Rolling bearing fault diagnosis method and system based on controllable sparsity morphological decomposition

Abstract

A rolling bearing fault diagnosis method and system based on controllable sparsity morphological decomposition, wherein the bearing vibration signal and speed signal are collected, and the vibration signal segment under stable working conditions is selected according to the speed signal; based on the frequency domain sparsity of discrete frequency components, an index for adaptively controlling the sparsity of the Fourier dictionary is constructed; based on the shift invariance of the fault pulse, a cyclic dictionary learning algorithm for determining the sparsity is designed; based on the discrete frequency components and the fault pulse distinguishing characteristics, a morphological decomposition algorithm model with controllable sparsity is established; according to the constraints of adaptively controlling the sparsity, a greedy algorithm is used to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model to separate the discrete frequencies; according to the constraints of determining the sparsity, the frequency domain least squares method and the greedy algorithm are used alternately to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively; the reconstructed fault pulse is subjected to envelope spectrum analysis, compared with the characteristic frequency of the bearing fault, the CI index is calculated, and the diagnosis is completed.

Description

Technical Field

[0001] The present invention relates to the technical field of rolling bearing faults, and in particular to a rolling bearing fault diagnosis method and system based on controllable sparsity morphological decomposition. Background Art

[0002] In rolling bearing fault diagnosis, due to the increasing complexity of mechanical systems, multiple transmission components, strong background noise interference, etc., the vibration signal collected by the sensor is coupled by fault pulses, discrete frequency components and environmental noise, and the fault characteristics of interest are difficult to observe. Generalized morphological component decomposition is an effective sparse signal decoupling method. The algorithm uses different dictionaries to sparsely represent signal components according to the morphological differences of signal components, and has been widely used in various scenarios. However, the dictionary sparsity and sparsity parameters of morphological component decomposition based on greedy algorithms and morphological component decomposition based on convex optimization theory are often determined empirically. Therefore, it is necessary to propose a morphological component decomposition method that can control the sparsity and reliably extract effective fault components.

[0003] The above information disclosed in this Background section is only for enhancement of understanding of the background of the invention and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the invention

[0004] The present invention provides a rolling bearing fault diagnosis method and system based on controllable sparsity morphological decomposition, which can control the dictionary sparsity of the morphological decomposition method, realize adaptive vibration signal decoupling, and complete rolling bearing fault diagnosis.

[0005] A rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, comprising:

[0006] S100, collecting the vibration signal and the speed signal of the rolling bearing, and selecting the vibration signal under the stable working condition with a sample number of N according to the speed signal , When a bearing fails, the vibration signal Contains fault pulses with periodic shift invariant characteristics , and discrete frequency components with frequency domain sparsity , , discrete frequency components Including periodic signals caused by gear or shaft excitation;

[0007] S200, based on discrete frequency components The frequency domain sparsity of the Fourier dictionary is adaptively controlled by constructing the index constraint for evaluating the sparsity.

[0008] S300, based on fault pulse Based on the shift invariance of the fault cycle, a cyclic dictionary is constructed with the failure cycle as the atomic length and a cyclic dictionary learning algorithm is designed to determine the sparsity.

[0009] S400, based on discrete frequency components and fault pulse Based on the distinguishing characteristics of the morphological decomposition algorithm, a morphological decomposition algorithm model with controllable sparsity is established;

[0010] S500, according to the constraint of adaptively controlling sparsity, using a greedy algorithm to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model and separate the discrete frequency components ;

[0011] S600, according to the constraint of determining sparsity, alternately use the frequency domain least square method and the greedy algorithm to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively, and realize the fault pulse extract;

[0012] S700, for the extracted fault pulse Perform envelope spectrum analysis, compare it with the characteristic frequency of bearing fault, calculate CI index, and complete fault diagnosis. CI index is the characteristic fundamental frequency amplitude FFA and characteristic energy ratio FER.

[0013] Preferably, in S200, according to the discrete frequency components The frequency domain sparsity of , using an explicit Fourier dictionary , and greedy algorithm to reconstruct discrete frequency components :

[0014] ;

[0015] The data matrix , From the vibration signal By length No overlap Segment structure, , is the sparse representation matrix of the Fourier dictionary, which adaptively controls the sparsity constraint of the Fourier dictionary It is composed of the following indicators that reflect the changes in frequency components:

[0016] 1) Based on the maximum difference to mean ratio MR:

[0017] ;

[0018] Constraints on the adaptive control sparsity of MR for:

[0019] ;

[0020] 2) Based on spectrum flatness SF:

[0021] ;

[0022] Constraints for adaptive control of sparsity of SF for:

[0023] ;

[0024] 3) Based on the peak factor PP:

[0025] ;

[0026] Constraints on adaptive control sparsity of PP composition for:

[0027] ,

[0028] in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, represents the mean operation; To reconstruct the spectrum of discrete frequency components The first The power spectrum of the interval, ; Indicates the maximum value calculation, Indicates minimum value calculation; is the reconstructed discrete frequency component Length.

[0029] Preferably, in S300, the atomic length of the loop dictionary design is ,in,

[0030] ,

[0031] In the formula is the sampling frequency, is the fault characteristic frequency, Indicates rounding up, according to the fault pulse The shift invariance of Vector Circular shift to construct a circular dictionary , :

[0032] ,

[0033] in,

[0034] , is the length of the pulse of interest and is a positive integer,

[0035] Vector is a circulatory vector, and ,in,

[0036] ,

[0037] Among them, vector Including multiple 0s, the number of 0s is equal to and The difference between

[0038] Further solving the circular dictionary And its sparse representation:

[0039] ,

[0040] The data matrix , The vibration signal Divide Each segment is and are constructed without overlapping; , is a sparse representation matrix; is the sparsity constraint of the cyclic dictionary, set to be deterministic ,

[0041] The solution steps include:

[0042] S301, first use the unitary Fourier dictionary Convert the loop dictionary to frequency domain solution:

[0043] ,

[0044] in, and ; , is the data matrix The result after Fourier transformation; , is a sparse representation matrix The result after Fourier transformation is

[0045] S302. Decompose the least squares problem into Sub-questions:

[0046] ,

[0047] in, yes No. elements, yes No. List, yes No. List,

[0048] S303, Solution:

[0049] ,

[0050] S304, before loop vector Solve for values:

[0051] ,

[0052] in, , Represents the front of the loop vector elements, is the inverse Fourier transform,

[0053] S305. Solving sparse representation by greedy algorithm and sparse constraints .

