CN115031966A

CN115031966A - Rolling bearing fault diagnosis method and device, electronic equipment and storage medium

Info

Publication number: CN115031966A
Application number: CN202210563536.9A
Authority: CN
Inventors: 张琪萱; 郑毅贤; 吴文超; 王达一
Original assignee: Siemens Ltd China
Current assignee: Siemens Ltd China
Priority date: 2022-05-23
Filing date: 2022-05-23
Publication date: 2022-09-09

Abstract

The embodiment of the application provides a rolling bearing fault diagnosis method, a rolling bearing fault diagnosis device, electronic equipment and a storage medium, wherein signal reconstruction is performed on a vibration signal of a rolling bearing, and each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band is obtained; and calculating the spectral kurtosis of each reconstructed sub-signal to determine the optimal sub-signal in each reconstructed sub-signal, acquiring the sub-signal to be analyzed of the vibration signal based on the preset signal characteristics of the optimal sub-signal, analyzing the sub-signal to be analyzed by using a fault analysis model, and determining the fault diagnosis result of the rolling bearing. Therefore, the accuracy of the fault diagnosis result can be improved, and better fault diagnosis and processing efficiency is achieved.

Description

Rolling bearing fault diagnosis method and device, electronic equipment and storage medium

Technical Field

The embodiment of the application relates to the technical field of signal analysis, in particular to a rolling bearing fault diagnosis method and device, electronic equipment and a computer storage medium.

Background

The rolling bearing is particularly important for normal operation of mechanical equipment, and in fault diagnosis of the rolling bearing, an early fault signal (early wear signal) of the rolling bearing is often generated in a high-frequency band and is easily submerged in a high-frequency noise signal, so that the early fault signal of the rolling bearing is difficult to be found and early warning is performed.

Secondly, in the actual industrial scene, due to the reasons of large noise, high equipment complexity and the like in the industrial field, when the rolling bearing is found to be in fault, the specific fault type (such as the faults of bearing inner ring abrasion, bearing outer ring abrasion and the like) of the rolling bearing is difficult to be subdivided by using the existing diagnosis rule.

In addition, most of the current rolling bearing fault diagnosis methods are performed by using a vibration analyzer in an off-line analysis manner, and the following problems mainly exist: real-time online analysis cannot be performed on the rolling bearing, so that fault early warning cannot be provided for the rolling bearing in time; different types of machines have different mechanical characteristics, the traditional vibration analysis rules are too general for analyzing the vibration signals of the machines, the analysis cost of a vibration analyzer is high, and all necessary expert experience knowledge of various types of rotary machines cannot be mastered; the traditional fault diagnosis method is realized based on fixed theoretical significance and mechanism, and is difficult to adapt to various different practical application scenes quickly and automatically.

In view of the above, there is a need for an improved fault diagnosis scheme for a rolling bearing, which can solve various problems in the prior art.

Disclosure of Invention

In order to solve the above problems, embodiments of the present application provide a rolling bearing fault diagnosis method, device, electronic device, and computer storage medium, so as to at least partially solve the above problems.

According to a first aspect of embodiments of the present application, there is provided a rolling bearing failure diagnosis method including: performing signal reconstruction aiming at a vibration signal of a rolling bearing to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band; calculating the spectral kurtosis of each reconstructed sub-signal to determine the optimal sub-signal in each reconstructed sub-signal, and acquiring the sub-signal to be analyzed of the vibration signal based on the preset signal characteristic of the optimal sub-signal; and analyzing the sub-signal to be analyzed by using a fault analysis model, and determining a fault diagnosis result of the rolling bearing.

Optionally, the performing signal reconstruction on the vibration signal of the rolling bearing, and acquiring each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band includes: obtaining an integer power of a preset base number closest to the fault frequency according to the fault frequency of the rolling bearing, and determining the number of division layers based on the integer power; and performing frequency spectrum division on the vibration signal according to the number of the divided layers to obtain a plurality of divided frequency bands, and constructing a filter bank of each divided frequency band to perform filtering on the vibration signal of the rolling bearing to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band.

Optionally, the step of obtaining each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band further includes: performing frequency band division on a bearing frequency band of the rolling bearing according to the division layer number, and determining each division frequency band; defining an empirical scale function and an empirical wavelet function of the filter bank according to each divided frequency band so as to filter the vibration signal of the rolling bearing and obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band;

the empirical scale function is defined as:

the empirical wavelet function is defined as:

wherein ω is frequency, ω is _n For frequency division of the nth divided band, said τ _n And the beta is a coefficient function for the transition frequency band of the nth divided frequency band.

Optionally, the calculating the spectral kurtosis of each reconstructed sub-signal to determine an optimal sub-signal of the reconstructed sub-signals includes: calculating the linear spectrum kurtosis of each reconstructed sub-signal, and determining the reconstructed sub-signal corresponding to the maximum one of the linear spectrum kurtosis as the optimal sub-signal; wherein the step of calculating the linear spectral kurtosis of each reconstructed sub-signal comprises: determining a reconstructed sub-signal as a current sub-signal; determining a random variable sequence of the current sub-signal corresponding to each candidate phase based on a plurality of candidate phases in succession; sequentially arranging the random variable sequence of each candidate stage according to the segment value of each candidate stage, and determining the candidate linear moment of any one candidate stage; determining a first linear moment of a first preset stage and a second linear moment of a second preset stage according to the candidate linear moment of any one candidate stage; determining a linear spectral kurtosis of the current sub-signal based on the first and second linear moments.

Optionally, the candidate linear moments of any one of the candidate phases are expressed as:

wherein, said λ _r Is a candidate linear moment of the r-th candidate stage, the E [ X _r-k|r ]Is a random variable sequence X _r-k Is a desired value of

And randomly selecting the combination number of k variables from the r-1 variables.

