CN112414713A - Rolling bearing fault detection method based on measured signals - Google Patents
Rolling bearing fault detection method based on measured signals Download PDFInfo
- Publication number
- CN112414713A CN112414713A CN202011217106.9A CN202011217106A CN112414713A CN 112414713 A CN112414713 A CN 112414713A CN 202011217106 A CN202011217106 A CN 202011217106A CN 112414713 A CN112414713 A CN 112414713A
- Authority
- CN
- China
- Prior art keywords
- rolling bearing
- fault
- support vector
- signal
- new
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005096 rolling process Methods 0.000 title claims abstract description 42
- 238000001514 detection method Methods 0.000 title claims abstract description 18
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 64
- 238000012706 support-vector machine Methods 0.000 claims abstract description 31
- 238000005457 optimization Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 22
- 241001481710 Cerambycidae Species 0.000 claims abstract description 18
- 238000010845 search algorithm Methods 0.000 claims abstract description 13
- 238000005070 sampling Methods 0.000 claims description 38
- 239000013598 vector Substances 0.000 claims description 35
- 238000000034 method Methods 0.000 claims description 34
- 230000008569 process Effects 0.000 claims description 19
- 230000001133 acceleration Effects 0.000 claims description 9
- 238000012952 Resampling Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 abstract description 29
- 238000005259 measurement Methods 0.000 abstract description 6
- 230000008859 change Effects 0.000 abstract description 5
- 239000000306 component Substances 0.000 description 13
- 238000004458 analytical method Methods 0.000 description 9
- 238000003745 diagnosis Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002040 relaxant effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
Landscapes
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a rolling bearing fault detection method based on an actual measurement signal, which comprises the following steps of firstly, converting a rolling bearing fault time domain vibration signal into an angular domain by adopting an order ratio tracking technology; then, parameter optimization is carried out on the variation modal decomposition by utilizing a longicorn beard search algorithm, vibration signals of all states of the rolling bearing are decomposed, a series of inherent modal functions are obtained, and when the bearing has different faults, frequency band energy in different inherent modal functions can change; secondly, extracting Renyi entropy characteristics from the modal components containing the main fault information, and constructing a characteristic subset; and finally, training by using the vibration signals in the normal state which are easy to obtain, extracting fault characteristic quantity, establishing a fault data sample and an incremental learning data sample, training by adopting an incremental learning algorithm of a single-class support vector machine to obtain a fault identification model, accurately judging whether the rolling bearing has a fault or not, and realizing fault early warning.
Description
Technical Field
The invention belongs to the technical field of electrical fault detection, and particularly relates to a rolling bearing fault detection method based on an actual measurement signal.
Background
The rolling bearing is used as a core component of the wind turbine, the wind turbine is continuously influenced by impact force and load due to the severe working environment of the wind turbine, and meanwhile, the wind speed has intermittency and strong fluctuation, so that the wind turbine faces complex working conditions. The fan rolling bearing vibration signal fault characteristics under the variable working condition scene are unstable, the working condition is complex, and the fault sample data is less, so that the bearing fault analysis is difficult. After long-term operation, the bearing of the wind turbine generator is easy to break down, and once the bearing fails, noise and abnormal sound are generated slightly, and the driving system is broken down seriously, so that the operation of the wind turbine generator is seriously influenced. Due to the restriction of high-altitude, low-speed and heavy-load working conditions of the wind turbine generator, the bearing is not easy to observe and disassemble, the difficulty and inconvenience are often brought to workers in fault analysis and judgment, the great difficulty is brought to the operation and maintenance of wind turbine equipment, the maintenance cost of the wind turbine generator is increased, the fault can be found and arranged and maintained as soon as possible in the early fault diagnosis and fault trend prediction of the bearing, and the method has important significance in guaranteeing the safe and reliable operation of the wind turbine generator and the economical efficiency of a wind farm.