[0054] Preferably, in S400,

[0055] Establish a morphological decomposition algorithm model with controllable sparsity:

[0056] ,

[0057] Sparsity Constraints Different situations make the morphological decomposition algorithm model include:

[0058] 1) Model based on MR adaptive control sparsity constraint:

[0059] ,

[0060] 2) Model based on SF adaptive control sparsity constraint:

[0061] ,

[0062] 3) Model based on SF adaptive control sparsity constraint:

[0063] ,

[0064] in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, Represents the mean operation.

[0065] Preferably, in S400, in the morphological decomposition algorithm model with controllable sparsity, the Fourier dictionary The atomic length Set to:

[0066] ,

[0067] in, Yes Select data The length of An integer multiple of .

[0068] Preferably, in S500, the discrete frequency components are first separated, including:

[0069] S501, firstly By length No overlap Segment, construct data matrix , , construct an explicit Fourier dictionary , , separating the discrete frequency components is equivalent to solving the subproblem:

[0070] ,

[0071] Initialize the residual matrix , , initialize the support set ,initialization Satisfy the constraints and set the upper limit of sparsity to ;

[0072] S502, select explicit Fourier dictionary , The residual , Most relevant atoms:

[0073] ,

[0074] in, , is a dictionary matrix No. List,

[0075] Add the newly selected index to the backing set middle:

[0076] ,

[0077] S503, using the least squares method to update the sparse representation matrix , :

[0078] ,

[0079] in, is the support set The corresponding set of atoms, is the current support set Compute the sparse representation matrix;

[0080] Update the residual matrix:

[0081] ,

[0082] Update iteration count ,until ;

[0083] S504. Determine the constraints of sparsity Which of the following is met:

[0084] 1) Constraints of adaptive control sparsity based on MR:

[0085] ,

[0086] 2) Adaptive control of sparsity constraints based on SF:

[0087] ,

[0088] 3) Constraints of adaptive control sparsity based on PP:

[0089] ,

[0090] in, , Represents the coefficient of the current sparsity solution, , Represents the coefficient of the previous sparsity solution;

[0091] If the constraints are not met, keep , separate discrete frequency components from the data :

[0092] ,

[0093] in, , is the residual signal after separating the discrete frequency components.

[0094] Otherwise, the sparsity , return to step S502.

[0095] Preferably, in S600, extracting the pulse includes the steps of:

[0096] S601, the residual signal By length No overlap Segment, construct data matrix , ; Random selection A column in is used as a cyclic vector to construct a cyclic dictionary , randomly initialized , , extracting the fault pulse is equivalent to solving the subproblem:

[0097] ,

[0098] right and Perform Fourier transform to get , and , , initialize the residual matrix , , initialize the support set ;

[0099] S602: Calculate the elements of the dictionary :

[0100] ,

[0101] in, yes No. elements, yes No. OK, yes No. OK;

[0102] Calculate the front of the loop vector Values:

[0103] ,

[0104] Get the loop vector And construct the loop dictionary ;

[0105] S603: Select a circular dictionary based on the determined sparsity constraint Zhongyu Most relevant atoms:

[0106] ,

[0107] in, , is a dictionary matrix No. List,

[0108] Add the newly selected index to the backing set middle:

[0109] ,

[0110] Update the sparse representation matrix using the least squares method , :

[0111] ,

[0112] in, is the support set The corresponding set of atoms;

[0113] S604, reconstruct the current loop dictionary and the sparse representation matrix Fault pulse , :

[0114] ,

[0115] Reconstruct the loop dictionary from the previous iteration , and the sparse representation matrix , Fault pulse , :

[0116] ,

[0117] in, , is the circular dictionary obtained by the previous iteration. , is the sparse representation matrix obtained by the previous iteration;

[0118] S605, Calculation and The root mean square error when Or when the number of iterations Nit is reached, the reconstructed pulse is obtained , , otherwise, return to step S602.

[0119] In addition, the present invention also discloses a rolling bearing fault diagnosis system based on controllable sparsity morphological decomposition, comprising:

[0120] The acquisition unit is used to collect the vibration signal and speed signal of the rolling bearing, and select the vibration signal segment under stable working conditions according to the speed signal , ;

[0121] The construction unit is used to generate a discrete frequency component based on The frequency domain sparsity of , construct an index constraint to evaluate sparsity, so as to adaptively control the sparsity of Fourier dictionary; based on fault pulse The shift invariance of , construct a cyclic dictionary and design a cyclic dictionary learning algorithm to determine the sparsity; based on discrete frequency components and fault pulse Based on the distinguishing characteristics of the morphological decomposition algorithm, a morphological decomposition algorithm model with controllable sparsity is established;

[0122] The extraction unit is used to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model using a greedy algorithm according to the constraints of adaptively controlling the sparsity, and to separate the discrete frequency components According to the constraints of determining sparsity, the frequency domain least squares method and greedy algorithm are used alternately to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively to realize the fault pulse extract;

[0123] Diagnostic unit, which is used to extract fault pulses Perform envelope spectrum analysis, compare it with the characteristic frequency of bearing fault, calculate CI index, and complete fault diagnosis. CI index is the characteristic fundamental frequency amplitude FFA and characteristic energy ratio FER.