Optionally, the obtaining a sub-signal to be analyzed of the vibration signal based on the preset signal feature of the optimal sub-signal includes: and according to the central frequency spectrum and the optimal band-pass of the optimal sub-signal, performing signal extraction on the vibration signal to determine a sub-signal to be analyzed of the vibration signal.

Optionally, the analyzing the sub-signal to be analyzed by using a fault analysis model to determine a fault diagnosis result of the rolling bearing includes: extracting a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed; and performing prediction on a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on a plurality of given fault categories by using the fault analysis model, and determining a fault category prediction result of the sub-signal to be analyzed.

Optionally, the fault analysis model comprises a random forest model having a plurality of decision trees; wherein the determining a fault category prediction result of the sub-signal to be analyzed by performing prediction on a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed based on a given plurality of fault categories by using the fault analysis model comprises: performing prediction on a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on each given fault type by utilizing each decision tree in the random forest model to obtain a preliminary fault type prediction result of each decision tree; and counting the ticket number of each fault category according to the preliminary fault category prediction result of each decision tree, and determining the fault category with the largest ticket number as the final fault category prediction result of the sub-signal to be analyzed.

Optionally, the random forest model may be trained by: generating each decision tree in the random forest model by using the training set of the random forest model until a preset growth stopping condition is met, and obtaining the random forest model based on each generated decision tree; performing verification aiming at the random forest model by using the verification set of the random forest model to obtain a loss training loss function; and optimizing the random forest model based on the training loss function, returning to execute the training set using the random forest model, and generating each decision tree in the random forest model until the training loss function meets a preset convergence condition so as to obtain the trained random forest model.

Optionally, the generating, by using the training set of the random forest model, each decision tree in the random forest model until a preset growth stopping condition is met includes: a step of determining a kini coefficient, which is to take all the features which are not determined as the optimal features in the training set as candidate features and calculate the kini coefficient of the training set according to each segmentation feature of each candidate feature; a segmentation determining step, namely determining the candidate feature with the minimum kini coefficient and the segmentation feature of the candidate feature as an optimal feature and an optimal segmentation point respectively according to each kini coefficient, dividing the training set into a first subset and a second subset based on the optimal feature, the optimal segmentation point and each candidate feature in the training set, and distributing the first subset and the second subset to two newly generated child nodes in a decision tree; and recursively calling the step of determining the kini coefficients and the step of determining the segmentation for each of the two child nodes until a preset growth stopping condition is met.

Optionally, the preset growth stopping condition comprises any one of the following conditions: the number of training samples corresponding to the candidate features currently contained in the training set is less than a preset sample number threshold; the kini coefficient of the training set is smaller than a preset coefficient threshold value; the number of candidate features currently contained in the training set does not satisfy the number of features required for generating a new node.

According to a second aspect of the embodiments of the present application, there is provided a rolling bearing failure diagnosis device including: the signal extraction module is used for executing signal reconstruction aiming at a vibration signal of a rolling bearing, acquiring each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band, calculating the spectral kurtosis of each reconstructed sub-signal to determine the optimal sub-signal in each reconstructed sub-signal, and acquiring a sub-signal to be analyzed of the vibration signal based on the preset signal characteristic of the optimal sub-signal; and the fault analysis module is used for analyzing the sub-signal to be analyzed by using a fault analysis model and determining a fault diagnosis result of the rolling bearing.

According to a third aspect of embodiments of the present application, there is provided an electronic apparatus, including: the system comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete mutual communication through the communication bus; the memory is used for storing at least one executable instruction, and the executable instruction enables the processor to execute the operation corresponding to the rolling bearing fault diagnosis method of the first aspect.

According to a fourth aspect of embodiments of the present application, there is provided a computer storage medium having stored thereon a computer program which, when executed by a processor, implements the rolling bearing fault diagnosis method described above in the first aspect.

According to the rolling bearing fault diagnosis scheme provided by the embodiment of the application, the vibration signals of the rolling bearing are reconstructed, the spectral kurtosis of each reconstructed sub-signal is calculated, the sub-signals to be analyzed, which are in a fault high-frequency generation band and have background noise filtered, in the vibration signals are obtained from the vibration signals, and fault analysis of the rolling bearing is executed according to the sub-signals, so that real-time online analysis of the rolling bearing faults can be realized, the accuracy of the rolling bearing fault diagnosis result can be improved, and timely early warning of early fault signals of the rolling bearing can be realized.

According to the fault diagnosis scheme of the rolling bearing, the number of the divided frequency bands of the vibration signal is determined based on the fault frequency of the rolling bearing, so that the problem that the operation load of a system is too large due to too many divided frequency bands is solved, or the problem that the analysis result is inaccurate due to too few divided frequency bands is solved.

According to the rolling bearing fault diagnosis scheme provided by the embodiment of the application, the empirical scale function and the empirical wavelet function are constructed according to the divided frequency bands, and the reconstructed sub-signals corresponding to the divided frequency bands in the vibration signal are extracted, so that the background noise in the vibration signal can be effectively filtered on the premise of not losing important information components in the original vibration signal, the clearer reconstructed sub-signals are obtained, the time resolution of high frequency is higher, the frequency resolution of low frequency is better, the accuracy of subsequent fault diagnosis results is improved, the band-pass of each divided frequency band can be automatically given under the condition of no human intervention, and the labor cost is saved.

According to the rolling bearing fault diagnosis scheme provided by the embodiment of the application, the optimal sub-signals in the reconstruction sub-signals are determined by calculating the linear spectrum kurtosis of the reconstruction sub-signals, so that the frequency bands with high fault risks in the vibration signals can be more conveniently located, and the waveforms of other frequency bands in the vibration signals are filtered, so that more accurate fault reason analysis and fault location are realized.

