Rolling bearing vibration signal feature extraction and analysis method in composite fault state

Abstract

The invention belongs to the technical field of rolling bearing diagnosis, and discloses a rolling bearing vibration signal feature extraction and analysis method in a composite fault state, which comprises the following steps: firstly, establishing a fault and frequency relation library, and obtaining and preprocessing an original vibration signal; carrying out frequency band division on the preprocessed signal through a multi-resolution signal decomposition method to obtain a frequency band signal group; for each fault type, related frequency band signals are selected from the frequency band signal group according to the theoretical frequency of the fault type to form a target frequency band signal group; by calculating a fault characteristic index, selecting an optimal frequency window as a target frequency range; and based on the target frequency range and the original vibration signal, obtaining a high-order statistical characteristic spectrum, and analyzing whether a corresponding fault exists according to the high-order statistical characteristic spectrum. Through a dynamic frequency band selection mechanism, the problem of improper selection of signal frequency bands in a composite fault state is effectively solved, and the accuracy and reliability of composite fault diagnosis in a complex environment are further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of rolling bearing diagnosis, and more specifically, to a method for extracting and analyzing the vibration signal characteristics of rolling bearings in a compound fault state. Background Art

[0002] As a key component in rotating machinery, the health monitoring and fault diagnosis of rolling bearings have become increasingly important. However, in the actual industrial environment, rolling bearings often have multiple faults simultaneously, and this compound fault state poses a great challenge to accurate diagnosis. In the face of compound faults, some existing methods diagnose by analyzing the vibration signals in the entire frequency band. Although this full-frequency band analysis method attempts to capture all possible fault characteristics, it also brings a series of problems. Analyzing the entire frequency band inevitably generates a large amount of redundant data, significantly increasing the computational complexity and storage burden. More critically, since different types of fault characteristics may be distributed in different frequency band ranges, full-frequency band analysis is prone to having useful fault information masked by noise and interference, especially when the characteristic frequencies of certain faults are very close, and it is difficult for this method to effectively distinguish different fault types.

[0003] To address the limitations of full-frequency band analysis, some techniques attempt to analyze by selecting specific frequency bands. This method aims to focus on the frequency range that may contain key fault information to improve the accuracy of diagnosis. However, these techniques often rely too much on experience or use fixed frequency ranges when selecting appropriate frequency bands for analysis. This static frequency band selection method is difficult to cope with the diversity of actual situations and cannot accurately capture the dynamic changes of various fault characteristics in the compound fault state. This analysis method is difficult to adapt to the complex and changeable industrial environment, and ultimately makes it difficult to accurately analyze the actual faults in the compound fault state. Summary of the Invention

[0004] In order to overcome the problem that it is difficult to accurately analyze the actual faults in the prior art under the compound fault state, the present invention proposes a method for extracting and analyzing the vibration signal characteristics of rolling bearings in the compound fault state to solve the above problems.

[0005] The present invention provides the following technical solutions: A method for extracting and analyzing the vibration signal characteristics of rolling bearings in a compound fault state, comprising: Establishing a fault-frequency relationship library, where the fault-frequency relationship library includes fault types and corresponding theoretical frequencies; Obtaining the original vibration signal of the rolling bearing and preprocessing the original vibration signal to obtain a preprocessed signal; The frequency band of the preprocessed signal is divided by a multi-resolution signal decomposition method to obtain a set of frequency band signals consisting of a series of frequency band signals; For each fault type, several frequency band signals are selected from the set of frequency band signals according to its corresponding theoretical frequency to form a target set of frequency band signals corresponding to each fault type; Calculate the fault feature index of each frequency band signal in the target set of frequency band signals corresponding to each fault type; obtain the frequency range of the frequency band signal with the largest fault feature index as the target frequency range corresponding to the fault type; According to the target frequency range corresponding to each fault type and the original vibration signal, obtain the high-order statistical characteristic spectrum corresponding to each fault type; Analyze whether there is a corresponding fault according to the high-order statistical characteristic spectrum, and count the analysis results of all fault types to obtain the final analysis result.