[0124] Compared with the prior art, the present invention has the following advantages: the present invention uses sensors to collect vibration signals and speed signals of bearings in a transmission system of mechanical equipment; designs constraints on sparsity of Fourier dictionary adaptive control based on the frequency domain sparsity of discrete frequency components, and designs a cyclic dictionary learning algorithm for determining sparsity based on the shift invariance of fault pulses; establishes a morphological decomposition algorithm model with controllable sparsity according to morphological distinction characteristics; uses a greedy algorithm to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model to separate discrete frequencies, and uses the least squares method and the greedy algorithm to alternately solve the cyclic dictionary and its sparse representation to extract pulses; finally, performs envelope spectrum analysis on the reconstructed fault pulse, compares it with the characteristic frequency of the bearing fault, calculates the CI index to complete the fault diagnosis, and significantly improves the accuracy and reliability of the diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0125] By reading the detailed description of the preferred specific embodiments below, various other advantages and benefits of the present invention will become clear to those of ordinary skill in the art. The drawings in the specification are only for the purpose of illustrating the preferred embodiments and are not considered to be limitations of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative work. Moreover, the same reference numerals are used to represent the same components throughout the drawings.

[0126] In the attached picture:

[0127] Figure 1 It is a rolling bearing fault diagnosis method and system flow chart based on controllable sparsity morphological decomposition according to an embodiment of the present invention;

[0128] Figure 2 is a schematic diagram of the overall structure of a Safran engine according to an embodiment of the present invention;

[0129] Figure 3 is a rotation frequency curve diagram of the L5 axis according to an embodiment of the present invention;

[0130] Figure 4 (a) is an original time domain waveform diagram of a vibration signal of a slow train according to an embodiment of the present invention;

[0131] Figure 4 (b) is an original envelope spectrum of a vibration signal of a slow train according to an embodiment of the present invention;

[0132] Figure 5 (a) is a time domain waveform diagram of a low-speed slow train vibration signal after controllable sparsity morphological decomposition processing according to an embodiment of the present invention;

[0133] Figure 5 (b) is an envelope spectrum of a low-speed slow train vibration signal after controllable sparsity morphological decomposition processing according to an embodiment of the present invention;

[0134] Figure 6 (a) is an original time domain waveform diagram of a vibration signal at a maximum speed state according to an embodiment of the present invention;

[0135] Figure 6 (b) is an original envelope spectrum of a vibration signal at a maximum speed state according to an embodiment of the present invention;

[0136] Figure 7 (a) is a time domain waveform diagram of a vibration signal in a maximum speed state after being processed by controllable sparsity morphological decomposition according to an embodiment of the present invention;

[0137] Figure 7 (b) is an envelope spectrum of a vibration signal in a maximum speed state after being processed by controllable sparsity morphological decomposition according to an embodiment of the present invention.

[0138] The present invention is further explained below in conjunction with the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0139] The specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the specific embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0140] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present invention, but the description is based on the general principles of the specification and is not intended to limit the scope of the present invention. The scope of protection of the present invention shall be determined by the attached claims.

[0141] To facilitate understanding of the embodiments of the present invention, further explanation will be given below by taking specific embodiments as examples in conjunction with the accompanying drawings, and each of the accompanying drawings does not constitute a limitation on the embodiments of the present invention.

[0142] like Figures 1 to 7 As shown, the rolling bearing fault diagnosis method includes the following steps:

[0143] S100, collecting the vibration signal and the speed signal of the rolling bearing, and selecting the vibration signal under the stable working condition with a sample number of N according to the speed signal , When a bearing fails, the vibration signal Contains fault pulses with periodic shift invariant characteristics , and discrete frequency components with frequency domain sparsity , , discrete frequency components Including periodic signals caused by gear or shaft excitation;

[0144] S200, based on discrete frequency components The frequency domain sparsity of the Fourier dictionary is adaptively controlled by constructing the index constraint for evaluating the sparsity.

[0145] S300, based on fault pulse Based on the shift invariance of the fault cycle, a cyclic dictionary is constructed with the failure cycle as the atomic length and a cyclic dictionary learning algorithm is designed to determine the sparsity.

[0146] S400, based on discrete frequency components and fault pulse Based on the distinguishing characteristics of the morphological decomposition algorithm, a morphological decomposition algorithm model with controllable sparsity is established;

[0147] S500, according to the constraint of adaptively controlling sparsity, using a greedy algorithm to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model and separate the discrete frequency components ;

[0148] S600, according to the constraint of determining sparsity, alternately use the frequency domain least square method and the greedy algorithm to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively, and realize the fault pulse extract;

[0149] S700, for the extracted fault pulse Perform envelope spectrum analysis, compare it with the characteristic frequency of bearing fault, calculate CI index, and complete fault diagnosis. CI index is the characteristic fundamental frequency amplitude FFA and characteristic energy ratio FER.

[0150] In a preferred embodiment of the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, in S200, according to the discrete frequency components The frequency domain sparsity of , using an explicit Fourier dictionary and greedy algorithm to reconstruct discrete frequency components :

[0151] ;

[0152] The data matrix , From the vibration signal By length No overlap Segment structure, , is the sparse representation matrix of the Fourier dictionary, which adaptively controls the sparsity constraint of the Fourier dictionary It is composed of the following indicators that reflect the changes in frequency components:

[0153] 1) Based on the maximum difference to mean ratio MR:

[0154] ,

[0155] Constraints on the adaptive control sparsity of MR for:

[0156] ,

[0157] 2) Based on spectrum flatness SF:

[0158] ,

[0159] Constraints for adaptive control of sparsity of SF for:

[0160] ,

[0161] 3) Based on the peak factor PP:

[0162] ,

[0163] Constraints on adaptive control sparsity of PP composition for:

[0164] ,

[0165] in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, represents the mean operation; To reconstruct the spectrum of discrete frequency components The first The power spectrum of the interval, ; Indicates the maximum value calculation, Indicates minimum value calculation; is the reconstructed discrete frequency component Length.