According to the rolling bearing fault diagnosis scheme provided by the embodiment of the application, the frequency domain characteristics and the time domain characteristics of the sub-signals to be analyzed are extracted, so that relatively complete fault signal characteristics can be provided, and a fault analysis model can perform fault diagnosis and analysis more accurately and efficiently.

According to the fault diagnosis scheme of the rolling bearing, the random forest algorithm is used for executing fault analysis and prediction aiming at the extracted frequency domain characteristics and time domain characteristics, so that fault diagnosis of the rolling bearing is more intelligent and convenient, the rolling bearing is suitable for various different practical application scenes, and a more refined fault category diagnosis result can be provided, so that personnel can timely perform troubleshooting and equipment maintenance.

According to the rolling bearing fault diagnosis scheme provided by the embodiment of the application, various growth stopping conditions are set, so that each decision tree in a random forest model can be accurately and efficiently generated, and the training efficiency of the model is improved.

According to the rolling bearing fault diagnosis scheme provided by the embodiment of the application, the random forest model is trained by using the cross validation algorithm, so that the robustness of the model prediction result can be improved.

Drawings

The drawings are only for purposes of illustrating and explaining the present application and are not to be construed as limiting the scope of the present application. Wherein the content of the first and second substances,

fig. 1 is a process flow chart of a rolling bearing failure diagnosis method according to an exemplary embodiment of the present application.

Fig. 2 is a process flow chart of a rolling bearing fault diagnosis method according to another exemplary embodiment of the present application.

Fig. 3 is a process flow chart of a rolling bearing fault diagnosis method according to another exemplary embodiment of the present application.

Fig. 4 is a process flow chart of a rolling bearing fault diagnosis method according to another exemplary embodiment of the present application.

Fig. 5 is a process flow chart of a rolling bearing fault diagnosis method according to another exemplary embodiment of the present application.

Fig. 6 is a process flow chart of a rolling bearing failure diagnosis method according to another exemplary embodiment of the present application.

Fig. 7 is a block diagram showing the structure of a rolling bearing failure diagnosis apparatus according to an exemplary embodiment of the present application.

Fig. 8 is a block diagram of an electronic device according to an exemplary embodiment of the present application.

Description of the reference numerals:

700. a rolling bearing failure diagnosis device; 702. an electronic device; 704. a fault analysis model 704; 800. an electronic device; 802. a processor; 804. a communication interface; 806. a memory; 808. a communication bus; 810. a computer program.

Detailed Description

In order to make the technical features, objects and effects of the embodiments of the present application more clearly understood, specific embodiments of the present application will now be described with reference to the accompanying drawings.

Fig. 1 shows a process flow of a rolling bearing failure diagnosis method of an exemplary embodiment of the present application. As shown in the figure, the present embodiment mainly includes the following processing steps:

and step S102, signal reconstruction is carried out on the vibration signals of the rolling bearing, and each reconstructed sub-signal of each divided frequency band corresponding to the vibration signals is obtained.

Alternatively, the number of divided layers may be determined according to the failure frequency of the rolling bearing.

Alternatively, frequency spectrum division may be performed on the vibration signal according to the number of divided layers to obtain a plurality of divided frequency bands, and filtering may be performed on the vibration signal of the rolling bearing based on each divided frequency band to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band.

Specifically, an Empirical Wavelet Transform (EWT) algorithm may be used to perform adaptive segmentation of a frequency spectrum for a vibration signal based on the number of division layers to obtain a plurality of segmented frequency bands, and a filter bank may be constructed on each segmented frequency band to perform filtering for the vibration signal of the rolling bearing to obtain each reconstructed sub-signal of the vibration signal corresponding to each segmented frequency band.

The filter bank may include a plurality of filters, and each filter corresponds to each divided frequency band.

And step S104, calculating the spectral kurtosis of each reconstructed sub-signal to determine the optimal sub-signal in each reconstructed sub-signal, and acquiring the sub-signal to be analyzed of the vibration signal based on the preset signal characteristics of the optimal sub-signal.

Alternatively, the linear spectral kurtosis of each reconstructed sub-signal may be calculated, and the reconstructed sub-signal corresponding to the largest one of the linear spectral kurtosis is determined as the optimal sub-signal.

Optionally, a center frequency spectrum and an optimal band pass of the optimal sub-signal may be determined, and signal extraction may be performed on the vibration signal according to the determined center frequency and optimal band pass to determine a sub-signal to be analyzed of the vibration signal.

And S106, analyzing the sub-signals to be analyzed by using the fault analysis model, and determining the fault diagnosis result of the rolling bearing.

Optionally, a plurality of time domain features and a plurality of frequency domain features in the sub-signal to be analyzed may be extracted for the fault analysis model to perform the prediction.

Alternatively, the fault analysis model may perform prediction for each time domain feature and each frequency domain feature in the sub-signal to be analyzed based on a given plurality of fault categories, output a probability value corresponding to each fault category, and determine a fault diagnosis result of the sub-signal to be analyzed based on the fault category with the highest probability value.

Optionally, the fault analysis model may include a random forest model.

In summary, according to the rolling bearing fault diagnosis method in the embodiment of the present application, the reconstructed sub-signals of the vibration signal corresponding to the divided frequency bands are obtained, the sub-signal to be analyzed is obtained from the vibration signal according to the spectral kurtosis of the reconstructed sub-signals, and the fault analysis of the rolling bearing is performed according to the sub-signal to be analyzed. Therefore, the fault risk high-frequency-emission frequency band in the vibration signal can be conveniently positioned, the waveforms of other frequency bands in the vibration signal are filtered, the sub-signal to be analyzed, with background noise effectively filtered, can be obtained from the vibration signal, fault cause analysis and fault positioning can be accurately executed subsequently, and the technical effect of early warning at the early stage of the fault signal is achieved.