[0006] Preferably, the preprocessing of the original vibration signal to obtain the preprocessed signal includes: Obtain the original vibration signal, perform noise reduction processing on the original vibration signal; calculate the Hilbert transform of the noise-reduced original vibration signal; take the logarithm of the result of the Hilbert transform to obtain the preprocessed signal.

[0007] Preferably, the division of the frequency band of the preprocessed signal by the multi-resolution signal decomposition method to obtain a series of frequency band signals includes: Perform wavelet packet decomposition on the preprocessed signal; During the wavelet packet decomposition process, perform n-layer decomposition, where n is an integer greater than 1. The n-layer decomposition specifically includes: the first layer of decomposition divides the entire preprocessed signal into two sub-frequency bands with equal width; the nth layer of decomposition equally divides each sub-frequency band of the n-1th layer into two new sub-frequency bands; Retain the frequency band information of all decomposition levels to obtain a series of frequency band signals.

[0008] Preferably, the composition of the target set of frequency band signals corresponding to each fault type includes: Obtain the theoretical frequency corresponding to the fault type, and construct a harmonic sequence according to the theoretical frequency. The harmonic sequence is expressed as , where represents the theoretical frequency, is a preset integer greater than or equal to 5; Traverse all frequency band signals in the set of frequency band signals, and mark the frequency band signals containing at least one frequency in the harmonic sequence as candidate frequency band signals; For each candidate frequency band, use the following formula to calculate the matching degree of the frequencies it contains: , where Indicates the matching degree including frequency, Indicates the frequency value in the harmonic sequence included in the candidate frequency band, Indicates the center frequency of the candidate frequency band, Indicates the bandwidth of the candidate frequency band; Calculate the final matching degree of each candidate frequency band signal, where the final matching degree is the average value of the matching degrees of all included frequencies within the candidate frequency band signal; Select the candidate frequency band signals with the final matching degree greater than the preset threshold, and combine the selected candidate frequency band signals into a target frequency band signal group.

[0009] Preferably, the step of calculating the fault feature index of each frequency band signal in the target frequency band signal group corresponding to each fault type includes: For each frequency band signal in the target frequency band signal group, obtain the corresponding harmonic sequence, and obtain the frequencies included in the frequency band signal according to the harmonic sequence; For each included frequency, calculate the frequency saliency, frequency energy degree, and frequency distortion degree; Use the formula Calculate the frequency contribution index, where in the formula, Indicates the frequency contribution index, Indicates the frequency saliency, Indicates the frequency energy degree, Indicates the frequency distortion degree; Add up the frequency contribution indexes of all included frequencies of the frequency band signal to obtain the fault feature index of the frequency band signal.

[0010] Preferably, the frequency saliency is equal to the ratio of the included frequency amplitude to the maximum amplitude in the frequency band signal; the frequency distortion degree is equal to the square root of the ratio of the square of the included frequency amplitude to the square of the theoretical frequency amplitude; The calculation formula of the frequency energy degree is as follows: , Where in the formula, Indicates the frequency included in the frequency band signal, Indicates the preset frequency range parameter, Indicates the power spectral density function of the frequency band signal, Indicates the minimum frequency in the frequency band signal, Indicates the maximum frequency in the frequency band signal.

[0011] Preferably, the step of obtaining the high-order statistical feature spectrum corresponding to each fault type according to the target frequency range corresponding to each fault type and the original vibration signal includes: For each fault type, according to its corresponding target frequency range, perform band-pass filtering on the original vibration signal to obtain the filtered signal; calculate the high-order cumulants of the filtered signal, and obtain the high-order statistical feature spectrum according to the high-order cumulants; The high-order cumulants include second-order cumulants, third-order cumulants, and fourth-order cumulants; The obtaining of the high-order statistical feature spectrum according to the high-order cumulants includes performing Fourier transform on the second-order cumulants to obtain the power spectrum; performing Fourier transform on the third-order cumulants to obtain the bispectrum; performing Fourier transform on the fourth-order cumulants to obtain the trispectrum, and combining the power spectrum, bispectrum, and trispectrum to obtain the high-order statistical feature spectrum corresponding to this fault type.