[0166] In a preferred implementation of the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, in S300,

[0167] The atomic length of the loop dictionary design is ,in,

[0168] ,

[0169] In the formula is the sampling frequency, is the fault characteristic frequency, Indicates rounding up, according to the fault pulse The shift invariance of Vector Circular shift to construct a circular dictionary , :

[0170] ,

[0171] in,

[0172] , is the length of the pulse of interest and is a positive integer,

[0173] Vector is a circulatory vector, and ,in,

[0174] ,

[0175] Among them, vector Including multiple 0s, the number of 0s is equal to and The difference between

[0176] Further solving the circular dictionary And its sparse representation:

[0177] ,

[0178] The data matrix , The vibration signal Divide segments, each segment length is and are constructed without overlapping; , is a sparse representation matrix; is the sparsity constraint of the cyclic dictionary, set to be deterministic ,

[0179] The solution steps include:

[0180] S301, first use the unitary Fourier dictionary Convert the loop dictionary to frequency domain solution:

[0181] ,

[0182] in, and ; , is the data matrix The result after Fourier transformation; , is a sparse representation matrix The result after Fourier transformation is

[0183] S302. Decompose the least squares problem into Sub-questions:

[0184] ,

[0185] in, yes No. elements, yes No. List, yes No. List,

[0186] S303, Solution:

[0187] ,

[0188] S304, before loop vector Solve for values:

[0189] ,

[0190] in, , Represents the front of the loop vector elements, is the inverse Fourier transform,

[0191] S305. Solving sparse representation by greedy algorithm and sparse constraints .

[0192] Preferably, in S400,

[0193] Establish a morphological decomposition algorithm model with controllable sparsity:

[0194] ,

[0195] Sparsity Constraints Different situations make the morphological decomposition algorithm model include:

[0196] 1) Model based on MR adaptive control sparsity constraint:

[0197] ,

[0198] 2) Model based on SF adaptive control sparsity constraint:

[0199] ,

[0200] 3) Model based on SF adaptive control sparsity constraint:

[0201] ,

[0202] in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, Represents the mean operation.

[0203] In a preferred embodiment of the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, in S400, in the controllable sparsity morphological decomposition algorithm model, the Fourier dictionary The atomic length Set to:

[0204] ,

[0205] in, Yes Select data The length of An integer multiple of .

[0206] In a preferred implementation of the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, in S500, discrete frequency components are first separated, including:

[0207] S501, firstly By length No overlap Segment, construct data matrix , , construct an explicit Fourier dictionary , , separating the discrete frequency components is equivalent to solving the subproblem:

[0208] ,

[0209] Initialize the residual matrix , , initialize the support set ,initialization Satisfy the constraints and set the upper limit of sparsity to ;

[0210] S502, select explicit Fourier dictionary , The residual , Most relevant atoms:

[0211] ,

[0212] in, , is a dictionary matrix No. List,

[0213] Add the newly selected index to the backing set middle:

[0214] ,

[0215] S503, using the least squares method to update the sparse representation matrix , :

[0216] ,

[0217] in, is the support set The corresponding set of atoms, is the current support set Compute the sparse representation matrix;

[0218] Update the residual matrix:

[0219] ,

[0220] Update iteration count ,until ;

[0221] S504. Determine the constraints of sparsity Which of the following is met:

[0222] 1) Constraints of adaptive control sparsity based on MR:

[0223] ,

[0224] 2) Adaptive control of sparsity constraints based on SF:

[0225] ,

[0226] 3) Constraints of adaptive control sparsity based on PP:

[0227] ,

[0228] in, , Represents the coefficient of the current sparsity solution, , Represents the coefficient of the previous sparsity solution;

[0229] If the constraints are not met, keep , separate discrete frequency components from the data :

[0230] ,

[0231] in, , is the residual signal after separating the discrete frequency components.

[0232] Otherwise, the sparsity , return to step S502.

[0233] In a preferred implementation of the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, in S600, extracting pulses includes the steps of:

[0234] S601, the residual signal By length No overlap Segment, construct data matrix , ; Random selection A column in is used as a cyclic vector to construct a cyclic dictionary , randomly initialized , , extracting the fault pulse is equivalent to solving the subproblem:

[0235] ,

[0236] right and Perform Fourier transform to get , and , , initialize the residual matrix , , initialize the support set ;

[0237] S602: Calculate the elements of the dictionary :

[0238] ,

[0239] in, yes No. elements, yes No. OK, yes No. OK;

[0240] Calculate the front of the loop vector Values:

[0241] ,

[0242] Get the loop vector And construct the loop dictionary ;

[0243] S603: Select a circular dictionary based on the determined sparsity constraint Zhongyu Most relevant atoms:

[0244] ,

[0245] in, , is a dictionary matrix No. List,

[0246] Add the newly selected index to the backing set middle:

[0247] ,

[0248] Update the sparse representation matrix using the least squares method , :

[0249] ,

[0250] in, is the support set The corresponding set of atoms;

[0251] S604, reconstruct the current loop dictionary and the sparse representation matrix Fault pulse , :

[0252] ,

[0253] Reconstruct the loop dictionary from the previous iteration , and the sparse representation matrix , Fault pulse , :

[0254] ,

[0255] in, , is the circular dictionary obtained by the previous iteration. , is the sparse representation matrix obtained by the previous iteration;

[0256] S605, Calculation and The root mean square error when Or when the number of iterations Nit is reached, the reconstructed pulse is obtained , , otherwise, return to step S602.

[0257] In another embodiment, the present invention further discloses a rolling bearing fault diagnosis system based on controllable sparsity morphological decomposition, comprising:

[0258] The acquisition unit is used to collect the vibration signal and speed signal of the rolling bearing, and select the vibration signal segment under stable working conditions according to the speed signal , ;

[0259] The construction unit is used to generate a discrete frequency component based on The frequency domain sparsity of , construct an index constraint to evaluate sparsity, so as to adaptively control the sparsity of Fourier dictionary; based on fault pulse The shift invariance of , construct a cyclic dictionary and design a cyclic dictionary learning algorithm to determine the sparsity; based on discrete frequency components and fault pulse Based on the distinguishing characteristics of the morphological decomposition algorithm, a morphological decomposition algorithm model with controllable sparsity is established;

[0260] The extraction unit is used to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model using a greedy algorithm according to the constraints of adaptively controlling the sparsity, and to separate the discrete frequency components According to the constraints of determining sparsity, the frequency domain least squares method and greedy algorithm are used alternately to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively to realize the fault pulse extract;

[0261] Diagnostic unit, which is used to extract fault pulses Perform envelope spectrum analysis, compare it with the characteristic frequency of bearing fault, calculate CI index, and complete fault diagnosis. CI index is the characteristic fundamental frequency amplitude FFA and characteristic energy ratio FER.

