CN113203565A

CN113203565A - Bearing fault identification method and system based on EEMD sparse decomposition

Info

Publication number: CN113203565A
Application number: CN202110316920.4A
Authority: CN
Inventors: 张健; 宋文广; 雷鸣; 林德树
Original assignee: Yangtze University
Current assignee: Yangtze University
Priority date: 2021-03-25
Filing date: 2021-03-25
Publication date: 2021-08-03

Abstract

The invention relates to a bearing fault identification method and a system based on EEMD sparse decomposition, firstly, a bearing vibration signal is obtained and converted into an electric signal; decomposing the electric signal by using an EMD method to obtain a plurality of IMF components and a remainder; performing sparse processing on the plurality of IMF components; and calculating the energy entropy of the vibration signal of the bearing by utilizing the IMF component after the ensemble averaging by the EEMD method, and identifying the fault type of the bearing by combining the energy distribution condition. The method is used as an effective self-adaptive algorithm to decompose particularly aiming at non-stationary signal processing, and for respective independent IMF components, the IMF components have the scale characteristics of signals and have the characteristic of changing along with the signals.

Description

Bearing fault identification method and system based on EEMD sparse decomposition

Technical Field

The invention relates to the field of signal and information processing, in particular to a bearing fault identification method and system based on EEMD sparse decomposition.

Background

Under the conditions of high load, strong high temperature, strong high pressure, high humidity, strong electromagnetic interference and strong coupling, the large-scale bearing which runs for a long time under the severe working condition environment is often in the condition of material aging, high temperature and high pressure, sudden loading action and design defects in the running process, particularly abnormal vibration factors can generate irreversible damage accumulation on the bearing. Meanwhile, the normal operation of the whole bearing is seriously affected by faults, and most of the faults are caused by abnormal operation of some key parts, so that permanent damage is caused to the bearing, and huge economic loss is brought. How to ensure the stable, reliable and safe operation of the bearing becomes a hotspot problem which needs to be concerned when the bearing is operated and maintained.

Since vibration signals directly represent the dynamic behavior of a faulty bearing, which leads to their susceptibility to faults, they are widely used to detect bearing faults. At present, fault diagnosis of large mechanical bearings mainly focuses on collecting vibration state parameters of equipment operation so as to realize state monitoring and fault diagnosis. Mechanical faults are researched and judged by analyzing the change situation of various parameters of the mechanical vibration signals, such as the change situation of a time domain, a frequency domain and an amplitude domain, so that early warning is implemented. Meanwhile, the extraction method of fault features has been developed from conventional methods (such as fast fourier transform, power spectrum estimation, time-frequency analysis and axial locus) to a higher hierarchy direction (angular domain analysis, holographic spectrum, fractal dimension, etc.). Such as: the holographic spectrum theory which combines multiple information such as amplitude, frequency, phase, angular momentum and the like of mechanical vibration together for research and judgment has great significance for improving the diagnosis level of mechanical faults.

Disclosure of Invention

The invention provides a bearing fault identification method and system based on EEMD sparse decomposition, wherein signals are processed according to the rule that the frequency of the signals changes along with time, the EEMD sparse decomposition method is used as an effective self-adaptive algorithm to decompose particularly aiming at non-stationary signal processing, and for IMF components which are independent of each other, the IMF components have the scale characteristics of the signals and the characteristics of changing along with the IMF components. In order to further clarify the difference between the fault and the non-fault event, the important research oriented to the independent IMF component is developed, then the energy distribution of the fault vibration signal in different frequency bands is analyzed according to the characteristic distribution of the IMF, and then the energy distribution is compared with the energy distribution in the normal state. Meanwhile, the type of the fault behavior is judged by means of a judgment method of energy entropy.

The technical scheme for solving the technical problems is as follows:

in a first aspect, the invention provides a bearing fault vibration signal feature extraction and identification method, which comprises the following steps:

step 100, acquiring a bearing vibration signal and converting the bearing vibration signal into an electric signal;

step 200, adding Gaussian white noise with a certain amplitude into the electric signal, and performing EMD (empirical mode decomposition) on the electric signal added with the white noise to obtain a plurality of IMF (intrinsic mode function) components and a remainder;

step 300, performing sparse processing on a plurality of IMF components;

step 400, performing iteration of preset times on the step 200 and the step 300, wherein the added white gaussian noise is different white gaussian noise with the same root mean square at each iteration, and then performing overall average calculation on the same-order IMF component in the iteration result;

and 500, calculating the energy entropy of the vibration signal of the bearing by using the IMF component after the ensemble averaging, and identifying the fault type of the bearing by combining the energy distribution condition.