The fault diagnosis of the fan bearing generally analyzes and processes a bearing vibration signal, and common bearing vibration signal time-frequency analysis methods comprise wavelet transformation, empirical mode decomposition, local mean modal decomposition and variational modal decomposition. The wavelet transform inherits and develops the idea of short-time Fourier transform localization, overcomes the defects that the window size does not change along with the frequency and the like, can provide a time-frequency window changing along with the frequency, but can obtain the best effect only by selecting different basis functions when processing complex vibration signals, and the parameter selection has no unified standard. The empirical mode decomposition decomposes a non-stationary vibration signal into a series of eigenmode functions with different characteristic scales, and the energy of the eigenmode functions is input as a neural network, but the problems of inaccurate envelope fitting, divergence at the boundary and the like exist. The local mean modal decomposition is an improvement on empirical modal decomposition, optimizes the capability of inhibiting an endpoint, but is easy to generate modal aliasing phenomenon and reduces the accuracy of signal decomposition. The variational modal decomposition is a multi-component signal self-adaptive decomposition method, has good anti-noise performance when processing main bearing signals, but the modal number needs to be estimated according to prior knowledge, and if the modal number is selected unreasonably, a larger decomposition error is easily caused.
Disclosure of Invention
The invention aims to provide a rolling bearing fault detection method based on an actual measurement signal, and solves the problems that in the prior art, the fault characteristics of a bearing signal based on a variable working condition unbalanced small sample are difficult to extract and the identification accuracy is low.
The technical scheme adopted by the invention is that a rolling bearing fault detection method based on an actual measurement signal is implemented according to the following steps:
and 5, adopting a new single-type support vector machine to carry out fault identification on the bearing vibration signal.
The invention is also characterized in that:
the specific process of the step 1 is as follows:
step 1.1, sampling a vibration signal and a rotating speed signal at equal time intervals delta t in two paths by a constant sampling rate, wherein the vibration signal is set as x (n), and the rotating speed signal is set as s (n);
step 1.2, calculating a time sequence t corresponding to the equal angle increment delta theta through a rotating speed signal s (n)ik(ii) a Firstly, searching the points of s (n) which satisfy the condition that s (n) is less than or equal to 0 and s (n +1) is greater than or equal to O in s (n), and assuming the corresponding time value tsnAnd tsn+1The waveform in between is linear;
the exact zero-crossing time tz of s (n) is determined from equation (1)iThe time value corresponding to each rotation of the reference shaft;
for time series tziThe difference is calculated to obtain the time T required by each rotation of the reference shaftiAssuming that the reference axis does a uniform variable speed motion within one revolution, and assuming that the angular accelerations of the first two revolutions are equal; and (3) calculating the equiangular sampling time t in the cycle by recursion in combination with a kinematic lawikAnd instantaneous speed of rotation vik；
Step 1.3, according to the equal-angle sampling time value tik( i 1, 2..) interpolating the vibration signal x (n) to obtain the value t of the vibration signal at the equal angular sampling time tikCorresponding amplitude value, obtaining an equiangular resampling signal x (t) of the vibration signalik) I.e. the angular domain stationary signal x (t)ik)。
Step 1.2, combining the law of kinematics, calculating the i-th-3, 4-th equiangular sampling time t in the week in a recursion mannerikAnd instantaneous speed of rotation vikThe specific process is as follows:
the first two weeks are estimated by equation set (2)Initial velocity v of10And angular acceleration a1Comprises the following steps:
setting equal-angle sampling M points of each rotation of a reference shaft, namely equal-angle increment delta theta is 2 pi/M; therefore, the angle of the k-th sampling point rotating corresponding to the reference axis is 2k pi/M, and the time value t corresponding to each equiangular sampling point in the first two weeks is obtained by the formula (3) and the formula (4)ikInstantaneous rotational speed vikAnd initial velocity v of week 330；
v30＝v10+a1(T1+T2) (4)
Wherein, the angle mark i represents the number of revolutions; the corner mark k represents the kth sample point in each revolution;
the angular acceleration ai and the initial velocity v of the i-th cycle of 3,4 and … are obtained by recursion of the formula (5)i+1,0；
The equiangular sampling time t in the week can be recursively calculated according to equation (6), wherein the i-th is 3,4ikAnd instantaneous speed of rotation vik；
The specific process of the step 2 is as follows: by initializing the variational modal decomposition number K and the penalty factor alpha, setting a set of the decomposition number K and the penalty factor alpha as a search space of a search algorithm of a Tianniu searching place, setting a network error function MSF as a target optimization function, and iterating to obtain the optimal variational modal decomposition number K and the optimal penalty factor alpha; and carrying out self-adaptive decomposition on the angular domain stationary signals through a variational modal decomposition algorithm with the optimal variational modal decomposition number K and the optimal penalty factor alpha to obtain an inherent modal function component.