Specifically, the embodiment of the present application uses spectral kurtosis to indicate a frequency band where a transient occurs, and uses an optimal band-pass filter to remove background noise in a vibration signal. And determining a low-frequency region according to the bandwidth of the band-pass filter, and calculating a bispectrum of the envelope of the band-pass filter signal on the low-frequency region so as to diagnose and analyze the fault of the rolling bearing. Therefore, the embodiment of the application can effectively inhibit background noise so as to improve the accuracy of fault diagnosis of the rolling bearing.

Fig. 2 shows a process flow of a rolling bearing failure diagnosis method according to another exemplary embodiment of the present application. This example is a specific implementation of step S102. As shown in the figure, the present embodiment mainly includes the following steps:

step S202, obtaining an integer power of a preset base number closest to the fault frequency according to the fault frequency of the rolling bearing, and determining the number of division layers based on the integer power.

Alternatively, the preset base number may be set to 2.

Specifically, assuming that the failure frequency of the rolling bearing is 100, the preset base number is 2, wherein the 6 th power of 2 is 64, and the difference between the failure frequency and the base number is 36; the power of 7 of 2 is 128, which is a difference of 28 from the failure frequency 100, and thus it can be determined that the integer power of 2 (M) closest to the failure frequency 100 should be 7, thereby determining the number of division layers.

In this embodiment, the number of division layers and the integer power satisfy the following relation:

the number of division layers (M) < integer power (M) + 1.

Step S204, performing frequency spectrum division on the vibration signal according to the number of divided layers to obtain a plurality of divided frequency bands, and constructing a filter bank of each divided frequency band to perform filtering on the vibration signal of the rolling bearing to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band.

Alternatively, frequency division can be performed on a bearing frequency band of the rolling bearing according to the number of divided layers, each divided frequency band is determined, an empirical scale function and an empirical wavelet function of a filter bank are defined according to each divided frequency band, filtering is performed on a vibration signal of the rolling bearing, and each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band is obtained, so that a high-frequency signal in the vibration signal can be better separated and extracted.

The empirical scaling function is defined as the following equation 1:

the empirical wavelet function is defined as the following equation 2:

in the above equations 1 and 2, ω is a frequency, ω is _n For frequency division of the nth divided band, τ _n And beta is a coefficient function for the transition frequency band of the nth divided frequency band.

In summary, according to the fault frequency of the rolling bearing, the integer power of the preset base number closest to the fault frequency is obtained, and the appropriate number of the dividing layers is determined, so that the problem of overlarge system operation load caused by too many dividing layers or the problem of inaccurate analysis result caused by too few dividing layers can be avoided.

In addition, the embodiment of the application utilizes the empirical wavelet empirical algorithm, and can perform self-adaptive decomposition and reconstruction on the vibration signal without manual intervention, so that the time resolution of high frequency is higher, the frequency resolution of low frequency is better, the accuracy of signal diagnosis results is favorably improved, and the labor cost can be reduced.

Fig. 3 shows a process flow of a rolling bearing failure diagnosis method according to another exemplary embodiment of the present application. This embodiment is a specific implementation mainly showing the calculation of the linear spectral kurtosis of each reconstructed sub-signal in step S104. As shown in the figure, the present embodiment mainly includes the following steps:

in step S302, a reconstructed sub-signal is determined as the current sub-signal.

Specifically, one reconstructed sub-signal may be sequentially acquired and determined as the current sub-signal.

Step S304, determining a random variable sequence of the current sub-signal corresponding to each candidate phase based on a plurality of consecutive candidate phases.

In particular, for a plurality of successive candidate phases 1, 2, …, r, it may be determined that the current sub-signal corresponds to a random variable sequence X for each candidate phase _1|r ，X _2|r ，…，X _r|r 。

Step S306, according to the segment value of each candidate stage, sequentially arranging the random variable sequence of each candidate stage, and determining the candidate linear moment of any one candidate stage.

For example, the random variable sequence of each candidate stage may be sequentially arranged according to the segment value size of each candidate stage, resulting in an ordered sequence, which is expressed as: x _1|r ≤X _2|r ≤…≤X _r|r And based on the sequence, obtaining a candidate linear moment of any one candidate stage, which can be expressed as the following formula 3:

wherein λ is _r Is a candidate linear moment of the r-th candidate stage, E [ X ] _r-k|r ]Is a random variable sequence X _r-k The expected value of (c) is,

In this example, E [ X ] _r-k|r ]The value of (b) is the sum and average of all random variable sequences.

Step S308, determining a first linear moment of the first preset stage and a second linear moment of the second preset stage according to the candidate linear moment of any one of the candidate stages.

Optionally, the first preset stage and the second preset stage may be a second stage and a fourth stage, respectively, and the first linearity of the second stage is determinedMoment lambda ₂ And a second linear moment λ of the fourth stage ₂ 。

Step S310, determining the linear spectrum kurtosis of the current subsignal based on the first linear moment and the second linear moment.

Optionally, the first linear moment λ may be according to the second stage ₂ And a second linear moment λ of the fourth stage ₂ Determines the linear spectral kurtosis of the current subsignal, which is expressed as the following equation 4:

in the above equation 4, τ is the linear spectral kurtosis of the current sub-signal

Step S312, determining whether the linear spectral kurtosis of all the reconstructed sub-signals is determined, if yes, executing step S314, otherwise, returning to step S302 to obtain the next reconstructed sub-signal, and executing the calculation of the linear spectral kurtosis.

In step S314, the reconstructed sub-signal with the highest linear spectral kurtosis is determined as the optimal sub-signal.