[0262] In another embodiment, Figure 1 As shown, a rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition includes the following steps:

[0263] Use sensors to collect the vibration signal and speed signal of the rolling bearing, and select the vibration signal segment under stable working conditions based on the speed signal , ;

[0264] Based on discrete frequency components The frequency domain sparsity of the Fourier dictionary is adaptively controlled by constructing the index constraint for evaluating the sparsity.

[0265] Based on fault pulse Shift invariance, design a cyclic dictionary learning algorithm with determined sparsity;

[0266] Based on fault pulse Based on the shift invariance of the fault cycle, a cyclic dictionary is constructed with the failure cycle as the atomic length and a cyclic dictionary learning algorithm is designed to determine the sparsity.

[0267] According to the sparse constraints, greedy algorithms (including matching pursuit algorithm, weak matching pursuit algorithm, orthogonal matching pursuit algorithm, batch orthogonal matching pursuit algorithm and compressed sampling matching pursuit algorithm, etc.) are used to solve the sparse representation of the Fourier dictionary in the model and separate the discrete frequency components. ;

[0268] According to the constraints of determining sparsity, the frequency domain least squares method and greedy algorithm are used alternately to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively to realize the fault pulse extract;

[0269] The extracted fault pulse is analyzed by envelope spectrum, compared with the characteristic frequency of the bearing fault, and the CI index is calculated to complete the fault diagnosis.

[0270] In this embodiment, the sensor collects vibration signals and speed signals to obtain the speed information and vibration information of the rolling bearing, wherein the speed information is obtained by installing various speed sensors on any shaft in the entire transmission system, and the vibration information is obtained by installing various vibration sensors near the component of interest or on the outer casing; the obtained vibration signal and speed signal are converted into analog / digital to facilitate further processing; in some other cases, the vibration signal or pulse signal is pre-processed by low-pass, high-pass or band-pass filtering, and the signal in this case is still a vibration signal and a pulse signal that meets the requirements of this document; the embodiment is further explained as follows:

[0271] According to the discrete frequency components The frequency domain sparsity of , using an explicit Fourier dictionary , and greedy algorithm to reconstruct discrete frequency components :

[0272] ;

[0273] The data matrix , From the vibration signal By length No overlap Segment structure, , is the sparse representation matrix of the Fourier dictionary;

[0274] The following mainly focuses on the sparsity constraints of the three indicators reflecting the change of frequency components in this embodiment of the present invention. To introduce:

[0275] 1) Based on the maximum difference to mean ratio MR:

[0276] ,

[0277] Constraints on the adaptive control sparsity of MR for:

[0278] ,

[0279] 2) Based on spectrum flatness SF:

[0280] ,

[0281] Constraints for adaptive control of sparsity of SF for:

[0282] ,

[0283] 3) Based on the peak factor PP:

[0284] ,

[0285] Constraints on adaptive control sparsity of PP composition for:

[0286] ,

[0287] in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, represents the mean operation; To reconstruct the spectrum of discrete frequency components The first The power spectrum of the interval, ; Indicates the maximum value calculation, Indicates minimum value calculation; is the reconstructed discrete frequency component Length.

[0288] This provides the basis for separating discrete frequency components.

[0289] Among them, the atomic length of the loop dictionary design is ,and:

[0290] ,

[0291] In the formula is the sampling frequency, is the fault characteristic frequency, Indicates rounding up. According to the fault pulse The shift invariance of Vector Circular shift to construct a circular dictionary , :

[0292] ,

[0293] It should be noted that the present invention mainly focuses on the length The main pulse of is taken as the pulse of interest, and the circulant vector of the constructed circulant matrix is:

[0294]

[0295] In the formula, , the sum of the number of 0 and the positive integer n is equal to ,

[0296] Further solving the circular dictionary And its sparse representation:

[0297] ,

[0298] The data matrix , The vibration signal Divide segments, each segment length is and are constructed without overlapping; , is a sparse representation matrix; is the sparsity constraint of the cyclic dictionary, set to be deterministic ;

[0299] Furthermore, the solution steps include:

[0300] Step a: Use the implicit unitary Fourier dictionary Convert the loop dictionary to frequency domain solution:

[0301] ,

[0302] in, and ; , is the data matrix The result after Fourier transformation; , is a sparse representation matrix The result after Fourier transformation.

[0303] Step b: Decompose the least squares problem into Sub-questions:

[0304] ,

[0305] in, yes The jth element of yes The jth column of yes The jth column of .

[0306] Step c. It can be solved:

[0307] ,

[0308] Step d, before looping vector The values ​​can be solved:

[0309] ,

[0310] in, , Represents the front of the loop vector elements, is the inverse Fourier transform,

[0311] Among them, sparse representation Solving through greedy algorithms and sparse constraints;

[0312] This provides a basis for extracting fault pulses;

[0313] Furthermore, based on the distinguishing characteristics of discrete frequency components and fault pulses, a morphological decomposition algorithm model with controllable sparsity is established:

[0314] ,

[0315] Sparsity constraints (also called constraints) The difference makes the model have the following specific forms:

[0316] a. Model based on MR adaptive control sparsity constraint:

[0317] ,

[0318] b. Model based on SF adaptive control sparsity constraint:

[0319] ,

[0320] c. Model based on PP adaptive control sparsity constraint:

[0321] ,

[0322] in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, Represents the mean operation.