The specific process of obtaining the optimal variational modal decomposition number K and the optimal penalty factor alpha comprises the following steps:
step A, setting x as the central position of the longicorn, xleftAnd xrightThe two whiskers serving as the longicorn are respectively positioned on the left side and the right side of a central position x at a distance d; wherein, xleft，xrightAre all n-dimensional vectors;
step B, setting the direction of each previous step of the longicorn to be a random direction, namely the direction of the right beard pointing to the left beard has uncertainty; each n-dimensional unit random vector is generated, and dir represents the direction of the right whisker pointing to the left whisker:
in equation (7), rands (n,1) represents that an n-dimensional random vector is generated, and the positions of the left and right whiskers are:
xleft＝x(k)+dir·d(k)/2 (8)
xright＝x(k)-dir·d(k)/2 (9)
in formulas (8) and (9): x (k) represents the center position at the k-th iteration, and d (k) represents the two-whisker distance at the k-th iteration;
c, calculating the corresponding fitness values f of the two whiskers according to the current position and the target optimization function fleft，frightAnd determining the advancing direction of the next step according to the magnitude relation:
x(k+1)＝x(k)-S(k)·dir·sgn(fleft-fright) (10)
in equation (10), s (k) represents the advance step at the k-th iteration;
updating the decomposition number K and the penalty factor alpha according to the advancing direction;
and D, judging whether the optimization precision is met or is greater than the maximum iteration number, finishing the optimization if the optimization precision is met, outputting the optimal variational modal decomposition number K and the optimal penalty factor alpha, and otherwise, returning to the step B.
The formula for the Renyi entropy is:
in the formula, a is the order of entropy, and when a is 1, the order Renyi entropy is Shannon entropy; pi is the probability density (i ═ 1,2, …).
The specific process of the step 4 is as follows:
4.1, randomly selecting one group of feature subsets as initial samples and the other group as newly added samples from two groups of feature subsets obtained by the rolling bearing time domain vibration signals through the steps 2 and 3;
step 4.2, inputting the initial sample into a classifier of a support vector machine to obtain a classifier phi1And a set of support vectors a 1;
inputting the newly added sample into a classifier of a support vector machine to obtain a misclassified sample set B;
step 4.3, inputting the support vector set A1 and the misclassification sample set B into a classifier of the support vector machine to obtain a new classifier phi2And a set of support vectors a 2;
step 4.4, taking the union of the support vector set A1 and the support vector set A2 as a new training set, comparing the new training set with the misclassification sample set B, taking the number set difference as a new misclassification sample set B, returning to the step 4.3, judging whether the new misclassification sample set B is empty, and outputting a new classifier phi if the new misclassification sample set B is empty2And supporting vector set A2, otherwise, returning to step 4.3;
step 4.5, passing a new classifier phi2And the support vector set A2 constructs a new single-class support vector machine.
The invention has the beneficial effects that:
the invention relates to a rolling bearing fault detection method based on an actual measurement signal, which considers that the rotating speed of a bearing of a wind turbine generator set can change under a variable working condition and the vibration signal of the bearing of the wind turbine generator set has the obvious nonlinear non-stationary characteristic under the variable working condition, and resamples a time domain vibration signal by adopting an order tracking technology to obtain an angular domain signal; parameter optimization is carried out on the variation modal decomposition by using a longicorn stigma search algorithm, Renyi entropy characteristics of IMF components are extracted to construct a characteristic subset, an Ocsvm increment algorithm is used for training a classifier, and the fault identification accuracy rate of the bearing vibration signal based on the variable working condition non-stable small sample is improved; the bearing fault detection method provided by the invention provides a new effective technical scheme for solving the problems that the bearing fault analysis is difficult due to unstable vibration signal fault characteristics, complex working conditions and less fault sample data of the fan rolling bearing under the variable working condition scene.
