parts acoustic emission monitoring signals to comprehensively evaluate the performance of the proposed method, such as planetary gearbox failure cases and rolling bearing failure cases, which fully demonstrate the state monitoring and fault diagnosis capabilities of the proposed method in various types of rotating parts acoustic emission monitoring signals; the proposed method deeply explores the unique advantages and potential of acoustic emission technology in the field of fault diagnosis, realizes high-precision monitoring and fault diagnosis of a variety of mechanical rotating parts, and fully meets the needs of complex and changeable industrial application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Other features, objects and advantages of the present application will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings.

[0048] Figure 1 It is a flow chart of the method for diagnosing mechanical rotating parts acoustic emission faults based on bidirectional weighted cyclostationarity of the present invention;

[0049] Figure 2 is a schematic diagram of the structure of a test bench in a specific embodiment of the present invention;

[0050] Figure 3 The acoustic emission waveform flow when the planetary gearbox is running healthily in the first specific embodiment of the present invention;

[0051] Figure 4 The acoustic emission waveform flow when the planetary gear of the planetary gearbox has a weak fault in the first specific embodiment of the present invention;

[0052] Figure 5 is a spectral kurtosis ratio graph feature matrix of a fault signal in a first specific embodiment of the present invention;

[0053] Figure 6 In the first specific embodiment of the present invention, Figure 5 The optimal frequency band of the filter signal is obtained by

[0054] Figure 7 is a cyclic spectrum coherence matrix of a fault signal in a first specific embodiment of the present invention;

[0055] Figure 8 In the first specific embodiment of the present invention, Figure 7 The enhanced envelope spectrum obtained by performing full-band integration strategy;

[0056] Fig. 9 A histogram of cyclic spectrum coherence weight redistribution vectors in the first specific embodiment of the present invention;

[0057] Fig.10In the first specific embodiment of the present invention, Fig. 9 The envelope spectrum obtained after the bidirectional reweighted difference operation;

[0058] Fig.11 The acoustic emission waveform flow of the harmonic reducer in healthy operation in the second specific embodiment of the present invention;

[0059] Fig.12 The acoustic emission waveform flow before the final failure of the harmonic reducer in the second specific embodiment of the present invention;

[0060] Fig.13 is a spectral kurtosis ratio graph feature matrix of a fault signal in a second specific embodiment of the present invention;

[0061] Fig.14 In the second specific embodiment of the present invention, Fig.13 The optimal frequency band of the filter signal is obtained by

[0062] Fig.15 is a cyclic spectrum coherence matrix of a fault signal in a second specific embodiment of the present invention;

[0063] Fig.16 In the second specific embodiment of the present invention, Fig.15 The enhanced envelope spectrum obtained by performing full-band integration strategy;

[0064] Fig.17 A histogram of cyclic spectrum coherence weight redistribution vectors in a second specific embodiment of the present invention;

[0065] Fig.18 In the second specific embodiment of the present invention, Fig.17 The envelope spectrum obtained after the bidirectional reweighted difference operation. DETAILED DESCRIPTION

[0066] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It will be appreciated that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the inventions. It should also be noted that, for ease of description, only portions related to the relevant inventions are shown in the accompanying drawings. It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0067] Figure 1 The present invention shows a method for diagnosing mechanical rotating parts acoustic emission faults based on bidirectional weighted cyclostationarity, which fully exploits the cyclostationarity difference between signals in a healthy state and a faulty state. The method comprises the following steps:

[0068] S1. Acquire the acoustic emission signal to be diagnosed, obtain the cyclic spectrum coherence function and obtain the baseline discrete spectrum coherence map and the discrete spectrum coherence map to be diagnosed.

[0069] S11. In the continuous monitoring scenario, use the acoustic emission monitoring system to obtain the historical health operation baseline data of the machine x H (t), collect the acoustic emission signal x(t) to be diagnosed. To ensure the accuracy and effectiveness of signal acquisition, the acoustic emission sensor should be placed in a position close to the mechanical rotating part of the tested equipment and with a flat surface, and fixed by gluing or magnetic clamp adsorption. If magnetic clamp adsorption is used, it is necessary to apply high vacuum silicone grease or other acoustic coupling agents to optimize signal transmission. Before the formal test, a lead break test is also required to verify the reliability and stability of the sensor connection. The acoustic emission signal to be diagnosed is preprocessed and the preprocessing operation includes high-frequency sampling, data truncation and de-averaging to ensure the accuracy and efficiency of subsequent analysis.