[0012] Preferably, the analyzing whether there is a corresponding fault according to the high-order statistical feature spectrum includes: extracting analysis features from the high-order statistical feature spectrum of each fault type; inputting the analysis features and the fault type into a pre-trained fault discrimination model; obtaining the result of whether there is a fault output by the model.

[0013] Preferably, the extraction steps of the analysis features include: Calculating the peak value, mean value, variance, skewness, and kurtosis of the power spectrum; obtaining the bispectrum peak value and bispectrum entropy in the bispectrum; obtaining the trispectrum peak value and trispectrum energy in the trispectrum; combining the peak value, mean value, variance, skewness, and kurtosis of the power spectrum, the bispectrum peak value and bispectrum entropy in the bispectrum, and the trispectrum peak value and trispectrum energy in the trispectrum to obtain the analysis features.

[0014] Preferably, the pre-trained fault discrimination model includes: collecting vibration signal samples of rolling bearings with known fault types; for each sample, obtaining the target frequency range and obtaining the high-order statistical feature spectrum according to the target frequency range; extracting analysis features from the high-order statistical feature spectrum; using the extracted analysis features and the corresponding fault type labels as training data; Constructing a machine learning model, and using the training data to train the machine learning model to obtain the fault discrimination model.

[0015] The present invention provides a method for extracting and analyzing the characteristics of the vibration signal of a rolling bearing under a compound fault state, and has the following beneficial effects: For each type of fault, a harmonic sequence is constructed according to its theoretical frequency, and relevant band signals are selected from the band signal group through matching degree calculation. Subsequently, in the relevant band signals, each frequency component is evaluated multi-dimensionally by calculating the frequency significance, frequency energy degree, and frequency distortion degree. Considering the signal strength, energy distribution, and deviation from the theoretical frequency, the most suitable band for fault diagnosis is effectively identified. The optimal frequency window with the highest distinguishability of the fault characteristic signal relative to the interference signal is found. This dynamic selection mechanism effectively overcomes the problem of more interference signals in the band due to inappropriate selection of the signal band in the compound fault state. Finally, by analyzing different types of fault signals in different frequency bands, this solution can better separate and identify each component in the compound fault, thus significantly improving the accuracy and reliability of compound fault diagnosis in complex environments. Brief Description of the Drawings

[0016] Figure 1 It is a schematic flow chart of the method for extracting and analyzing the vibration signal characteristics of a rolling bearing under the compound fault state of the present invention. Detailed Embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1 Please refer to Figure 1 , in this embodiment, the method for extracting and analyzing the vibration signal characteristics of a rolling bearing under the compound fault state includes: S1. Establish a fault-frequency relationship library, where the fault-frequency relationship library includes fault types and corresponding theoretical frequencies; In this embodiment, the process of establishing the fault-frequency relationship library can be as follows: First, determine the specific fault types to be included in the relationship library. These fault types include but are not limited to: outer race pitting, outer race crack, inner race pitting, inner race crack, rolling element pitting, rolling element crack, cage fracture, etc. For each specific fault type, determine its corresponding theoretical frequency through experiments or actual operation data. These theoretical frequencies are usually the frequencies at which significant changes occur in the bearing vibration signal when this type of fault occurs. For example, for the outer race pitting fault, its theoretical frequency may be the frequency at which a certain amplitude in the vibration signal increases significantly.

[0019] S2. Obtain the original vibration signal of the rolling bearing and preprocess the original vibration signal to obtain a preprocessed signal; The preprocessing of the original vibration signal to obtain the preprocessed signal includes: Obtain the original vibration signal and perform noise reduction processing on the original vibration signal; calculate the Hilbert transform of the noise-reduced original vibration signal; take the logarithm of the result of the Hilbert transform to obtain the preprocessed signal.