Drawings
FIG. 1 is a flow chart of a rolling bearing fault detection method based on measured signals according to the present invention;
FIG. 2(a) is a graph of a vibration signal sampled in the present invention;
FIG. 2(b) is a diagram of the resampling transform angular domain signal of the time domain signal of the order ratio tracking technique of the present invention;
FIG. 3 is a flow chart of optimization variational modal decomposition of the longicorn whisker algorithm of the present invention;
FIG. 4 is a diagram of a single-class support vector machine increment algorithm in the present invention;
FIG. 5 is a graph comparing the results of three tests in the example of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention provides a fan rolling bearing diagnosis method based on variable-working-condition non-stable normal small sample actual measurement signals, which aims to solve the problems that the fan rolling bearing vibration signal fault characteristics are unstable, the working conditions are complex, and the fault sample data are few under the variable-working-condition scene, so that the bearing fault analysis is difficult. Firstly, converting a rolling bearing fault time domain vibration signal into an angular domain by adopting an order ratio tracking technology; then, parameter optimization is carried out on Variation Modal Decomposition (VMD) by utilizing a celestial cow whisker search algorithm (BAS), vibration signals of all states of the rolling bearing are decomposed, a series of inherent modal functions are obtained, and when different faults occur to the bearing, frequency band energy in different inherent modal functions can change; secondly, extracting Renyi entropy characteristics from the modal components containing the main fault information, and constructing a characteristic subset; and finally, training by using the vibration signals in the normal state which are easy to obtain, extracting fault characteristic quantity, establishing a fault data sample and an incremental learning data sample, training by adopting a single-class support vector machine (ocsvm) incremental learning algorithm to obtain a fault identification model, accurately judging whether the rolling bearing has faults or not, and realizing fault early warning.
As shown in fig. 1, the specific process is as follows:
The specific process of the step 1 is as follows:
step 1.1, sampling a vibration signal and a rotating speed signal at equal time intervals delta t in two paths by a constant sampling rate, wherein the vibration signal is set as x (n), and the rotating speed signal is set as s (n);
step 1.2, calculating a time sequence t corresponding to the equal angle increment delta theta through a rotating speed signal s (n)ik(ii) a Firstly, searching the points of s (n) which satisfy the condition that s (n) is less than or equal to 0 and s (n +1) is greater than or equal to O in s (n), and assuming the corresponding time value tsnAnd tsn+1The waveform in between is linear;
the exact zero-crossing time tz of s (n) is determined from equation (1)iThe time value corresponding to each rotation of the reference shaft;
for time series tziThe difference is calculated to obtain the time T required by each rotation of the reference shaftiAssuming that the reference axis makes a uniform motion during one revolution and assuming the first two angular revolutionsThe accelerations are equal; and (3) calculating the equiangular sampling time t in the cycle by recursion in combination with a kinematic lawikAnd instantaneous speed of rotation vik；
Step 1.2, combining the law of kinematics, calculating the i-th-3, 4-th equiangular sampling time t in the week in a recursion mannerikAnd instantaneous speed of rotation vikThe specific process is as follows:
the initial velocity v of the first two weeks is estimated by equation set (2)10And angular acceleration a1Comprises the following steps:
setting equal-angle sampling M points of each rotation of a reference shaft, namely equal-angle increment delta theta is 2 pi/M; therefore, the angle of the k-th sampling point rotating corresponding to the reference axis is 2k pi/M, and the time value t corresponding to each equiangular sampling point in the first two weeks is obtained by the formula (3) and the formula (4)ikInstantaneous rotational speed vikAnd initial velocity v of week 330；
v30＝v10+a1(T1+T2) (4)
Wherein, the angle mark i represents the number of revolutions; the corner mark k represents the kth sample point in each revolution;
the angular acceleration a of the cycle of i 3,4, … is obtained by recursion of the formula (5)iAnd an initial velocity vi+1,0；
The equiangular sampling time t in the week can be recursively calculated according to equation (6), wherein the i-th is 3,4ikAnd instantaneous speed of rotation vik；
Step 1.3, according to the equal-angle sampling time value tik( i 1, 2..) interpolating the vibration signal x (n) to obtain the value t of the vibration signal at the equal angular sampling time tikCorresponding amplitude value, obtaining an equiangular resampling signal x (t) of the vibration signalik) I.e. the angular domain stationary signal x (t)ik) As shown in fig. 2 (b).