[0323] Among them, in the morphological decomposition algorithm model with controllable sparsity, the Fourier dictionary The atomic length Set to:

[0324] ,

[0325] in, Yes Select data The length of An integer multiple of ;

[0326] It should be noted that separation of discrete frequency components specifically includes:

[0327] Step a: First, the data By length No overlap Segment, construct data matrix , , construct an explicit Fourier dictionary , , separating the discrete frequency components is equivalent to solving the subproblem:

[0328] ,

[0329] Initialize the residual matrix , , initialize the support set ,initialization Satisfy the constraints and set the upper limit of sparsity to ;

[0330] Step b: Select the explicit Fourier dictionary , The residual , Most relevant atoms:

[0331] ,in, is a dictionary matrix No. Add the newly selected index to the backing set middle:

[0332] ,

[0333] Step c: Update the sparse representation matrix using the least squares method :

[0334] ,

[0335] in, is the support set The corresponding set of atoms, is the current support set Compute the sparse representation matrix;

[0336] Update the residual matrix:

[0337] ,

[0338] Update iteration count ,until ;

[0339] Step d: Determine the sparsity constraint , involving the following multiple judgments to determine which one is met or satisfied:

[0340] a. Constraints of adaptive control sparsity based on MR:

[0341] ,

[0342] b. Constraints of adaptive control sparsity based on SF:

[0343] ,

[0344] c. Constraints of adaptive control sparsity based on PP:

[0345] ,

[0346] in, , Represents the coefficient of the current sparsity solution, , Represents the coefficient of the previous sparsity solution;

[0347] If the constraints are not met, keep , which can separate discrete frequency components from the data :

[0348] ,

[0349] in, , It is the residual signal after separating the discrete frequency components;

[0350] Otherwise, the sparsity , return to step b;

[0351] Further, after separating the discrete frequency components, extracting the pulse specifically includes:

[0352] Step a: The residual signal By length No overlap Segment, construct data matrix , ; Random selection A column in is used as a cyclic vector to construct a cyclic dictionary , randomly initialized , , extracting the fault pulse is equivalent to solving the subproblem:

[0353] ,

[0354] right and Perform Fourier transform to get , and , , initialize the residual matrix , , initialize the support set ;

[0355] Step b: Calculate the elements of the dictionary :

[0356] ,

[0357] in, yes No. elements, yes No. List, yes No. List;

[0358] Calculate the front of the loop vector Values:

[0359] ,

[0360] get And construct ;

[0361] Step c: Select the cyclic dictionary according to the determined sparsity constraint The residual Most relevant atoms:

[0362] ,

[0363] in, , is a dictionary matrix No. List.

[0364] Add the newly selected index to the backing set middle:

[0365] ,

[0366] Update the sparse representation matrix using the least squares method :

[0367] ,

[0368] in, is the support set The corresponding set of atoms;

[0369] Step d: Reconstruct the current loop dictionary and the sparse representation matrix Fault pulse , : Reconstruct the loop dictionary from the previous iteration , and the sparse representation matrix , Fault pulse , :

[0370] in, , is the circular dictionary obtained by the previous iteration. , is the sparse representation matrix obtained by the previous iteration;

[0371] Step e: Calculation and The root mean square error when Or when the number of iterations Nit is reached, the reconstructed pulse is obtained , , otherwise, return to step b;

[0372] Furthermore, for the reconstructed fault pulse Perform envelope analysis, where the envelope spectrum is obtained by the following processing:

[0373] ,

[0374] ,

[0375] in, represents the Hilbert transform, is the result of the Hilbert transform, represents the discrete Fourier transform, is the obtained envelope spectrum;

[0376] Further observe whether there is an obvious inner race fault characteristic frequency in the envelope spectrum , Outer ring fault characteristic frequency , rolling element fault characteristic frequency and cage failure characteristic frequency , using CI indicators for quantitative evaluation;

[0377] The CI index consists of the following indicators:

[0378] 1) Characteristic fundamental frequency amplitude FFA:

[0379] ,

[0380] 2) Characteristic energy ratio FER:

[0381] ,

[0382] in, Represents the envelope spectrum The amplitude of the first fundamental frequency of the bearing fault characteristic frequency; Represents the envelope spectrum The characteristic frequency of the bearing fault The amplitude of the order frequency.

[0383] In another embodiment, this embodiment involves the diagnosis of a rolling bearing fault of an aircraft engine. The diagnosis method adopts the method described above, and the above method is specifically implemented using matlab. The diagnosis object is as follows: Figure 2 The Safran engine shown in the figure, wherein the engine accessory transmission system includes bearings such as L1, L4, and L5, wherein the outer ring of the L1 shaft ball bearing is severely scratched, and the outer ring of the L5 shaft roller bearing is peeled off. Three acceleration sensors Acc0, Acc1, and Acc2, as well as a speed sensor Tacho are arranged on the engine to collect vibration signals and speed signals respectively, with a sampling frequency of 44100Hz. This embodiment focuses on the analysis of the outer ring peeling fault of the L5 shaft bearing, and the ratio of the characteristic frequency of the outer ring fault of the L5 shaft bearing to the speed is 7.759.

[0384] Figure 3 is a rotation frequency curve of the L5 axis, which represents the rotation speed change of the L5 axis within 200 seconds. This embodiment uses a rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition to analyze 65536 sampling points in the low-speed slow vehicle state (the first 47 seconds) and 65536 sampling points in the maximum speed state (after 155 seconds) as vibration signals.

[0385] In this embodiment, the fault pulse extraction of the vibration signal is performed, and the constraint condition (also called constraint) of Fourier dictionary adaptive control sparsity based on the maximum difference mean ratio MR is selected. Model:

[0386] ,

[0387] It should be noted that the constraints of the sparsity are adaptively controlled using the Fourier dictionary based on the spectrum flatness SF and the peak factor PP. Similar results can be obtained with the model. Explicit Fourier dictionary The atomic length In this embodiment, it is set to 512, and the corresponding Fourier dictionary sparse matrix is .