In a second aspect, the present invention provides a bearing fault identification system based on EEMD sparse decomposition, which is characterized by comprising:

the preprocessing module is used for acquiring a bearing vibration signal and converting the bearing vibration signal into an electric signal;

the EEMD sparse decomposition module is used for adding Gaussian white noise with a certain amplitude into the electric signal and performing EMD decomposition on the electric signal added with the white noise to obtain a plurality of IMF components and a remainder; performing sparse processing on the plurality of IMF components; performing iteration for a preset number of times, wherein the added Gaussian white noise is different Gaussian white noises with equal root-mean-square in each iteration, and then performing overall average calculation on IMF components of the same order in an iteration result;

and the classification identification module is used for calculating the energy entropy of the bearing vibration signal by utilizing the IMF component after the ensemble averaging and identifying the fault type of the bearing by combining the energy distribution condition.

In a third aspect, the present invention provides an electronic device comprising:

a memory for storing a computer software program;

and the processor is used for reading and executing the computer software program stored in the memory so as to realize the bearing fault identification method based on EEMD sparse decomposition in the first aspect of the invention.

In a fourth aspect, the present invention provides a computer-readable storage medium, wherein a computer software program for implementing the bearing fault identification method based on EEMD sparse decomposition according to any one of the first aspect of the present invention is stored in the storage medium.

The invention has the beneficial effects that: after the vibration signal is obtained, EMD decomposition is carried out, then IMF components obtained through EMD decomposition are subjected to sparse processing, a sparse processing method in image processing is used for reference, a plurality of IMF components obtained after EMD decomposition are subjected to fine processing, fault characteristics contained in the IMF components are used, then overall averaging is carried out, the problem is well avoided by utilizing the characteristic that the frequency of white noise is uniformly distributed after the white noise is added, and meanwhile, the real attribute of time frequency is obtained. The added white noise component can also realize complete noise reduction under the operation of solving the overall mean value by the same-order IMF. However, the actual decomposition effect and the increment of the iteration turns do not have a linear relation, so the selection of the iteration number should be specifically selected according to the experimental result. Compared with the similar method, the method can better realize the denoising and the refining processing aiming at the vibration signal.

Drawings

FIG. 1 is a flow chart of a method provided by an embodiment of the present invention;

FIG. 2 is a time domain and frequency domain diagram of a bearing fault vibration signal at a rotation speed of 1200r/min according to an embodiment of the present invention.

Detailed Description

The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.

Fig. 1 is a method for extracting and identifying a bearing fault vibration signal feature provided in an embodiment of the present invention, including the following steps:

step 100, obtaining a bearing vibration signal and converting the bearing vibration signal into an electric signal y (t).

Step 200, adding Gaussian white noise with a certain amplitude into the electric signal after the sparse processing, and performing EMD on the electric signal added with the white noise to obtain a plurality of IMF components and a remainder.

And step 300, performing sparse processing on the plurality of IMF components.

The sparse processing process utilizes a high-amplitude, low-amplitude dictionary { A }_h,A_lAnd (d) reconstructing the electric signal y (t) to obtain a reconstructed signal x (t). Specifically, the method comprises the following steps:

the electrical signal y (t) is restored to low amplitude samples y with the same size as the target high amplitude component using a cubic interpolation process_l(t)；

Using S high-pass filters on low-amplitude samples y_l(t) performing filtering processing;

high-pass filtered y_l(t) decomposition into dimensions of

To obtain corresponding high amplitude features y_kAnd calculating a sparse coefficient:

sparse coefficient { alpha) is obtained based on OMP method_k}；

Using high-amplitude dictionaries { A_hAnd a sparse coefficient { alpha }_kGet the high amplitude component block x_k＝A_hα_k(ii) a Splicing the obtained high-amplitude component blocks, and carrying out average value processing on the overlapped part;

combined with interpolated low-amplitude component y_l(t), reconstructing the high amplitude component:

in the formula, R_kRepresenting component blocksAnd (5) extracting a matrix from the features.