Specifically, an optimal sub-signal of each reconstructed sub-signal may be determined according to a maximum value of each linear spectral kurtosis corresponding to each reconstructed sub-signal.

And step S316, performing signal extraction on the vibration signal according to the central frequency spectrum and the optimal band-pass of the optimal sub-signal to determine a sub-signal to be analyzed of the vibration signal.

Specifically, the central frequency spectrum and the optimal band pass of the optimal sub-signal can be applied to the vibration signal of the rolling bearing to obtain the sub-signal to be analyzed corresponding to the fault risk frequency band in the vibration signal.

In summary, the embodiment of the present application can facilitate fast and accurate positioning of a frequency band with the highest fault risk in a vibration signal by calculating the linear spectral kurtosis of each reconstructed sub-signal, and perform filtering on waveforms of other frequency bands in the vibration signal, thereby facilitating improvement of subsequent fault analysis and fault positioning processing effects.

Fig. 4 shows a process flow of a rolling bearing failure method according to another exemplary embodiment of the present application. As shown, the present embodiment is mainly a specific implementation of the step S106. As shown in the figure, the present embodiment mainly includes the following steps:

step S402, extracting a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed.

Alternatively, the number of frequency domain features and the number of time domain features extracted from the sub-signal to be analyzed may be the same or different.

And S404, predicting a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on a plurality of given fault categories by using the fault analysis model, and determining the fault category of the sub-signal to be analyzed.

Alternatively, the fault analysis model may be a random forest model comprising a plurality of decision trees.

Specifically, the random forest may be a decision tree model based on a bagging algorithm framework.

Optionally, prediction may be performed on a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed based on given fault categories by using each decision tree in the random forest model, so as to obtain a preliminary fault category prediction result of each decision tree, count the number of votes of each fault category according to the preliminary fault category prediction result of each decision tree, and determine the fault category with the largest number of votes as the final fault category prediction result of the sub-signal to be analyzed.

For example, assuming that the random forest model includes 10 decision trees, each decision tree may be used to perform prediction on a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed based on given fault categories (for example, a given fault category includes three categories, namely, fault category a, fault category B, and fault category C), output a preliminary fault prediction result of each decision tree (i.e., output one of fault category a, fault category B, and fault category C), and determine the fault category B as a final fault category prediction result of the sub-signal to be analyzed if the number of votes obtained for the fault category B is the largest among all 10 preliminary fault category prediction results.

To sum up, this application embodiment can provide comparatively complete fault signal characteristic through a plurality of frequency domain characteristics and a plurality of time domain characteristics of extracting the sub-signal of waiting to analyze to the fault analysis model can be more accurate and high-efficient carry out fault diagnosis analysis.

Moreover, the fault analysis and prediction are executed by utilizing the random forest algorithm, so that the fault diagnosis of the rolling bearing is more intelligent and convenient, and the method can be adaptive to various application scenes.

Fig. 5 shows a process flow of a rolling bearing failure diagnosis method according to another exemplary embodiment of the present application. This example shows a specific training scheme for a random forest model. As shown in the figure, the present embodiment mainly includes the following steps:

and S802, generating each decision tree in the random forest model by using the training set of the random forest model until a preset growth stopping condition is met, and acquiring the random forest model based on each generated decision tree.

Specifically, N training samples may be extracted from the sample set of the random forest model based on a random and playback manner, and a training set D composed of K sets of training samples may be obtained by repeating the extraction operation of the training samples K times.

Optionally, a CART algorithm may be used to generate each decision tree in the random forest model based on the training set D and a preset growth stopping condition, so as to obtain the random forest model.

Alternatively, the sample set of random forest models may be obtained from a bearing internal fault dataset, a bearing external fault dataset, a bearing ball fault dataset, and the like.

And step S804, verifying the random forest model by using a verification set of the random forest model to obtain a training loss function.

Optionally, the samples of the random forest model may be collected, and the training samples that are not extracted as the training set may be used as the material of the verification set.

Optionally, the current random forest model may be used to perform prediction on the training samples in the verification set to obtain a prediction result of the training samples, and the prediction result of the training samples is compared with the real label to obtain a training loss function.

Step S806, determine whether the training loss function satisfies a predetermined convergence condition, if yes, execute step S808, and if not, return to execute step S802.

Alternatively, the determination result that the training loss function satisfies the preset convergence condition may be obtained when the training loss function satisfies the preset threshold.

Alternatively, the determination result that the training loss function satisfies the preset convergence condition may be obtained when the training loss function tends to be stable.

And step S808, obtaining the trained random forest model.

In summary, the random forest model is trained by using the cross validation algorithm, so that the robustness of the model prediction result can be effectively improved.

Fig. 6 shows a process flow of a rolling bearing failure diagnosis method according to another exemplary embodiment of the present application. This example is a specific implementation of step S804. As shown in the figure, the present embodiment mainly includes the following steps:

step S602, executing a kini coefficient determining step, taking all features in the training set that are not determined as optimal features as candidate features, and calculating the kini coefficient of the training set according to each segmented feature of each candidate feature.

In particular, the number of segmented features M may be given according to the number M of all features in the training set, wherein the number M of segmented features should be much smaller than the number M of all features in the training set.

Alternatively, the equation for calculating the Gini (Gini) coefficient is shown in the following equation 5:

in the above equation 5, D is the training set, | D | represents the number of samples included in D, and a is the candidate feature, wherein, in the case that the candidate feature a is the segmentation feature a (i.e., a ═ a), the division is performed on the training set D, and two parts D1 and D2 are obtained, | D1| is the number of samples included in D1, | D2| is the number of samples included in D2.