[0070] S12, calculate the time-varying autocorrelation function R of the acoustic emission signal x(t) to be diagnosed xx (t,τ):

[0071]

[0072] Where, t represents time; τ represents delay; represents the ensemble average operator; x * It represents the conjugate operation on x.

[0073] S13, calculate the cyclic spectrum correlation function S of the acoustic emission signal x(t) to be diagnosed x (α,f), for the time-varying autocorrelation function R in formula (1) xx The bispectral representation of (t,τ) after two-dimensional Fourier transform is as follows:

[0074] S x (α,f)=∫∫R x (t,τ)e -j2π(αt+fτ) dtdτ (2)

[0075] Among them, α represents the cyclic frequency; f represents the spectral frequency; j represents the imaginary number; and e represents the exponential function.

[0076] The cycle frequency α and the spectrum frequency f correspond to the time t and the time delay τ before mapping, respectively, and also indicate the modulation frequency and the carrier frequency of the acoustic emission signal x(t) to be diagnosed, respectively.

[0077] S14, calculating the cyclic spectrum coherence function γ of the acoustic emission signal x(t) to be diagnosed x (α, f), for the cyclic spectrum correlation function S in formula (2) x(α,f) is normalized in energy, specifically expressed as:

[0078]

[0079] S15, based on the cyclic spectrum coherence function γ x (α, f), combined with the spectrum estimation method, obtain the baseline discrete spectrum coherence mapping of the acoustic emission signal x(t) to be diagnosed and the discrete spectrum coherence mapping to be diagnosed

[0080] S2. Obtain a weight redistribution vector based on the baseline discrete spectrum coherence mapping and the discrete spectrum coherence mapping to be diagnosed.

[0081] In the spectral frequency dimension of the spectral coherence matrix, the present invention dynamically determines the spectral frequency weight by evaluating the deviation of each sub-band monitoring signal relative to the baseline signal, aiming to enhance the diagnostic accuracy and efficiency of the one-dimensional mapping obtained after integration along the spectral frequency.

[0082] As a metric, Maximum Mean Discrepancy (MMD) can effectively quantify the distance between two statistical distributions in the reproducing kernel Hilbert space, and is particularly suitable for evaluating the degree of difference between the sensor signal to be diagnosed and the historical data of health status. Compared with other difference quantification techniques, MMD shows more excellent difference representation ability and numerical stability. At the same time, by properly selecting the kernel function, the mapping strategy of the feature space can be flexibly adjusted and optimized.

[0083] Set the first probability distribution p and the second probability distribution q, assume that the first group of signal data X obeys X~p, and the second group of signal data Y obeys Y~q, and the maximum mean difference MMD metric It is expressed as:

[0084]

[0085] in, represents the set of functions on the sample space; f(·) represents the mapping to the regenerated Hilbert space; n p and n q Respectively represent the number of samples of the first group of signal data X and the second group of signal data Y; i and j Represents the sample data of the first group of signal data X and the second group of signal data Y respectively. Maximum mean difference MMD metric The larger the value, the greater the distance between the first group of signal data X and the second group of signal data Y.

[0086] Based on the maximum mean difference (MMD) metric, the difference between the diagnosis and baseline of each spectral frequency component in the cyclic spectral coherence function is evaluated, and the spectral frequency weight redistribution vector R is calculated:

[0087]

[0088] in, represents the baseline discrete spectral coherence map, represents the discrete spectrum coherence map to be diagnosed, f n Indicates the resolution is F s / N w The nth spectral component in the spectral frequency of s Indicates the sampling frequency; N w The window length set for the cyclic coherence spectrum estimation.

[0089] S3. Based on the cyclic spectral coherence function, a differential spectral coherence matrix is ​​obtained.

[0090] In the direction of the cyclic axis of the spectral coherence matrix, the residual signal based on the healthy reference is used to characterize the cyclic component introduced by the fault.

[0091] S31. Calculate the differential demodulation spectrum: the narrowband demodulation spectrum γ based on the baseline signal H (α) and the narrowband demodulation spectrum γ of the signal to be diagnosed F (α), calculate the differential demodulation spectrum γ diff (α):

[0092]

[0093] Among them, ||·|| L1 represents the L1 norm of the matrix; c represents the separation factor, and adjusting its value can achieve the positive and negative separation of fault-inducing factors and basic components.