In the step, by taking a sparse processing method in image processing as a reference, a plurality of IMF components obtained after EMD decomposition are refined, so as to obtain fault characteristics contained in the IMF components.

Before sparse processing is carried out, an ultra-complete dictionary A needs to be constructed, and the dictionary learning process is as follows:

selecting IMF component with larger amplitude from IMF component set

(J ═ 1,2,3,4,5 …, J); obtaining IMF component with smaller amplitude by means of fuzzy and down sampling

Then based on cubic interpolation process

Reconstructing, namely performing interpolation processing on the IMF component with small amplitude with the same size as the IMF component with large amplitude to obtain the IMF component with small amplitude with the same size as the IMF component with large amplitude

The three differences are piecewise interpolation, and compared with cubic spline interpolation and other modes, the value of the interpolation polynomial at the node is equal to the value of the interpolation function at the node, and cubic spline interpolation also needs to know the derivative of the interpolation polynomial at some nodes, which is difficult to realize in practical application. The process of cubic interpolation is as follows:

H(x)＝α_j(x)f_j+α_j+1(x)f_j+1+β_j(x)f′_j+β_j+1(x)f′_j+1

the basis function value of the cubic interpolation is as follows:

the preprocessing process needs to filter information of a low-frequency part in a high-amplitude component set of the IMF, and ensures that a dictionary has sufficient expression on feature information of the part, and the preprocessing process can be performed in a differential recording manner as follows:

for the low-amplitude component set of the IMF, the preprocessing process needs to process the features of the high-frequency part in the set, and here, S groups of high-pass filters are used to extract the high-frequency features. The high frequency characteristic can be expressed as

By this preprocessing, the high and low amplitude components of the IMF can be decomposed into overlapping dimensions of

Then obtaining K component blocks randomly, and constructing IMF high-amplitude component and low-amplitude component samples

(1,2,...,K)。

IMF-based high-amplitude component and low-amplitude component samples

And the resulting training dictionary { A_h,A_lAnd (4) combining a pretreatment process:

first given the form

Then the dictionary training process of the IMF high-amplitude component is as follows:

similarly, the dictionary training process for IMF low amplitude components is:

combining the above two dictionary training processes, adopt

And

as dictionary training weights, the IMF high and low amplitude component dictionary training process can be expressed as:

the above formula can be simplified as:

parameters in the formula:

solving equation (8) directly results in a dictionary a solution with extremely high degrees of freedom and increases the requirement for the number of samples of dictionary a, which increases the computational effort for sample reconstruction. In order to reduce the degree of freedom of the dictionary A training process, the adopted method is to determine the dictionary A belongs to R^(S+1)n*mSparse representation is performed. If A ═ A₀Z，Z∈R^m0*mRepresents a sparse matrix, where A₀∈R^(S+1)n*m0For the base dictionary, the non-zero number of the elements of the matrix column needs to be less than T1, i.e. it is assumed that all atoms in the dictionary can be selected as A₀Based on the above assumptions, the dictionary training process described in equation (7) can be expressed as:

in the formula, Z_jThe j columns of elements of matrix Z are matched, and then the process of optimally solving problem (9) can be performed in two steps: (1) fixing a sparse matrix Z, and updating the sparse representation of the coefficient matrix Q; (2) and fixing the sparse matrix Q and updating the sparse matrix Z. First, fix the sparse matrix Z, then the optimization process of equation (9) is:

the OMP method is selected to solve the optimization problem of the formula (10). Secondly, determining a sparse coefficient Q, wherein the optimization process of the formula (9) is as follows:

is provided with

In the jth column of the sparse matrix Q, there are:

in the formula, the first and second sets of data are represented,

the optimized update process of equation (11) can be expressed as:

if it is satisfied with

Then:

in the formula, Tr (-) is a trajectory expression of the matrix. Due to the fact that

Value and element z_jIs irrelevant, therefore, is satisfying

In this case, the optimization process of equation (13) may be modified as follows:

equation (14) is a sparse coding form that can be solved based on the OMP method. Based on the two steps, a sparse matrix Z can be obtained, and then dictionaries { A ] with high amplitude and low amplitude are obtained_h,A_l}。

And step 400, performing preset times of iteration on the step 200 and the step 300, wherein the added white gaussian noise is different white gaussian noise with the same root mean square at each iteration, and then performing overall average calculation on the same-order IMF components in the iteration result.