Step S604, executing a segmentation determining step, determining the candidate feature with the minimum Gini coefficient and the segmentation feature of the candidate feature as an optimal feature and an optimal segmentation point according to each Gini coefficient, dividing the training set into a first subset and a second subset based on the optimal feature, the optimal segmentation point and each candidate feature in the training set, and distributing the first subset and the second subset to two newly generated child nodes in the decision tree.

Specifically, based on the optimal features and the optimal splitting points, an optimal splitting manner of the current node in the decision tree may be calculated, so as to perform splitting on the training set to obtain a first subset and a second subset, and the first subset and the second subset are allocated to two newly generated child nodes under the current node.

Step S606, for each of the two child nodes, recursively invoking a step of determining a kini coefficient and a step of determining a cut until a preset growth stop condition is satisfied.

Specifically, the kini coefficient determination processing of step S602 and the slicing determination processing of step S604 may be recursively invoked for the child node corresponding to the first subset and the child node corresponding to the second subset, respectively, until a preset growth stop condition is satisfied.

Optionally, if the number of training samples corresponding to the candidate features currently included in the training set is less than the threshold of the number of preset samples, a determination result meeting a preset growth stopping condition may be obtained.

Specifically, if the number of training samples currently remaining in the training set is smaller than the preset sample number threshold, which indicates that the number of training samples currently remaining in the training set cannot generate a new node, a determination result meeting a preset growth stop condition may be obtained.

Optionally, if the kini coefficient of the training set is smaller than the preset coefficient threshold, a judgment result meeting a preset growth stopping condition may be obtained.

Specifically, if the kini coefficient of the training set is smaller than the preset coefficient threshold value, which indicates that the currently remaining training samples in the training set belong to the same category, a judgment result meeting the preset growth stopping condition can be obtained.

Optionally, if the number of candidate features currently included in the training set does not satisfy the number of features required for generating a new node, a determination result satisfying a preset growth stop condition may be obtained.

For example, if the candidate features currently included in the training set are 3, and 5 features are required to be used for generating a new node, which means that the new node cannot be generated based on the number of candidate features currently included in the training set, a determination result satisfying a preset growth stop condition may be obtained.

In summary, according to the rolling bearing fault diagnosis method provided by the embodiment of the application, various growth stopping conditions are set, so that each decision tree in a random forest model can be accurately and efficiently generated, and the training efficiency of the model is improved.

Fig. 7 shows a block diagram of the structure of a rolling bearing failure diagnosis device of an exemplary embodiment of the present application. As shown in the drawing, the rolling bearing failure diagnosis device 700 of the present embodiment mainly includes: signal extraction module 702, fault analysis model 704.

The signal extraction module 702 is configured to perform signal reconstruction on a vibration signal of a rolling bearing, acquire each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band, calculate a spectral kurtosis of each reconstructed sub-signal to determine an optimal sub-signal in each reconstructed sub-signal, and acquire a sub-signal to be analyzed of the vibration signal based on a preset signal feature of the optimal sub-signal.

And the fault analysis module 704 is used for analyzing the sub-signal to be analyzed by using a fault analysis model and determining a fault diagnosis result of the rolling bearing.

Optionally, the signal extraction module 702 is further configured to: obtaining an integer power of a preset base number closest to the fault frequency according to the fault frequency of the rolling bearing, and determining the number of division layers based on the integer power; and performing frequency spectrum division on the vibration signal according to the number of the divided layers to obtain a plurality of divided frequency bands, and constructing a filter group of each divided frequency band to perform filtering on the vibration signal of the rolling bearing so as to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band.

Optionally, the signal extraction module 702 is further configured to: performing frequency band division on a bearing frequency band of the rolling bearing according to the division layer number, and determining each division frequency band; defining an empirical scale function and an empirical wavelet function of the filter bank according to each divided frequency band so as to filter the vibration signal of the rolling bearing and obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band;

the empirical scale function is defined as:

the empirical wavelet function is defined as:

Optionally, the signal extraction module 702 is further configured to: and calculating the linear spectrum kurtosis of each reconstructed sub-signal, and determining the reconstructed sub-signal corresponding to the maximum one of the linear spectrum kurtosis as the optimal sub-signal.

Optionally, the signal extraction module 702 is further configured to: determining a reconstructed sub-signal as a current sub-signal; determining a random variable sequence of the current sub-signal corresponding to each candidate phase based on a plurality of candidate phases in succession; sequentially arranging the random variable sequence of each candidate stage according to the segment value of each candidate stage, and determining the candidate linear moment of any one candidate stage; determining a first linear moment of a first preset stage and a second linear moment of a second preset stage according to the candidate linear moment of any one candidate stage; determining a linear spectral kurtosis of the current sub-signal based on the first and second linear moments.

And randomly selecting a combination number of k variables from the r-1 variables.

Optionally, the signal extraction module 702 is further configured to: and according to the central frequency spectrum and the optimal band-pass of the optimal sub-signal, performing signal extraction on the vibration signal to determine a sub-signal to be analyzed of the vibration signal.

Optionally, the fault analysis model 704 is further configured to: extracting a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed; and performing prediction on a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on a plurality of given fault categories by using the fault analysis model, and determining a fault category prediction result of the sub-signal to be analyzed.

Optionally, the fault analysis model comprises a random forest model having a plurality of decision trees.

Optionally, the fault analysis model 704 is further configured to: performing prediction on a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on given fault types by utilizing each decision tree in the random forest model to obtain a preliminary fault type prediction result of each decision tree; and counting the number of votes of each fault category according to the preliminary fault category prediction result of each decision tree, and determining the fault category with the largest number of votes as the final fault category prediction result of the sub-signal to be analyzed.