The basic principle of the order ratio tracking technology is that equal-angle sampling of M points is carried out in each rotation period of a rotating shaft, and rotating speed sampling of different periods can be obtained by changing the number M of the points, so 1/4 period rotating speed sampling commonly used in engineering is actually a special case of the order ratio tracking technology, and the order ratio tracking technology is more flexible than 1/4 period rotating speed sampling.
the specific process of the step 2 is as follows: by initializing the variational modal decomposition number K and the penalty factor alpha, setting a set of the decomposition number K and the penalty factor alpha as a search space of a search algorithm of a Tianniu searching place, setting a network error function MSF as a target optimization function, and iterating to obtain the optimal variational modal decomposition number K and the optimal penalty factor alpha; and carrying out self-adaptive decomposition on the angular domain stationary signals through a variational modal decomposition algorithm with the optimal variational modal decomposition number K and the optimal penalty factor alpha to obtain an inherent modal function component.
The specific process of obtaining the optimal variational modal decomposition number K and the optimal penalty factor alpha comprises the following steps:
step A, setting x as the central position of the longicorn, xleftAnd xrightThe two whiskers serving as the longicorn are respectively positioned on the left side and the right side of a central position x at a distance d; wherein, xleft，xrightAre all n-dimensional vectors;
step B, setting the direction of each previous step of the longicorn to be a random direction, namely the direction of the right beard pointing to the left beard has uncertainty; each n-dimensional unit random vector is generated, and dir represents the direction of the right whisker pointing to the left whisker:
in equation (7), rands (n,1) represents that an n-dimensional random vector is generated, and the positions of the left and right whiskers are:
xleft＝x(k)+dir·d(k)/2 (8)
xright＝x(k)-dir·d(k)/2 (9)
in formulas (8) and (9): x (k) represents the center position at the k-th iteration, and d (k) represents the two-whisker distance at the k-th iteration;
c, calculating the corresponding fitness values f of the two whiskers according to the current position and the target optimization function fleft，frightAnd determining the advancing direction of the next step according to the magnitude relation:
x(k+1)＝x(k)-S(k)·dir·sgn(fleft-fright) (10)
in equation (10), s (k) represents the advance step at the k-th iteration;
updating the decomposition number K and the penalty factor alpha according to the advancing direction;
and D, judging whether the optimization precision is met or is greater than the maximum iteration number, finishing the optimization if the optimization precision is met, outputting the optimal variational modal decomposition number K and the optimal penalty factor alpha, and otherwise, returning to the step B.
The Variational Modal Decomposition (VMD) can search the optimal solution of the constraint model, decompose the input signal into a series of modal components with sparse characteristics, and realize the self-adaptive decomposition of the signal. Assuming that each mode is a finite bandwidth with a center frequency, the center frequency and bandwidth are continuously updated during the decomposition process, the VMD decomposition is a mode function u seeking the minimum sum of K estimated bandwidthsk(t), the sum of the modes is the input signal f. The bandwidth of each modal function is determined by the following method:
1) to obtain an analytical signal of the mode functions, for each mode function uk(t) performing a Hilbert transform.
2) Estimating the center frequency e of each modal analysis signal-jwktMixing, modulating the spectrum of each mode to the corresponding fundamental band:
3) calculating the square L of the gradient of the above demodulated signal2And norm, namely estimating the bandwidth of each modal component. The corresponding constraint variational model expression is as follows:
wherein the content of the first and second substances,
the decomposition number K and the penalty factor alpha in the variation modal decomposition parameters need to be preset by experience, and if the setting is wrong, the over-decomposition or under-decomposition phenomenon can be caused. The BAS algorithm searches for an optimal solution by adopting a fixed step length, and the step length is multiplied by an attenuation coefficient in each iteration to improve the search precision. And setting the corresponding initial step size Si, the minimum step size S0i and the attenuation coefficient Wi according to the initialization parameters, wherein i is 1,2, …, n. In the iteration process, if the current step length can reduce the fitness value, which indicates that the current step length is effective, the step length is kept unchanged, the decomposition number K and the penalty factor a are efficiently optimized, and the BAS has the characteristics of low calculated amount, high speed and easiness in implementation, and is suitable for engineering application.
the formula for the Renyi entropy is:
in the formula, a is the order of entropy, and when a is 1, the order Renyi entropy is Shannon entropy; pi is the probability density (i ═ 1,2, …).