The iterative and ensemble averaging computation processes of

steps

200 and 300 constitute the EEMD sparse decomposition method,

the EEMD sparse method is adopted as an effective adaptive algorithm to decompose particularly aiming at the processing of non-stationary signals. For each component of the IMF that is independent of the other, it has a scale characteristic of the signal and a characteristic that varies with it.

The problem of mode aliasing can be shown when the traditional empirical mode decomposition method is applied, and the problem is well avoided by utilizing the characteristic that the frequency of the white noise is uniformly distributed after the white noise is added, and the real attribute of the time frequency is obtained at the same time.

The added white noise component can also realize complete noise reduction under the operation of solving the overall mean value by the same-order IMF. However, the actual decomposition effect and the increment of the iteration turns do not have a linear relation, so the selection of the iteration number should be specifically selected according to the experimental result. Compared with the similar method, the method can better realize the denoising and the refining processing aiming at the vibration signal.

The front-end sensor array has different frequencies of the detected vibration signals, and the propagation angles and directions of the received vibration signals have randomness and uncertainty.

In general terms: the vibration wave frequencies collected by all the vibration sensors are compared each time, and more than 3 vibration signal quantities with larger amplitude and lower frequency are selected from the vibration wave frequencies to be processed. The measured vibration wave frequency reflected to each sensor is different; the dominant frequency plays a crucial role in signal analysis. Therefore, the principle of selecting the main vibration frequency and other component frequencies from the selected plurality of vibration waves is as follows, and the vibration wave with the highest amplitude from the plurality of vibration waves measured in the sensor is used as the main vibration frequency, and the others are used as the components.

Step 500, calculating the energy entropy of the bearing vibration signal by utilizing the IMF component after the ensemble averaging, and applying C_kEach characterizing the energy of the IMF component, C ═ C_kK ∈ R } will be the energy distribution of the signal, expressed by

Calculating an energy entropy value, wherein V_k＝C_kand/C. And then identifying the type of the bearing fault by combining the energy distribution condition.

Under the normal working condition operation of the bearing, the representation of vibration signal energy distribution is uniform and has small fluctuation, if a fault condition occurs, resonance can be detected in a corresponding frequency band, meanwhile, energy can be converged in the frequency band range, and the energy entropy value can also change accordingly.

By calculating the energy entropy value and comparing the energy entropy value with data in the rolling bearing fault energy entropy experience database, the fault type can be preliminarily judged and early warning can be timely realized by integrating the energy distribution condition.

The characteristics of the vibration signal of the bearing under fault conditions are analyzed below. When the motor rotating speed is 1200r/min, the frequency domain waveform of the fault vibration signal acquired by the FBG acceleration sensor is shown in FIG. 2.

After the noise reduction and conditioning processes are performed on the signal, 9 IMF components and one survivor component appear when EEMD sparse decomposition is performed. The EEMD sparse method has the advantages that attenuation caused by the end effect on signal energy is reduced, typical IMF characteristic components are selected, and energy distribution and entropy values of the typical IMF characteristic components are calculated to achieve the purpose.

The IMF component simultaneously carries the characteristics of original signal locality and time-varying scalability, and is obtained through theoretical analysis: the energy change conditions under different conditions can be mastered by comparing the energy entropy so as to judge the fault event and the normal event.

Category 6 fault conditions were selected and their energy entropy values were calculated as shown in table 1.

TABLE 16 energy entropy distribution of class failure modes

As can be seen from Table 1, the energy entropy values of the working parts of the bearing are maximum and stable under normal working conditions, the energy entropy values are continuously decreased along with the continuous deterioration of the fault, and the energy entropy values are sequentially arranged into an inner ring, an outer ring and a rolling shaft according to the magnitude sequence of the energy entropy values under the same type of fault conditions. According to the rule analysis, the energy entropy value under various fault conditions can be used as the basis for fault diagnosis.

For fault identification of the bearing according to the energy entropy mode, the fault type of the bearing can be preliminarily judged by comparing the fault identification with a threshold value in an energy entropy sub-database in a rolling bearing standard database set of the university of Kaiser-Sichu in the United states.