Optionally, the rolling bearing fault diagnosis apparatus 700 further includes a model training module (not shown) configured to generate, by using a training set of the random forest model, each decision tree in the random forest model until a preset growth stopping condition is met, and obtain the random forest model based on each generated decision tree; and optimizing the random forest model based on the training loss function, returning to execute the training set using the random forest model, and generating each decision tree in the random forest model until the training loss function meets a preset convergence condition so as to obtain the trained random forest model.

Optionally, the model training module is further configured to: executing a kini coefficient determining step, taking all the features which are not determined as optimal features in the training set as candidate features, and calculating the kini coefficient of the training set according to each segmentation feature of each candidate feature; executing a segmentation determining step, namely determining the candidate feature with the minimum Gini coefficient and the segmentation feature of the candidate feature as an optimal feature and an optimal segmentation point according to each Gini coefficient, dividing the training set into a first subset and a second subset based on the optimal feature, the optimal segmentation point and each candidate feature in the training set, and distributing the first subset and the second subset to two newly generated child nodes in a decision tree; and recursively calling the step of determining the kini coefficient and the step of determining the segmentation for each of the two child nodes until a preset growth stopping condition is met.

Optionally, the preset growth stopping condition comprises any one of the following conditions: the number of training samples corresponding to the candidate features currently contained in the training set is less than a preset sample number threshold; the Gini coefficient of the training set is smaller than a preset coefficient threshold; the number of candidate features currently contained in the training set does not satisfy the number of features required for generating a new node.

The rolling bearing fault diagnosis device provided by the embodiment of the invention corresponds to the rolling bearing fault diagnosis method provided by the embodiment of the invention, and other descriptions can refer to the description of the rolling bearing fault diagnosis method provided by the embodiment of the invention, and are not repeated herein.

Another embodiment of the present invention provides an electronic device, including: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface are communicated with each other through the communication bus.

Fig. 8 is a block diagram of an electronic device according to an exemplary embodiment of the invention, and as shown in fig. 8, the electronic device 800 of this embodiment may include a processor (processor) 802, a communication interface (communication interface)804, and a memory (memory) 806.

The processor 802, communication interface 804, and memory 806 may communicate with one another via a communication bus 808.

The communication interface 804 is used for communication with other electronic devices such as a terminal device or a server.

The processor 802, configured to execute the computer program 810, may specifically execute the relevant steps in the embodiments of the methods described above, that is, execute the steps in the fault diagnosis method for the rolling bearing described in the embodiments described above.

In particular, the computer program 810 may comprise program code comprising computer operating instructions.

The processor 802 may be a central processing unit CPU, or an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits configured to implement embodiments of the present invention. The electronic device includes one or more processors, which may be the same type of processor, such as one or more CPUs; or may be different types of processors such as one or more CPUs and one or more ASICs.

A memory 806 for storing a computer program 810. The memory 806 may include high-speed RAM memory, and may also include non-volatile memory (non-volatile memory), such as at least one disk memory.

Another embodiment of the present invention provides a computer storage medium having a computer program stored thereon, which when executed by a processor, can implement the rolling bearing fault diagnosis method described in each of the above embodiments.

It should be noted that, according to implementation requirements, each component/step described in the embodiment of the present invention may be divided into more components/steps, and two or more components/steps or partial operations of the components/steps may also be combined into a new component/step to achieve the purpose of the embodiment of the present invention.

The above-described method according to an embodiment of the present invention may be implemented in hardware, firmware, or as software or computer code storable in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk, or a magneto-optical disk, or as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium downloaded through a network and to be stored in a local recording medium, so that the method described herein may be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware such as an ASIC or FPGA. It is understood that the computer, processor, microprocessor controller or programmable hardware includes memory components (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code that when accessed and executed by the computer, processor or hardware, implements the rolling bearing fault diagnostic methods described herein. Further, when a general-purpose computer accesses code for implementing the rolling bearing fault diagnosis method shown herein, execution of the code converts the general-purpose computer into a special-purpose computer for executing the rolling bearing fault diagnosis method shown herein.

Those of ordinary skill in the art will appreciate that the various illustrative elements and method steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the technical solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present embodiments.

The above embodiments are only for illustrating the embodiments of the present invention and not for limiting the embodiments of the present invention, and those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention, so that all equivalent technical solutions also belong to the scope of the embodiments of the present invention, and the scope of patent protection of the embodiments of the present invention should be defined by the claims.

Claims

1. A rolling bearing failure diagnosis method characterized by comprising:

performing signal reconstruction aiming at a vibration signal of a rolling bearing to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band;

calculating the spectral kurtosis of each reconstructed sub-signal to determine the optimal sub-signal in each reconstructed sub-signal, and acquiring the sub-signal to be analyzed of the vibration signal based on the preset signal characteristic of the optimal sub-signal;

and analyzing the sub-signals to be analyzed by using a fault analysis model, and determining a fault diagnosis result of the rolling bearing.

2. The method according to claim 1, wherein the performing signal reconstruction for the vibration signal of the rolling bearing, and obtaining each reconstructed sub-signal corresponding to each divided frequency band for the vibration signal comprises:

obtaining an integer power of a preset base number closest to the fault frequency according to the fault frequency of the rolling bearing, and determining the number of division layers based on the integer power;

and performing frequency spectrum division on the vibration signal according to the number of the divided layers to obtain a plurality of divided frequency bands, and constructing a filter bank of each divided frequency band to perform filtering on the vibration signal of the rolling bearing to obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band.

3. The method of claim 2, wherein the step of obtaining the reconstructed sub-signals of the vibration signal corresponding to the divided frequency bands further comprises:

performing frequency band division on a bearing frequency band of the rolling bearing according to the division layer number, and determining each division frequency band;

defining an empirical scale function and an empirical wavelet function of the filter bank according to each divided frequency band so as to filter the vibration signal of the rolling bearing and obtain each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band;

the empirical scale function is defined as:

the empirical wavelet function is defined as:

wherein ω is a frequency, ω is _n For the frequency division of the nth divided frequency band, said τ _n And the beta is a coefficient function for the transition frequency band of the nth divided frequency band.