[0388] First, according to the fifth step, the sparse representation matrix of the Fourier dictionary is calculated using the greedy algorithm under the sparsity constraint , reconstructing the discrete frequency components And realize the separation of discrete frequency components , and get the residual signal Next, follow the sixth step to solve the cyclic dictionary under the sparsity constraint And use the greedy algorithm to solve the sparse representation matrix , extract the fault pulse under the condition of meeting the error or maximum number of iterations .

[0389] Figure 4 (a) is the original time domain waveform of the vibration signal under the low-speed slow train state. Figure 4 (b) is the envelope spectrum of the vibration signal under low speed and slow train condition. Figure 4 It can be seen from (b) that the characteristic frequency of the outer race fault is It is very weak and difficult to observe directly. The calculated characteristic fundamental frequency amplitude FFA value in the envelope spectrum is 0.006, and the characteristic energy ratio FER value is 0.0041. The rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition is used to decompose the original signal and extract the fault pulse component in the signal. The results are as follows: Figure 5 .from Figure 5 As can be seen in (b), the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition extracts the obvious outer ring fault characteristic frequency , the characteristic fundamental frequency amplitude FFA value is 0.009, and the characteristic energy ratio FER value is 0.0382. This proves that the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition extracts effective outer ring fault characteristic frequency , diagnosed a fault in the outer ring of the accessory casing of the Safran engine at low speed.

[0390] The original time domain waveform and envelope spectrum of the vibration signal under the maximum speed state are plotted on Figure 6 In. From Figure 6 It can be seen from (b) that the characteristic frequency of the outer race fault is The vibration signal is weak and is easily misjudged by other frequency components around it. By calculation, the characteristic fundamental frequency amplitude FFA value of the original envelope spectrum of the vibration signal at the maximum speed state is 0.0105, and the characteristic energy ratio FER value is 0.0084. The rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition is continued to decompose the original signal and extract the fault pulse component in the signal. The results are as follows: Figure 7 .from Figure 7 As can be seen in (b), the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition still extracts the obvious outer ring fault characteristic frequency. , the characteristic fundamental frequency amplitude FFA value of the envelope spectrum is 0.0158, and the characteristic energy ratio FER value is 0.0382. This proves that the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition has extracted effective outer ring fault characteristic frequency , diagnosed a fault in the outer ring of the accessory case at maximum speed of the Safran engine.

[0391] In another embodiment, the greedy algorithm includes a matching pursuit algorithm, a weak matching pursuit algorithm, an orthogonal matching pursuit algorithm, a batch orthogonal matching pursuit algorithm, and a compressed sampling matching pursuit algorithm.

[0392] In another embodiment, the bearing fault characteristic frequency includes the inner race fault characteristic frequency , Outer ring fault characteristic frequency , rolling element fault characteristic frequency and cage failure characteristic frequency .

[0393] Although the embodiments of the present invention are described above in conjunction with the accompanying drawings, the present invention is not limited to the above specific embodiments and application fields, and the above specific embodiments are only illustrative and instructive, rather than restrictive. A person of ordinary skill in the art can also make many forms under the guidance of this specification and without departing from the scope of protection of the claims of the present invention, all of which belong to the protection of the present invention.

Claims

1. A rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition, characterized in that: The steps include: S100, collecting the vibration signal and the speed signal of the rolling bearing, and selecting the vibration signal under the stable working condition with a sample number of N according to the speed signal , When a bearing fails, the vibration signal Contains fault pulses with periodic shift invariant characteristics , and discrete frequency components with frequency domain sparsity , , discrete frequency components Including periodic signals caused by gear or shaft excitation; S200, based on discrete frequency components The frequency domain sparsity of the Fourier dictionary is adaptively controlled by constructing the index constraint for evaluating the sparsity. S300, based on fault pulse Based on the shift invariance of the fault cycle, a cyclic dictionary is constructed with the failure cycle as the atomic length and a cyclic dictionary learning algorithm is designed to determine the sparsity. S400, based on discrete frequency components and fault pulse Based on the distinguishing characteristics of the morphological decomposition algorithm, a morphological decomposition algorithm model with controllable sparsity is established; S500, according to the constraint of adaptively controlling sparsity, using a greedy algorithm to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model and separate the discrete frequency components ; S600, according to the constraint of determining sparsity, alternately use the frequency domain least square method and the greedy algorithm to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively, and realize the fault pulse extract; S700, for the extracted fault pulse Perform envelope spectrum analysis, compare with the characteristic frequency of bearing fault, calculate CI index, and complete fault diagnosis. CI index is the characteristic fundamental frequency amplitude FFA and characteristic energy ratio FER; Among them, S400, Establish a morphological decomposition algorithm model with controllable sparsity: ， Sparsity Constraints Different situations make the morphological decomposition algorithm model include: 1) Model based on MR adaptive control sparsity constraint: ， 2) Model based on SF adaptive control sparsity constraint: ， 3) Model based on SF adaptive control sparsity constraint: ， in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, Represents the mean operation.

2. A rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition according to claim 1, characterized in that: In S200, according to the discrete frequency components The frequency domain sparsity of , using an explicit Fourier dictionary , and greedy algorithm to reconstruct discrete frequency components : ； The data matrix , From the vibration signal By length No overlap Segment structure, , is the sparse representation matrix of the Fourier dictionary, which adaptively controls the sparsity constraint of the Fourier dictionary It is composed of the following indicators that reflect the changes in frequency components: 1) Based on the maximum difference to mean ratio MR: ； Constraints on the adaptive control sparsity of MR for: ； 2) Based on spectrum flatness SF: ； Constraints for adaptive control of sparsity of SF for: ； 3) Based on the peak factor PP: ； Constraints on adaptive control sparsity of PP composition for: ， in, , Represents the sparse representation matrix of the current solution; , Represents the discrete frequency components of the current reconstruction; It is an operation that concatenates the columns of a matrix into a one-dimensional vector; , The sparse representation matrix representing the solution of the previous sparsity level; , Represents the discrete frequency components reconstructed at the previous sparsity level; Represents the result after Fourier transform; , is the spectral envelope of the bearing signal and its standard deviation ; The value range of ; represents the absolute value operation, is the infinite norm, represents the mean operation; To reconstruct the spectrum of discrete frequency components The first The power spectrum of the interval, ; Indicates the maximum value calculation, Indicates minimum value calculation; is the reconstructed discrete frequency component Length.