Rolling bearing data of the university of kas storage, usa, which is a global and widely-focused and referred to as a fault diagnosis standard database, is widely used domestically for reference, and signals are acquired through a 16-channel DAT recorder and are subjected to post-processing in MATLAB.

The comparison result of the energy entropy value obtained by selecting a plurality of typical fault characteristics and calculating with the database is shown in the table 3-2.

The results of table 2 indicate that the energy entropy determination method can be effectively used as a feature value in the determination of a typical failure of a rolling bearing.

Table 2 fault type database comparison table

Experimental verification

The experimental study mainly relates to EEMD sparse feature extraction and energy entropy calculation of a fault vibration signal in the rotation process of a bearing, and further preliminarily judging the fault category. And respectively calculating the energy distribution and the energy entropy condition of each fault signal by using an energy entropy method under different fault states and comparing the energy distribution and the energy entropy condition. Three groups of fault characteristic analysis and comparison are carried out in the experiment. As follows:

(1) and calculating the energy distribution and the entropy of the scratch damage of the inner ring, the outer ring and the rolling shaft, and respectively comparing and analyzing the energy distribution and the entropy with the energy entropy under the normal running condition of each part.

(2) And calculating the energy distribution and entropy of the abrasion damage of the rotating shaft, the belt and the ball, and respectively comparing and analyzing the energy distribution and entropy with the energy entropy under the normal operation condition of each part.

(3) And calculating the energy distribution and entropy of the abrasion damage of the driving end, the fan and the gear, and respectively comparing and analyzing the energy distribution and entropy with the energy entropy under the normal running condition of each part.

And finally, comparing the fault data with various fault data in a rolling bearing typical energy entropy value database of the university of western medicine storage in America, and preliminarily judging and determining the fault type.

The processing steps of the abnormal vibration signal of the bearing are as follows:

step 1: and carrying out noise filtering processing on the input signal.

Step 2: EEMD sparse decomposition is carried out on the fault vibration signal subjected to noise reduction and conditioning, and different IMF components and a residual value are obtained; analyzing the energy distribution state of each IMF component and calculating the energy entropy value of each IMF component;

and step 3: and comparing the calculated energy entropy value with threshold data in a vibration fault energy entropy database in the standard bearing database of the university of western medicine storage, and preliminarily judging the fault type of the vibration fault.

The EEMD sparse decomposition method can avoid the interference of the end effect on signal energy as much as possible, select limited IMF components for analysis, calculate the energy distribution and entropy value of the limited IMF components, and discard the residual IMF components. Here, the first seven IMF components are each time selected as features and their energy states are calculated.

Firstly, the scratch fault phenomena of the inner ring, the outer ring and the roller are selected as examples and evaluated by an energy entropy calculation mode, and the energy distribution and the entropy distribution of each state are shown in tables 3 and 4.

TABLE 3 energy distribution of first class example failure modes (inner, outer, roller)

The selected outer ring, roller and inner ring failure phenomena are calculated and evaluated by using an energy entropy example mode as shown in table 4.

TABLE 4 energy entropy distribution of first class failure modes (inner, outer, roller)

From table 3 it can be seen that the energy of the vibration signal is mainly concentrated in the first five components, and the energy distribution is different for each type of fault.

The following conclusions are reached from tables 3 and 4:

(1) the first five components mainly gather the main energy of the signal, and the energy distribution under each fault state is different.

(2) The energy entropy value of the bearing under the normal operation condition is maximum, the fault is more serious along with the continuous deterioration of the operation environment and the condition, and the corresponding energy entropy value is in a gradually descending trend.

(3) The energy entropy values of the same fault characteristics of the inner ring, the outer ring and the roller are in a changing trend from big to small.

Subsequently, the second classification (spindle, belt, ball) wear failure phenomenon was selected as an example and evaluated by means of energy entropy calculation, and the energy distribution and entropy values of each state are shown in tables 5 and 6 below.

TABLE 5 second classification (spindle, belt, ball) energy distribution for example failure modes

TABLE 6 energy entropy distribution of the second class failure modes (spindle, belt, ball)

It can be seen from table 5 that the energy of the vibration signal is mainly concentrated in the first four components, and the energy distribution is different for each type of fault.