4. The method of any one of claims 1 to 3, wherein the calculating the spectral kurtosis of each reconstructed sub-signal to determine an optimal sub-signal of the reconstructed sub-signals comprises:

calculating the linear spectrum kurtosis of each reconstructed sub-signal, and determining the reconstructed sub-signal corresponding to the maximum one of the linear spectrum kurtosis as the optimal sub-signal;

wherein the step of calculating the linear spectral kurtosis of each reconstructed sub-signal comprises:

determining a reconstructed sub-signal as a current sub-signal;

determining a random variable sequence of the current sub-signal corresponding to each candidate phase based on a plurality of candidate phases in succession;

sequentially arranging the random variable sequence of each candidate stage according to the segment value of each candidate stage, and determining the candidate linear moment of any one candidate stage;

determining a first linear moment of a first preset stage and a second linear moment of a second preset stage according to the candidate linear moment of any one candidate stage;

determining a linear spectral kurtosis of the current sub-signal based on the first and second linear moments.

5. The method of claim 4, wherein the candidate linear moments for any one of the candidate phases are represented as:

wherein, said λ _r Is a candidate linear moment of the r-th candidate stage, the E [ X _r-k|r ]Is a random variable sequence X _r-k To a desired value of

6. The method according to claim 4, wherein the obtaining of the sub-signal to be analyzed of the vibration signal based on the preset signal characteristic of the optimal sub-signal comprises:

and according to the central frequency spectrum and the optimal band-pass of the optimal sub-signal, performing signal extraction on the vibration signal to determine a sub-signal to be analyzed of the vibration signal.

7. The method according to claim 1 or 6, wherein the analyzing the sub-signal to be analyzed by using the fault analysis model to determine the fault diagnosis result of the rolling bearing comprises:

extracting a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed;

and performing prediction on a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on a plurality of given fault categories by using the fault analysis model, and determining a fault category prediction result of the sub-signal to be analyzed.

8. The method of claim 7, wherein the fault analysis model comprises a random forest model having a plurality of decision trees;

wherein the determining, by using the fault analysis model, a fault category prediction result of the sub-signal to be analyzed by performing prediction on a plurality of frequency domain features and a plurality of time domain features of the sub-signal to be analyzed based on a given plurality of fault categories includes:

performing prediction on a plurality of frequency domain characteristics and a plurality of time domain characteristics of the sub-signal to be analyzed based on given fault types by utilizing each decision tree in the random forest model to obtain a preliminary fault type prediction result of each decision tree;

and counting the ticket number of each fault category according to the preliminary fault category prediction result of each decision tree, and determining the fault category with the largest ticket number as the final fault category prediction result of the sub-signal to be analyzed.

9. A method as claimed in claim 1 or 8, wherein the random forest model is trained by:

generating each decision tree in the random forest model by using the training set of the random forest model until a preset growth stopping condition is met, and acquiring the random forest model based on each generated decision tree;

performing verification aiming at the random forest model by using a verification set of the random forest model to obtain a loss training loss function;

and optimizing the random forest model based on the training loss function, returning to execute the training set using the random forest model, and generating each decision tree in the random forest model until the training loss function meets a preset convergence condition so as to obtain the trained random forest model.

10. A method as claimed in claim 9, wherein the generating, using the training set of random forest models, decision trees in the random forest models until a preset stop growing condition is met comprises:

a step of determining a kini coefficient, which is to take all the features which are not determined as the optimal features in the training set as candidate features and calculate the kini coefficient of the training set according to each segmentation feature of each candidate feature;

a segmentation determining step, namely determining the candidate feature with the minimum kini coefficient and the segmentation feature of the candidate feature as an optimal feature and an optimal segmentation point respectively according to each kini coefficient, dividing the training set into a first subset and a second subset based on the optimal feature, the optimal segmentation point and each candidate feature in the training set, and distributing the first subset and the second subset to two newly generated child nodes in a decision tree;

and recursively calling the step of determining the kini coefficients and the step of determining the segmentation for each of the two child nodes until a preset growth stopping condition is met.

11. The method according to claim 9, wherein the preset growth stop condition comprises any one of the following conditions:

the number of training samples corresponding to the candidate features currently contained in the training set is less than a preset sample number threshold;

the Gini coefficient of the training set is smaller than a preset coefficient threshold;

the number of candidate features currently contained in the training set does not satisfy the number of features required for generating a new node.

12. A rolling bearing failure diagnosis device characterized by comprising:

the signal extraction module is used for executing signal reconstruction aiming at a vibration signal of a rolling bearing, acquiring each reconstructed sub-signal of the vibration signal corresponding to each divided frequency band, calculating the spectral kurtosis of each reconstructed sub-signal to determine the optimal sub-signal in each reconstructed sub-signal, and acquiring a sub-signal to be analyzed of the vibration signal based on the preset signal characteristic of the optimal sub-signal;

and the fault analysis module is used for analyzing the sub-signal to be analyzed by using a fault analysis model and determining a fault diagnosis result of the rolling bearing.

13. An electronic device, comprising: the system comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete mutual communication through the communication bus;

the memory is used for storing at least one executable instruction, and the executable instruction causes the processor to execute the operation corresponding to the rolling bearing fault diagnosis method according to any one of claims 1-11.

14. A computer storage medium characterized by having stored thereon a computer program which, when executed by a processor, can implement the rolling bearing fault diagnosis method according to any one of claims 1 to 11.