[0094] S32. Determine the spectral ratio. If the spectral ratio is greater than 1, execute c=0.9c and return to re-execute step S31; if the spectral ratio is less than 1, execute c=1.1c and return to re-execute step S31; if the spectral ratio is equal to 1, execute step S33.

[0095] The spectral ratio is defined as:

[0096]

[0097] Among them, min(γ diff (α)) and max(γ diff (α)) respectively represent the differential demodulation spectrum γ diff The minimum and maximum values ​​of (α).

[0098] The differential demodulation spectrum calculated by the above algorithm can ensure that the absolute amplitudes of the basic components and the fault inducing factors are approximately equal. Negative values ​​in the spectrum represent the basic components, and positive values ​​represent the fault inducing factors.

[0099] S33, based on differential decomposition spectrum γ diff (α), separate the cyclic spectrum coherence function γ of the acoustic emission signal x(t) to be diagnosed x The fault inducing factors of each spectral frequency component in (α, f) are used to obtain the differential spectral coherence matrix:

[0100]

[0101] Among them, Max(·,·) means taking the maximum value of the corresponding elements of the two vectors, the purpose of which is to remove the basic components in the differential decomposition spectrum.

[0102] S4. Based on the spectral frequency weight redistribution vector and the differential spectral coherence matrix, a bidirectionally reweighted differentiated spectral coherence matrix is ​​obtained.

[0103] The spectrum frequency weight redistribution vector R obtained in step S2 represents a frequency band rich in fault introduction information.

[0104] Based on the spectral frequency weights, the redistribution vector R is used as the differential spectral coherence matrix CSCoh diff The reintegration strategy is used to obtain the bidirectional reweighted differential spectral coherence matrix DRD:

[0105] DRD=(R×(CSCoh diff ) T ) T (9)

[0106] in,(·) T Represents the transpose of a matrix.

[0107] Acoustic emission technology has the characteristics of wide-band response, and its sampling frequency is generally set at the MHz level. On the one hand, the spectrum frequency weight redistribution vector can ensure the effective synchronous extraction of fault information components located in multiple frequency bands. On the other hand, the differential demodulation spectrum can retain the fault-inducing components in the acoustic emission signal and suppress the basic components that have always existed in the acoustic emission signal during the monitoring process, where positive values ​​represent fault-inducing components and zero values ​​represent basic components, i.e., inherent components. Fault-inducing components refer to signal components introduced in the signal due to the degradation of mechanical rotating parts, and inherent components refer to signal components generated by mechanical rotating parts in a healthy state.

[0108] S5. Obtain the fault characteristic components of mechanical rotating parts based on the bidirectional reweighted differential spectral coherence matrix and analyze the fault of mechanical rotating parts. Specifically, for the bidirectional reweighted differential spectral coherence matrix, integrate along the spectral frequency direction to form the corresponding envelope spectrum, search for the fault characteristic components in the enhanced envelope spectrum, determine the fault of the mechanical rotating parts, and accurately identify various types of damage, such as gear wear, gear cracks, bearing inner and outer ring faults, etc.

[0109] In a specific embodiment, the method proposed in the present invention is applied to a planetary gearbox fault diagnosis embodiment for verification.

[0110] Take a mechanical transmission system test bench as an example. Figure 2 As shown, the test bench includes a drive motor 1, several couplings 2, a speed torque measurement module 3, a parallel shaft gearbox 4, an input shaft support module 5, an input shaft speed measurement module 6, a planetary gearbox 7, an acoustic emission sensor 8, an output shaft speed measurement module 9, an output shaft support module 10 and a magnetic powder brake 11. Among them, the drive motor 1 has a rated voltage of 380V, a rated power of 3kW, a rated speed of 3000rpm and a rated torque of 9.55N·m. The coupling 2, the input shaft support module 5 and the output shaft support module 10 play a key connection and support role in the test bench. The speed torque measurement module 3 is configured to monitor the actual speed and torque value of the drive motor 1. The parallel shaft gearbox 4 is the first stage reduction device of the transmission system, and the planetary gearbox 7 is the second stage reduction device of the transmission system. The input shaft speed measurement module 6 and the output shaft speed measurement module 9 are respectively used to accurately measure the input and output actual speeds of the planetary gearbox 7. In addition, the acoustic emission sensor 8 is used to collect acoustic emission signals during the test process to achieve fault diagnosis. The magnetic powder brake 11 provides the necessary output load for the entire transmission system.