[0020] In this embodiment, the process of obtaining and preprocessing the original vibration signal of the rolling bearing can be as follows: First, use vibration sensors to collect the original vibration signal of the rolling bearing. These sensors are usually installed at the bearing housing or key positions of the mechanical equipment to capture the vibration generated during the operation of the bearing. Since the collected original signal may be mixed with various noises. Therefore, the next step is to perform noise reduction processing on the original vibration signal. This step can adopt methods such as wavelet threshold denoising or empirical mode decomposition; after noise reduction, perform the Hilbert transform on the processed signal. The Hilbert transform can help obtain the instantaneous frequency and instantaneous amplitude of the signal. Finally, take the logarithm of the result of the Hilbert transform. The purpose of this step is to compress the dynamic range of the signal, making it easier to observe small amplitude changes, and it is also beneficial for subsequent feature extraction. Through these preprocessing steps, an optimized preprocessed signal is obtained. This signal retains the key features in the original vibration signal, while removing most of the noise and interference, laying a foundation for subsequent analysis.

[0021] S3. Divide the frequency band of the preprocessed signal by a multi-resolution signal decomposition method to obtain a series of frequency band signals that form a frequency band signal group; The dividing of the frequency band of the preprocessed signal by the multi-resolution signal decomposition method to obtain a series of frequency band signals includes: Perform wavelet packet decomposition on the preprocessed signal; During the wavelet packet decomposition process, perform n-layer decomposition, where n is an integer greater than 1. The n-layer decomposition specifically includes: The first layer of decomposition divides the entire preprocessed signal into two sub-frequency bands of equal width; the nth layer of decomposition divides each sub-frequency band of the n - 1th layer into two new sub-frequency bands; Retain the frequency band information of all decomposition levels to obtain a series of frequency band signals.

[0022] In this embodiment, the process of dividing the frequency band of the preprocessed signal by the multi-resolution signal decomposition method can be as follows: First, wavelet packet decomposition is selected as the multi-resolution signal decomposition method. When performing wavelet packet decomposition on the preprocessed signal, the number of decomposition levels n needs to be determined, where n is an integer greater than 1. The decomposition process starts from the first layer, and the entire preprocessed signal is divided into two sub-frequency bands with equal widths. These two sub-frequency bands correspond to the low-frequency part and the high-frequency part of the signal respectively. Then, continue the second layer of decomposition, and each of the two sub-frequency bands obtained in the first layer is equally divided into two new sub-frequency bands again, so that 4 sub-frequency bands are obtained. This process continues until the nth layer. When performing the nth layer of decomposition, each sub-frequency band in the n-1th layer is equally divided into two new sub-frequency bands. For example, if n = 3 is selected, then finally 8 sub-frequency bands will be obtained. It should be noted that the frequency band information of all decomposition levels is retained. This means that not only the sub-frequency bands of the last layer (the nth layer) are retained, but also the sub-frequency band information of the previous layers is retained. The advantage of doing this is that frequency band information of different scales can be obtained simultaneously, which is beneficial to capturing the characteristics of the signal in different frequency ranges. In this way, a series of frequency band signals covering different frequency ranges are finally obtained, and they jointly form a frequency band signal group.

[0023] S4. For each fault type, select several frequency band signals from the frequency band signal group according to its corresponding theoretical frequency to form a target frequency band signal group corresponding to each fault type; The formation of the target frequency band signal group corresponding to each fault type includes: Obtain the theoretical frequency corresponding to the fault type, and construct a harmonic sequence according to the theoretical frequency. The harmonic sequence is expressed as , where represents the theoretical frequency, is a preset integer greater than or equal to 5; Traverse all the frequency band signals in the frequency band signal group, and mark the frequency band signals containing at least one frequency in the harmonic sequence as candidate frequency band signals; For each candidate frequency band, use the following formula to calculate the matching degree of the frequencies it contains: , where represents the matching degree of the contained frequencies, represents the frequency value in the harmonic sequence contained in the candidate frequency band, represents the center frequency of the candidate frequency band, represents the bandwidth of the candidate frequency band; Calculate the final matching degree of each candidate frequency band signal. The final matching degree is the average value of the matching degrees of all the frequencies contained in the candidate frequency band signal; Select candidate band signals with a final matching degree greater than a preset threshold, and combine the selected candidate band signals into a target band signal group.