3. The rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition according to claim 1 is characterized in that: In S300, the atomic length of the loop dictionary design is ,in, ， In the formula is the sampling frequency, is the fault characteristic frequency, Indicates rounding up, according to the fault pulse The shift invariance of Vector Circular shift to construct a circular dictionary , : ， in, , is the length of the pulse of interest and is a positive integer, Vector is a circulatory vector, and ,in, ， Among them, vector Including multiple 0s, the number of 0s is equal to and The difference between Further solving the circular dictionary And its sparse representation: ， The data matrix , The vibration signal Divide segments, each segment length is and are constructed without overlapping; , is a sparse representation matrix; is the sparsity constraint of the cyclic dictionary, set to be deterministic , The solution steps include: S301, first use the unitary Fourier dictionary Convert the loop dictionary to frequency domain solution: ， in, and ; , is the data matrix The result after Fourier transformation; , is a sparse representation matrix The result after Fourier transformation is S302. Decompose the least squares problem into Sub-questions: ， in, yes No. elements, yes No. List, yes No. List, S303, Solution: ， S304, before loop vector Solve for values: ， in, , Represents the front of the loop vector elements, is the inverse Fourier transform, S305. Solving sparse representation by greedy algorithm and sparse constraints .

4. The rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition according to claim 1 is characterized in that: In S400, the Fourier dictionary in the morphological decomposition algorithm model with controllable sparsity The atomic length Set to ， in, Yes Select data The length of An integer multiple of .

5. The rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition according to claim 1 is characterized in that: In S500, discrete frequency components are first separated. include, S501, firstly By length No overlap Segment, construct data matrix , , construct an explicit Fourier dictionary , , separating the discrete frequency components is equivalent to solving the subproblem: ， Initialize the residual matrix , , initialize the support set ,initialization Satisfy the constraints and set the upper limit of sparsity to ; S502, select explicit Fourier dictionary , The residual , Most relevant atoms: ， in, , is a dictionary matrix No. List, Add the newly selected index to the backing set middle: ， S503, using the least squares method to update the sparse representation matrix , : ， in, is the support set The corresponding set of atoms, is the current support set Compute the sparse representation matrix; Update the residual matrix: ， Update iteration count ,until ; S504. Determine the constraints of sparsity Which of the following is met: 1) Constraints of adaptive control sparsity based on MR: ， 2) Adaptive control of sparsity constraints based on SF: ， 3) Constraints of adaptive control sparsity based on PP: ， in, , Represents the coefficient of the current sparsity solution, , Represents the coefficient of the previous sparsity solution; If the constraints are not met, keep , separate discrete frequency components from the data : ， in, , is the residual signal after separating the discrete frequency components. Otherwise, the sparsity , return to step S502.

6. The rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition according to claim 1 is characterized in that: In S600, extracting the pulse includes the steps of: S601, the residual signal By length No overlap Segment, construct data matrix , ; Random selection A column in is used as a cyclic vector to construct a cyclic dictionary , randomly initialized , , extracting the fault pulse is equivalent to solving the subproblem: ， right and Perform Fourier transform to get , and , , initialize the residual matrix , , initialize the support set ; S602: Calculate the elements of the dictionary : ， in, yes No. elements, yes No. OK, yes No. OK; Calculate the front of the loop vector Values: ， Get the loop vector And construct the loop dictionary ; S603: Select a circular dictionary based on the determined sparsity constraint Zhongyu Most relevant atoms: ， in, , is a dictionary matrix No. List, Add the newly selected index to the backing set middle: ， Update the sparse representation matrix using the least squares method , : ， in, is the support set The corresponding set of atoms; S604, reconstruct the current loop dictionary and the sparse representation matrix Fault pulse , : ， Reconstruct the loop dictionary from the previous iteration , and the sparse representation matrix , Fault pulse , : ， in, , is the circular dictionary obtained by the previous iteration. , is the sparse representation matrix obtained by the previous iteration; S605, Calculation and The root mean square error when Or when the number of iterations Nit is reached, the reconstructed pulse is obtained , , otherwise, return to step S602.

7. A system for implementing the rolling bearing fault diagnosis method based on controllable sparsity morphological decomposition as claimed in any one of claims 1 to 6, characterized in that: These include, The acquisition unit is used to collect the vibration signal and speed signal of the rolling bearing, and select the vibration signal segment under stable working conditions according to the speed signal , ; The construction unit is used to generate a discrete frequency component based on The frequency domain sparsity of , construct an index constraint to evaluate sparsity, so as to adaptively control the sparsity of Fourier dictionary; based on fault pulse The shift invariance of , construct a cyclic dictionary and design a cyclic dictionary learning algorithm to determine the sparsity; based on discrete frequency components and fault pulse Based on the distinguishing characteristics of the morphological decomposition algorithm, a morphological decomposition algorithm model with controllable sparsity is established; The extraction unit is used to solve the sparse representation of the Fourier dictionary in the morphological decomposition algorithm model using a greedy algorithm according to the constraints of adaptively controlling the sparsity, and to separate the discrete frequency components ; According to the constraints of determining sparsity, the frequency domain least squares method and greedy algorithm are used alternately to solve the sparse representation of the cyclic dictionary and the cyclic dictionary respectively to realize the fault pulse extract; Diagnostic unit, which is used to extract fault pulses Perform envelope spectrum analysis, compare it with the characteristic frequency of bearing fault, calculate CI index, and complete fault diagnosis. CI index is the characteristic fundamental frequency amplitude FFA and characteristic energy ratio FER.

Images