[0111] In a specific embodiment, the reduction ratio of the parallel shaft gearbox 4 is 1:3.9, the reduction ratio of the planetary gearbox 7 is 1:5, the number of sun gear teeth is 21, the number of planet gear teeth is 31, and the number of outer ring teeth is 84. The power flow direction is the sun gear shaft as input and the planet carrier shaft as output. A weak fault is implanted on the planet gear tooth surface to simulate its degradation process. The acoustic emission sampling frequency is set to 2MHz, and the preamplifier is selected to 40dB. According to the kinematic model of the mechanical transmission system, the characteristic order of the planet gear fault is calculated to be 0.139. In the experiment, the motor input speed was set to 3000rpm, and the load level was maintained at about 20%, corresponding to the planetary gearbox input speed of 12.82Hz, and the characteristic frequency of the planet gear fault was 6.95Hz. The collected acoustic emission waveform flow of the planetary gearbox during healthy operation is as follows: Figure 3 As shown in the figure, the acoustic emission waveform flow when the planetary gear has a weak fault is as follows: Figure 4 shown.

[0112] In order to show the superiority of the proposed method, the signal is firstly subjected to the traditional 1 / 3-binary tree to determine the frequency division strategy of different layers to form a feature matrix, and the ratio of the feature matrix of the fault signal and the healthy signal is calculated to form a spectral kurtosis ratio diagram, highlighting the kurtosis difference between the waiting signal and the baseline signal. The feature matrix of the spectral kurtosis ratio diagram is as follows: Figure 5 As shown, the center frequency of the preferred frequency band is 812500 Hz and the bandwidth is 41667 Hz. However, the spectral kurtosis ratio diagram has limited performance in extracting fault features in this embodiment. Figure 6 The fault characteristics are almost unrecognizable in the image, and the characteristic frequency f corresponding to the planetary gear fault is not detected. p The frequency peaks and their multiples have extremely low signal-to-noise ratios, making it difficult to directly diagnose the fault type of the equipment. Next, the cyclic spectrum coherence matrix of the fault signal is calculated, as Figure 7 As shown, the enhanced envelope spectrum is obtained by integrating the full frequency band, as shown in Figure 8 Although the enhanced envelope spectrum contains certain fault information, the characteristic frequency f corresponding to the planetary gear fault p The frequency peaks of the fault and its multiples are relatively inconspicuous. In the absence of a priori fault characteristic frequencies, it is still challenging to directly identify this weak fault. Finally, in order to more effectively identify weak faults, the bidirectional reweighted cyclostationary analysis method proposed in the present invention is used. Fig. 9 The weight redistribution vector histogram for the bidirectional reweighted difference operation is shown. Based on this, the envelope spectrum guided by this method is obtained through envelope analysis. Fig.10 Although the envelope spectrum appears to be relatively complex due to the weak fault, the fault characteristic frequency and its harmonic components are successfully included in the five spectral lines with the highest peak values, which verifies the effectiveness of the proposed method in weak fault extraction.