[0024] In this embodiment, for each fault type, the process of selecting several band signals from the band signal group according to its corresponding theoretical frequency to form a target band signal group may be as follows: First, obtain the theoretical frequency corresponding to the fault type. According to this theoretical frequency, construct a harmonic sequence. For example, if the theoretical frequency is 100 Hz and k is set to 5, then the harmonic sequence is: {100 Hz, 200 Hz, 300 Hz, 400 Hz, 500 Hz}. Next, traverse all the band signals in the band signal group. If a band signal contains at least one frequency in the harmonic sequence, mark it as a candidate band signal. For each candidate band, calculate the matching degree of the frequencies it contains. This matching degree reflects the relationship between the frequencies of the harmonic sequence contained in the candidate band and the center frequency and bandwidth of the band. The closer the frequency is to the center of the band, the higher the matching degree. Then, calculate the final matching degree of each candidate band signal. This final matching degree is the average value of the matching degrees of all the contained frequencies in the candidate band signal. Finally, select the candidate band signals with a final matching degree greater than the preset threshold, and combine these selected candidate band signals to form a target band signal group. By this method, for each fault type, the most relevant band signals can be screened out from the original band signal group, and these signals are likely to contain the characteristic information of the fault type.

[0025] S5. Calculate the fault feature index of each band signal in the target band signal group corresponding to each fault type; obtain the frequency range of the band signal with the largest fault feature index as the target frequency range corresponding to the fault type; The step of calculating the fault feature index of each band signal in the target band signal group corresponding to each fault type includes: For each band signal in the target band signal group, obtain the corresponding harmonic sequence, and obtain the frequencies contained in the band signal according to the harmonic sequence; For each contained frequency, calculate the frequency significance, frequency energy degree, and frequency distortion degree; Use the formula to calculate the frequency contribution index, where represents the frequency contribution index, represents the frequency significance, represents the frequency energy degree, represents the frequency distortion degree; Add up the frequency contribution indexes of all the contained frequencies of the band signal to obtain the fault feature index of the band signal.

[0026] The frequency significance is equal to the ratio of the included frequency amplitude to the maximum amplitude in the band signal; the frequency distortion is equal to the square root of the ratio of the square of the included frequency amplitude to the square of the theoretical frequency amplitude. The calculation formula of the frequency energy measure is as follows: , In the formula, represents the frequency included in the band signal, represents the preset frequency range parameter, represents the power spectral density function of the band signal, represents the minimum frequency in the band signal, represents the maximum frequency in the band signal.

[0027] In this embodiment, the process of calculating the fault feature index of each band signal in the target band signal group corresponding to each fault type may be as follows: First, for each band signal in the target band signal group, obtain its corresponding harmonic sequence, and determine the frequency included in the band signal according to this harmonic sequence. Next, perform three aspects of calculations on each included frequency: frequency significance, frequency energy measure, and frequency distortion, to comprehensively evaluate the characteristics of each frequency component.

[0028] Under the condition of compound faults, the signal at the theoretical frequency is often interfered by various factors, which may cause the characteristics at the theoretical frequency to be not obvious enough or masked. In contrast, the signal at the harmonic frequency may be less interfered, so it is easier to be used for fault diagnosis. Therefore, by comprehensively considering the frequency significance, energy measure, and distortion, calculate the contribution index of each frequency, and then obtain the fault feature index of each band signal. Through this method, it is possible to effectively identify the bands that are most suitable for fault diagnosis, rather than just the band with the largest signal intensity. Specifically, the band with a high fault feature index may contain relatively clear harmonic components. Although the amplitude of these components may not be the largest, they are relatively less interfered, so they are more conducive to accurately diagnosing faults.

[0029] Select the band signal with the largest fault feature index as the target frequency range, which means finding an optimal frequency window. Within this window, the distinguishability of the fault feature signal relative to the interference signal is the highest. By analyzing within this optimized frequency range, the fault features can be more accurately identified and quantified, and more accurate fault analysis results can be obtained even in the case of compound faults.