The following conclusions are drawn from tables 5, 6:

(1) the first four components mainly gather the main energy of the signal, and the energy distribution under each fault state is different.

(2) The energy entropy value of the bearing under the normal operation condition is the largest, the fault is more serious along with the continuous deterioration of the operation environment and the condition, and the corresponding energy entropy value is in a gradually descending trend.

(3) The energy entropy values of the same fault characteristics of the rotating shaft, the belt and the ball are in a gradually descending trend.

Finally, the fault phenomena of the third classification (driving end, fan and gear) are selected as an example and are evaluated by using an energy entropy calculation mode, and the energy distribution and entropy value of each state are shown in the following tables 7 and 8.

TABLE 7 energy distribution of third class example failure modes (drive end, fan, gears)

TABLE 8 energy entropy of third class example failure modes (drive end, fan, gears)

It can be seen from table 7 that the energy of the vibration signal is mainly concentrated in the first three components. The energy distribution under various fault conditions is different.

The following conclusions are drawn from tables 7, 8:

(1) the first three components mainly converge the main energy of the signal, and the energy distribution under each fault state is different.

(2) The energy entropy value under the normal operation condition is the biggest, and the trouble is more serious along with the continuous deterioration of operational environment and condition, and the energy entropy value that corresponds is the decline trend gradually.

(3) The energy entropy values of the same fault characteristics of the driving end, the fan and the gear are also in a gradually descending trend.

Here, the energy distribution and the energy entropy value of the three types of fault phenomena are calculated and compared to obtain: the scheme of predicting the bearing fault phenomenon by using the energy entropy as the characteristic value is completely feasible.

The rolling bearing fault type preliminary judgment method based on the energy entropy value can realize preliminary judgment of the rolling bearing fault type based on the energy entropy value by comparing the rolling bearing fault preliminary judgment with a threshold value in an energy entropy sub-database in a rolling bearing standard database set of the university of western storage in America.

The rolling bearing data of the university of Kaiser storage, USA, as a database widely adopted globally and leading to a fault diagnosis standard, is widely applied domestically, and vibration signals of the rolling bearing data are acquired through a DAT recorder with 16 channels and are subjected to post-processing in an MATLAB environment. The sampling frequency of the digital signal is 12000S/S, and each fault data of the bearing is collected at the sampling rate of 48000S/S. The ultra-high sampling rate can completely guarantee the situation capture for the vibration signal change.

The energy entropy values calculated by the part of typical fault phenomena are selected and input into a database for query, and the judgment result is shown in table 9.

Table 9 fault type database comparison table

The result shows that the input energy entropy value of each type of typical fault is basically consistent with the energy entropy threshold range of the corresponding fault type in the standard database, and the energy entropy value judgment method can be used as a characteristic value to be effectively used for judging the typical fault of the rolling bearing.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A bearing fault identification method based on EEMD sparse decomposition is characterized by comprising the following steps:

step 300, performing sparse processing on a plurality of IMF components;

2. The method as claimed in claim 1, wherein the bearing vibration signal is a low frequency vibration signal not higher than 1000 Hz.

3. The method for identifying the bearing fault based on EEMD sparse decomposition as claimed in claim 1, wherein the sparse processing of the IMF components comprises:

high-pass filtered y_l(t) decomposition into dimensions of

sparse coefficient { alpha) is obtained based on OMP method_k}；

in the formula, R_kA feature extraction matrix representing the component blocks.

4. The method for identifying the bearing fault based on the EEMD sparse decomposition as recited in claim 1, wherein in the step 400, the number of iterations is 100-400.

5. The method as claimed in claim 1, wherein the bearing fault type is identified according to the energy distribution condition, and the identification method comprises a classification identification method based on SVM.

6. The method for identifying the bearing fault based on the EEMD sparse decomposition as recited in claim 1, wherein a dictionary adopted when the electric signal is subjected to the sparse processing is obtained through a K-SVD dictionary learning algorithm.

7. A bearing fault identification system based on EEMD sparse decomposition, comprising:

8. An electronic device, comprising:

a memory for storing a computer software program;

a processor for reading and executing the computer software program stored in the memory to implement a bearing fault identification method based on EEMD sparse decomposition as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that the storage medium has stored therein a computer software program for implementing a bearing fault identification method based on EEMD sparse decomposition as claimed in any one of claims 1 to 6.