[0113] In another specific embodiment, the harmonic reducer fails after long-term use, which includes three failure modes: gear wear, flexspline inner wall wear, and fatigue cracking of the flexible thin-walled bearing in the wave generator. The acoustic emission signals collected during the degradation of the components are analyzed, and the collected acoustic emission waveforms of the harmonic reducer when it is running healthily are as follows: Fig.11 As shown, the acoustic emission waveform flow before the final failure is as follows Fig.12 First, the spectral kurtosis ratio graph is used to guide the signal processing process, and its characteristic matrix is ​​as follows Fig.13 As shown, the center frequency of the preferred frequency band is 104167 Hz and the bandwidth is 41667 Hz. However, the spectral kurtosis ratio diagram has limited performance in extracting fault features in this embodiment, although the envelope spectrum guided by it Fig.14 The 2 times rotation frequency f caused by gear tooth wear and flexspline inner wall wear rand its frequency multiples can be clearly displayed, but the fault characteristic frequency f corresponding to the outer ring of the flexible thin-walled bearing in the wave generator or The spectrum line at is relatively weak, and it is difficult to conclude that a fault has occurred. Next, this method calculates the cyclic spectrum coherence matrix of the fault signal, such as Fig.15 As shown, the enhanced envelope spectrum is obtained by integrating the full frequency band, as shown in Fig.16 As shown in the figure, its diagnostic effect is similar to that based on the spectral kurtosis ratio diagram, but the extraction of fault features corresponding to the outer ring of the flexible thin-walled bearing in the signal is still not ideal. Fig.17 The weight redistribution vector histogram for the bidirectional reweighted difference operation is shown. Based on this, the envelope spectrum guided by this method is obtained through envelope analysis. Fig.18 , not only can it identify the 2 times rotation frequency f caused by two uniform wear phenomena r , and can identify the fault characteristic frequency f of the outer ring of the flexible thin-walled bearing or , all three types of faults can be accurately identified, ensuring the comprehensive capture and accurate identification of fault information.

[0114] The present invention designs a mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationarity. On the basis of analyzing the characteristics of acoustic emission monitoring signals, the healthy baseline data is integrated into the cyclostationarity analysis framework as prior information, the fault characteristic components are effectively extracted, and the accuracy of fault diagnosis is improved. The differential and weighted ideas are innovatively integrated, and dual optimization is performed on the cyclic spectrum coherence feature plane. On the one hand, the maximum mean difference is used to measure the distribution difference between the healthy reference and the signal to be diagnosed, so as to achieve accurate reweighting in the spectral frequency direction. On the other hand, the differential processing on the cyclic frequency significantly enhances the diagnostic ability of the one-dimensional mapping after integration along the spectral frequency, and further improves the sensitivity of fault identification. In order to comprehensively evaluate the performance of the proposed method, a variety of mechanical rotating parts acoustic emission monitoring signals can be used for verification, such as planetary gearbox fault cases and rolling bearing fault cases, which fully proves the state monitoring and fault diagnosis capabilities of the proposed method in various rotating parts acoustic emission monitoring signals. The proposed method deeply explores the unique advantages and potentials of acoustic emission technology in the field of fault diagnosis, realizes high-precision monitoring and fault diagnosis of the states of various mechanical rotating parts, and fully meets the needs of complex and changeable industrial application scenarios.

[0115] Finally, it should be noted that the above embodiments are only intended to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the present invention can still be modified or replaced by equivalents. Any modification or partial replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for diagnosing mechanical rotating parts acoustic emission faults based on bidirectional weighted cyclostationarity, characterized in that: It includes the following steps: S1. Collect the acoustic emission signal to be diagnosed of the mechanical rotating part, obtain the cyclic spectrum coherence function and obtain the baseline discrete spectrum coherence mapping and the discrete spectrum coherence mapping of the acoustic emission signal to be diagnosed; S2. Based on the baseline discrete spectrum coherence map and the to-be-diagnosed discrete spectrum coherence map, a spectrum frequency weight redistribution vector is obtained; based on the maximum mean difference MMD metric, the to-be-diagnosed-baseline difference of each spectrum frequency component in the cyclic spectrum coherence function is evaluated, and the spectrum frequency weight redistribution vector R of the to-be-diagnosed acoustic emission signal is calculated and obtained: in, represents the baseline discrete spectrum coherence map, represents the discrete spectrum coherence map to be diagnosed, f n Indicates the resolution is F s / N w The nth spectral component in the spectral frequency of s Indicates the sampling frequency; N w The window length set in the cyclic spectrum coherence spectrum estimation; S3, based on the cyclic spectral coherence function, obtaining the differential spectral coherence matrix of the acoustic emission signal to be diagnosed; S31. Calculate the differential demodulation spectrum: the narrowband demodulation spectrum γ based on the baseline signal H (α) and the narrowband demodulation spectrum γ of the signal to be diagnosed F (α), calculate the differential demodulation spectrum γ diff (α): Among them, ||·|| L1 represents the L1 norm of the matrix; c represents the separation factor; S32, judging the spectral ratio, if the spectral ratio is greater than 1, executing c=0.9c and returning to re-execute step S31; if the spectral ratio is less than 1, executing c=1.1c and returning to re-execute step S31; if the spectral ratio is equal to 1, executing step S33; S33, based on the differential demodulation spectrum γ diff (α), separate the cyclic spectrum coherence function γ of the acoustic emission signal x(t) to be diagnosed x The fault-inducing component of each spectral frequency component in (α, f) is used to obtain the differential spectral coherence matrix of the acoustic emission signal to be diagnosed: Among them, Max(·,·) means taking the maximum value of the corresponding elements of the two vectors; the positive value of the differential decomposition spectrum represents the fault-induced component of the mechanical rotating parts, and the zero value represents the inherent component of the mechanical rotating parts; S4, based on the spectral frequency weight redistribution vector and the differential spectral coherence matrix, obtain a bidirectionally reweighted differentiated spectral coherence matrix: based on the spectral frequency weight redistribution vector R as the differential spectral coherence matrix CSCoh diff The reintegration strategy is used to obtain the bidirectional reweighted differential spectral coherence matrix DRD: DRD=(R×(CSCoh diff ) T ) T (9) in,(·) T Represents the transpose of a matrix; S5. Obtain fault characteristic components of mechanical rotating parts based on a bidirectionally reweighted differential spectral coherence matrix, and analyze the faults of the mechanical rotating parts.