[0030] S6. According to the target frequency range corresponding to each fault type and the original vibration signal, obtain the high-order statistical characteristic spectrum corresponding to each fault type; Obtaining the high - order statistical characteristic spectrum corresponding to each fault type based on the target frequency range corresponding to each fault type and the original vibration signal includes: For each fault type, according to its corresponding target frequency range, perform band - pass filtering on the original vibration signal to obtain the filtered signal; calculate the high - order cumulants of the filtered signal, and obtain the high - order statistical characteristic spectrum according to the high - order cumulants; The high - order cumulants include second - order cumulants, third - order cumulants, and fourth - order cumulants; Obtaining the high - order statistical characteristic spectrum according to the high - order cumulants includes performing Fourier transform on the second - order cumulants to obtain the power spectrum; performing Fourier transform on the third - order cumulants to obtain the bispectrum; performing Fourier transform on the fourth - order cumulants to obtain the trispectrum, and combining the power spectrum, bispectrum, and trispectrum to obtain the high - order statistical characteristic spectrum corresponding to this fault type.

[0031] In this embodiment, for each fault type, the process of processing the original vibration signal according to its corresponding target frequency range to obtain the high - order statistical characteristic spectrum can be as follows: First, use the previously determined target frequency range to perform band - pass filtering on the original vibration signal. The purpose of this step is to highlight the signal components within the target frequency range while suppressing the interference from other frequency ranges. Band - pass filtering can be implemented using digital filters, such as Butterworth filters or Chebyshev filters. Next, calculate the high - order cumulants of the filtered signal. High - order cumulants are a powerful signal - processing tool, especially suitable for analyzing non - Gaussian signals and nonlinear systems, such as the vibration signals generated by bearing faults. In this embodiment, second - order, third - order, and fourth - order cumulants are calculated. After calculating the high - order cumulants, further process these cumulants to obtain the high - order statistical characteristic spectrum. Specifically, perform Fourier transform on the second - order cumulants to obtain the power spectrum. Perform Fourier transform on the third - order cumulants to obtain the bispectrum. Perform Fourier transform on the fourth - order cumulants to obtain the trispectrum. Finally, combine the power spectrum, bispectrum, and trispectrum together to form the high - order statistical characteristic spectrum corresponding to this fault type. This comprehensive characteristic spectrum contains the linear and nonlinear characteristics of the signal and can comprehensively reflect the characteristics of the fault.

[0032] S7. Analyze whether there is a corresponding fault according to the high - order statistical characteristic spectrum, and statistically analyze the results of all fault types to obtain the final analysis result.

[0033] Analyzing whether there is a corresponding fault according to the high - order statistical characteristic spectrum includes: extracting analysis features from the high - order statistical characteristic spectrum of each fault type; inputting the analysis features and the fault type into a pre - trained fault discrimination model; obtaining the result of whether there is a fault output by the model.

[0034] The steps for extracting the analysis features include: Calculate the peak value, mean value, variance, skewness, and kurtosis of the power spectrum; obtain the bispectrum peak value and bispectrum entropy in the bispectrum; obtain the trispectrum peak value and trispectrum energy in the trispectrum; combine the peak value, mean value, variance, skewness, and kurtosis of the power spectrum, the bispectrum peak value and bispectrum entropy in the bispectrum, and the trispectrum peak value and trispectrum energy in the trispectrum to obtain the analysis features.

[0035] The pre-trained fault discrimination model includes: collecting vibration signal samples of rolling bearings with known fault types; for each sample, obtaining the target frequency range and obtaining the high-order statistical feature spectrum according to the target frequency range; extracting analysis features from the high-order statistical feature spectrum; using the extracted analysis features and the corresponding fault type labels as training data; Construct a machine learning model and use the training data to train the machine learning model to obtain a fault discrimination model.