The Renyi entropy is a tool for measuring information uncertainty, and for signals with stronger randomness, the larger the uncertainty of the signals is, the larger the corresponding entropy value is; for signals with strong certainty, the corresponding entropy value is also small. Corresponding to the time-frequency analysis result, the more concentrated the energy distribution is, the smaller the uncertainty is, and the smaller the corresponding entropy value is; the more spread its energy, the greater the uncertainty and the greater the corresponding entropy. The degree of the time-frequency spectrum energy convergence can be judged by utilizing the size of the entropy value, and the quality of a time-frequency analysis result is further judged.
the precondition for the incremental algorithm training Ocsvm is that there are a historical dataset a and an incremental sample B and it is assumed that these two datasets fulfill the condition a n B ═ Φ, Φ1And a1 are the initial SVM classifier on data set a and the corresponding set of support vectors, respectively. Then, an SVM classifier phi and a corresponding support vector set A based on the sample set A ═ B ═ phi are searched.
The specific process of the step 4 is as follows:
4.1, randomly selecting one group of feature subsets as initial samples and the other group as newly added samples from two groups of feature subsets obtained by the rolling bearing time domain vibration signals through the steps 2 and 3;
step 4.2, inputting the initial sample into a classifier of a support vector machine to obtain a classifier phi1And a set of support vectors a 1;
inputting the newly added sample into a classifier of a support vector machine to obtain a misclassified sample set B;
step 4.3, inputting the support vector set A1 and the misclassification sample set B into a classifier of the support vector machine to obtain a new classifier phi2And a set of support vectors a 2;
step (ii) of4.4, taking the union of the support vector set A1 and the support vector set A2 as a new training set, comparing the new training set with the misclassification sample set B, taking the number set difference as a new misclassification sample set B, returning to the step 4.3, judging whether the new misclassification sample set B is empty, and outputting a new classifier phi if the new misclassification sample set B is empty2And supporting vector set A2, otherwise, returning to step 4.3;
step 4.5, passing a new classifier phi2And the support vector set A2 constructs a new single-class support vector machine.
The flow chart of the single-class support vector machine increment algorithm is shown in figure 4.
And 5, adopting a new single-type support vector machine to carry out fault identification on the bearing vibration signal.
Examples
And constructing a bearing fault vibration signal test set sample by adopting actual fan bearing fault vibration data, and identifying the fault state by using the rolling bearing fault detection method disclosed by the invention. Experiments prove that the method not only can accurately identify the fault state of the fan bearing, but also can effectively identify unknown fault type data of a non-training sample.
The single-class support vector machine can complete training only by using the negative samples, and can accurately identify the non-negative samples. Assume that there are samples ziAnd i is 1,2, … N, the kernel function psi is mapped to a high-dimensional feature space, so that the feature space has better aggregation, and an optimal hyperplane can be solved in the feature space to realize the maximum separation of the target data from the coordinate origin. The decision function fsign (z) ═ sign (u · Ψ (z) - ρ) is to separate the sample set used for training as much as possible from the origin to maximize the distance between the hyperplane and the origin. The weight u and threshold ρ of the support vector can be described by solving the following quadratic programming problem:
s.t.(u·ψ(z))＞＞ρ-ξi，ξi≥0
wherein v is equal to (0,1) and is used for controlling the proportion of the support vector in the training sample; xiiAfter introducing the kernel function for relaxing the variable, the above problem can be converted into a dual problem
In the OCSVM, the OCSVM is,
to determine the threshold, a separate hyperplane is determined from the weight vector u, where k (z)i,zj) As a kernel function of the input space, αi、αzAre variables to be solved.