2. The mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationary according to claim 1 is characterized in that: Step S1 specifically includes the following steps: S11. In the continuous monitoring scenario, use the acoustic emission monitoring system to obtain the historical health operation baseline data of the machine x H (t), collect the acoustic emission signal x(t) to be diagnosed; S12, calculate the time-varying autocorrelation function R of the acoustic emission signal x(t) to be diagnosed xx (t,τ): Where, t represents time; τ represents delay; represents the ensemble average operator; x * Indicates the conjugate operation on x; S13, calculate the cyclic spectrum correlation function S of the acoustic emission signal x(t) to be diagnosed x (α, f), for the time-varying autocorrelation function R described in formula (1) xx The bispectral representation of (t,τ) after two-dimensional Fourier transform is as follows: Among them, α represents the cyclic frequency; f represents the spectral frequency; j represents the imaginary number; e represents the exponential function; S14, calculating the cyclic spectrum coherence function γ of the acoustic emission signal x(t) to be diagnosed x (α, f), for the cyclic spectrum correlation function S described in formula (2) x (α,f) is normalized in energy, specifically expressed as: S15, based on the cyclic spectrum coherence function γ x (α, f), combined with the spectrum estimation method, obtain the baseline discrete spectrum coherence mapping of the acoustic emission signal x(t) to be diagnosed and the discrete spectrum coherence mapping to be diagnosed 3. The mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationary according to claim 1 is characterized in that: In step S2, a first probability distribution p and a second probability distribution q are set, assuming that the first group of signal data X obeys X~p, and the second group of signal data Y obeys Y~q, and the maximum mean difference MMD metric It is expressed as: Where F represents the set of functions on the sample space; f(·) represents the mapping to the regenerated Hilbert space; n p and n q Respectively represent the number of samples of the first group of signal data X and the second group of signal data Y; i and j Respectively represent sample data of the first group of signal data X and the second group of signal data Y.

4. The method for diagnosing mechanical rotating parts acoustic emission faults based on bidirectional weighted cyclostationary according to claim 1 is characterized in that: The spectral ratio in step S32 is defined as: Among them, min(γ diff (α)) and max(γ diff (α)) respectively represent the differential demodulation spectrum γ diff The minimum and maximum values ​​of (α).

5. The mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationary according to claim 2 is characterized in that: In step S13, the cycle frequency α and the spectrum frequency f correspond to the time t and the delay τ before mapping respectively, and the cycle frequency α and the spectrum frequency f respectively indicate the modulation frequency and the carrier frequency of the acoustic emission signal x(t) to be diagnosed.

6. The mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationary according to claim 1 is characterized in that: In step S31, negative values ​​in the differential decomposition spectrum represent basic components, and positive values ​​represent fault-inducing factors.

7. The mechanical rotating parts acoustic emission fault diagnosis method based on bidirectional weighted cyclostationary according to claim 2 is characterized in that: In step S11, the acoustic emission signal to be diagnosed is subjected to a preprocessing operation, and the preprocessing operation includes high-frequency sampling, data truncation and de-averaging.

Images