[0036] In this embodiment, the process of analyzing whether there is a corresponding fault according to the high-order statistical feature spectrum and obtaining the final analysis result by counting the analysis results of all fault types can be as follows: First, perform feature extraction on the high-order statistical feature spectra of each fault type to obtain analysis features. This process includes multiple steps: calculating the peak value, mean value, variance, skewness, and kurtosis of the power spectrum, and these statistics can comprehensively describe the distribution characteristics of the power spectrum. For the bispectrum, extract the bispectrum peak value and bispectrum entropy, and these features can reflect the nonlinear characteristics and complexity of the signal. From the trispectrum, obtain the trispectrum peak value and trispectrum energy, and these features can capture the high-order nonlinear characteristics of the signal. Combine these features together to obtain a comprehensive analysis feature set, which contains the important features of the signal at different statistical orders.

[0037] Next, input the extracted analysis features and the corresponding fault types into the pre-trained fault discrimination model. This fault discrimination model is constructed by machine learning methods. The training process of the model is as follows: First, a large number of vibration signal samples of rolling bearings with known fault types are collected. For each sample, obtain the target frequency range according to the method described above and calculate the high-order statistical feature spectrum according to this range. Then, extract analysis features from these feature spectra. These extracted analysis features, together with the corresponding fault type labels, constitute the training data set.

[0038] Use this training data set to construct and train a machine learning model. This model can be a support vector machine, random forest, deep neural network, etc. The goal of model training is to enable it to accurately map the input analysis features to the corresponding fault types.

[0039] In practical applications, the newly acquired analysis features and fault types are input into this trained model. The model obtains a fault type based on the analysis features. If this fault type is the same as the input fault type, then the result output by the model indicates the existence of the corresponding fault; otherwise, the corresponding fault does not exist. Finally, the analysis results of all fault types are counted to obtain the final analysis result. This final result includes all detected fault types.

[0040] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one type, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0041] As mentioned above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

[0042] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included within the protection scope of the present invention.

Claims

1. A method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state, characterized in that: include: Establishing a fault and frequency relationship library, wherein the fault and frequency relationship library includes fault types and corresponding theoretical frequencies; Acquire the original vibration signal of the rolling bearing, and preprocess the original vibration signal to obtain a preprocessed signal; Dividing the frequency band of the preprocessed signal by a multi-resolution signal decomposition method to obtain a series of frequency band signals to form a frequency band signal group; For each fault type, a number of frequency band signals are selected from the frequency band signal group according to its corresponding theoretical frequency to form a target frequency band signal group corresponding to each fault type; Calculating a fault characteristic index of each frequency band signal in a target frequency band signal group corresponding to each fault type; Obtaining the frequency range of the frequency band signal with the largest fault characteristic index as the target frequency range corresponding to the fault type; According to the target frequency range and the original vibration signal corresponding to each fault type, a high-order statistical characteristic spectrum corresponding to each fault type is obtained; According to the high-order statistical characteristic spectrum analysis, whether there is a corresponding fault, the analysis results of all fault types are counted to obtain the final analysis result.

2. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 1 is characterized in that: The preprocessing of the original vibration signal to obtain the preprocessed signal comprises: The original vibration signal is obtained and noise reduction is performed on the original vibration signal; the Hilbert transform of the original vibration signal after noise reduction is calculated; and the logarithm of the result of the Hilbert transform is taken to obtain a preprocessed signal.

3. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 2 is characterized in that: The frequency band of the preprocessed signal is divided by a multi-resolution signal decomposition method to obtain a series of frequency band signals including: Performing wavelet packet decomposition on the preprocessed signal; In the wavelet packet decomposition process, n-layer decomposition is performed, where n is an integer greater than 1, and the n-layer decomposition specifically includes: the first-layer decomposition divides the entire preprocessed signal into two equal-width sub-bands; the n-layer decomposition divides each sub-band of the n-1th layer into two new sub-bands; The frequency band information of all decomposition levels is retained to obtain a series of frequency band signals.

4. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 3 is characterized in that: The target frequency band signal group corresponding to each fault type comprises: Obtain the theoretical frequency corresponding to the fault type, and construct a harmonic sequence based on the theoretical frequency. The harmonic sequence is expressed as ,in represents the theoretical frequency, is a preset integer greater than or equal to 5; Traversing all frequency band signals in the frequency band signal group, marking frequency band signals containing at least one frequency in the harmonic sequence as candidate frequency band signals; For each candidate frequency band, the matching degree of its included frequencies is calculated using the following formula: ， in Indicates the matching degree of the included frequency, represents the frequency value in the harmonic sequence contained in the candidate frequency band, represents the center frequency of the candidate frequency band, represents the bandwidth of the candidate frequency band; Calculating a final matching degree of each candidate frequency band signal, wherein the final matching degree is an average value of matching degrees of all included frequencies in the candidate frequency band signal; The candidate frequency band signals whose final matching degree is greater than a preset threshold are selected, and the selected candidate frequency band signals are combined into a target frequency band signal group.

5. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 4 is characterized in that: The step of calculating the fault characteristic index of each frequency band signal in the target frequency band signal group corresponding to each fault type comprises: For each frequency band signal in the target frequency band signal group, obtain a corresponding harmonic sequence, and obtain the frequency contained in the frequency band signal according to the harmonic sequence; For each included frequency, the frequency significance, frequency energy and frequency distortion are calculated; Using the formula Calculate the frequency contribution index, where represents the frequency contribution index, Indicates the frequency significance, represents the frequency energy measure, Indicates frequency distortion; The frequency contribution indexes of all frequencies included in the frequency band signal are added together to obtain the fault characteristic index of the frequency band signal.

6. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 5, characterized in that: The frequency significance is equal to the ratio of the included frequency amplitude to the maximum amplitude in the frequency band signal; the frequency distortion is equal to the square root of the ratio of the square of the included frequency amplitude to the square of the theoretical frequency amplitude; The calculation formula of the frequency energy metric is as follows: ， In the formula, Indicates the frequencies contained in the frequency band signal, Indicates the preset frequency range parameters, represents the power spectral density function of the frequency band signal, represents the minimum frequency in the band signal, Indicates the maximum frequency in the band signal.

7. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 1, characterized in that: The step of obtaining a high-order statistical characteristic spectrum corresponding to each fault type according to the target frequency range and the original vibration signal corresponding to each fault type includes: For each fault type, the original vibration signal is band-pass filtered according to its corresponding target frequency range to obtain a filtered signal; the high-order cumulative amount of the filtered signal is calculated, and the high-order statistical characteristic spectrum is obtained according to the high-order cumulative amount; The high-order cumulants include second-order cumulants, third-order cumulants and fourth-order cumulants; The method of obtaining a high-order statistical characteristic spectrum based on high-order cumulative quantities includes performing Fourier transform on second-order cumulative quantities to obtain a power spectrum; performing Fourier transform on third-order cumulative quantities to obtain a bispectrum; performing Fourier transform on fourth-order cumulative quantities to obtain a trispectrum, and combining the power spectrum, bispectrum and trispectrum to obtain a high-order statistical characteristic spectrum corresponding to the fault type.

8. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 7, characterized in that: The analyzing whether there is a corresponding fault based on the high-order statistical characteristic spectrum includes: extracting features from the high-order statistical characteristic spectrum of each fault type to obtain analysis features; inputting the analysis features and the fault type into a pre-trained fault discrimination model; and obtaining a fault result output by the model.

9. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 8, characterized in that: The step of extracting the analysis features comprises: Calculate the peak, mean, variance, skewness and kurtosis of the power spectrum; obtain the bispectral peak and bispectral entropy in the bispectrum; obtain the trispectral peak and trispectral energy in the trispectrum; combine the peak, mean, variance, skewness and kurtosis of the power spectrum, the bispectral peak and bispectral entropy in the bispectrum, and the trispectral peak and trispectral energy in the trispectrum to obtain analysis features.

10. The method for extracting and analyzing rolling bearing vibration signal characteristics under a composite fault state according to claim 9, characterized in that: The pre-trained fault discrimination model comprises: collecting rolling bearing vibration signal samples of known fault types; for each sample, obtaining a target frequency range and obtaining a high-order statistical characteristic spectrum according to the target frequency range; extracting analysis features from the high-order statistical characteristic spectrum; using the extracted analysis features and corresponding fault type labels as training data; Build a machine learning model, use the training data to train the machine learning model, and obtain a fault discrimination model.

Images