In the invention, the difficult problems of unstable fault characteristics, complex working conditions and less fault sample data of the vibration signal of the rolling bearing of the fan under the variable working condition scene are considered, firstly, the step ratio tracking technology is adopted to convert the vibration signal of the time domain of the fault of the rolling bearing into an angular domain; then, parameter optimization is carried out on Variation Modal Decomposition (VMD) by utilizing a celestial cow whisker search algorithm (BAS), vibration signals of all states of the rolling bearing are decomposed, a series of inherent modal functions are obtained, and when different faults occur to the bearing, frequency band energy in different inherent modal functions can change; secondly, extracting Renyi entropy characteristics from the modal components containing the main fault information, and constructing a characteristic subset; and finally, training by using a vibration signal of the bearing fault state, extracting fault characteristic quantity, establishing a fault data sample and an incremental learning data sample, training by using a one-class support vector machine (ocsvm) to obtain a fault identification model, accurately judging whether the rolling bearing has a fault, and realizing fault early warning.
The invention adopts a reference data set for monitoring the vibration state of the wind driven generator gearbox provided by NREL. The study of NREL condition monitoring was based on a GRC dynamometer, with data collected from identically designed "healthy" and "damaged" gearboxes by GRC dynamometer testing. Wherein vibration data is collected by an accelerometer together with a high speed axis RPM (rotational speed) signal table, data is collected at 40khz per channel using a national instrumentation PXI-4472b high speed data acquisition system (DAQ), and generator speed is recorded in addition to accelerometer data. Considering that the rotating speed of a gearbox bearing is high, the sampling frequency is determined to be 5kHZ, data separated by 8 sampling points form a new data set, and the sampling system samples about 160 points every time the gearbox high-speed bearing runs for one circle. Wherein, each fault sample is formed by adopting vibration signals of 800 sampling points in about 4 rotation periods. The training set comprises 2000 groups including four bearing fault states, and the test set comprises 500 groups including bearing fault states and normal states.
According to the determination of actual conditions, the optimization intervals of K and alpha are respectively [0,10] and [0,5000], an optimized parameter is found for variational modal decomposition by utilizing a longicorn whisker search algorithm, the maximum iteration frequency is set for 1000 times, and the error evaluation standard is the mean square error. The experimental results are analyzed, and finally, the number K of the components is 6 and the penalty factor alpha is 2022 are determined. The flow chart of optimization variational modal decomposition of the longicorn whisker search algorithm is shown in fig. 3.
The empirical mode decomposition method and the local mean mode decomposition signal processing method both have the performance of self-adaptive signal decomposition, and in order to verify that the improved variational mode decomposition method adopted by the invention has better decomposition performance, the three methods are respectively combined with a single-class support vector machine to be used for bearing fault state identification, and a comparison experiment is carried out to verify the effectiveness of the invention. To reduce the effect of random perturbations, each method was repeated 15 times. The results of comparing the accuracy of the experimental classification are shown in fig. 5.
Through comparison of experimental results, the bearing fault diagnosis method based on the longicorn whisker search algorithm optimization variational modal decomposition and the single-class support vector machine is adopted, the accuracy of bearing fault state identification reaches 97.24, and the accuracy of bearing fault state identification reaches 95.04 and 94.30 respectively based on the empirical modal decomposition and the single-class support vector machine and the fault diagnosis method based on the local mean modal decomposition and the single-class support vector machine.
Experiments prove that compared with the conventional bearing fault state diagnosis method, the bearing fault state diagnosis method adopted by the invention has the advantage that the bearing fault state identification accuracy is improved. The invention provides an effective technical scheme aiming at the problems that the bearing fault analysis is difficult due to unstable vibration signal fault characteristics, complex working conditions and less fault sample data of a fan rolling bearing under a variable working condition scene.
Claims (7)
1. A rolling bearing fault detection method based on measured signals is characterized by comprising the following steps:
step 1, acquiring a time domain vibration signal of a rolling bearing, and converting the time domain vibration signal into an angular domain stationary signal by adopting an order ratio tracking technology;
step 2, carrying out parameter optimization on a variational modal decomposition algorithm by using a longicorn stigma search algorithm, and decomposing steady signals of angular domains of each state of the rolling bearing to obtain an inherent modal function component;
step 3, extracting Renyi entropy characteristics from the inherent mode function component according to a formula of Renyi entropy, and constructing a characteristic subset;
step 4, establishing a fault data sample by using the extracted feature subset, training by using the increment algorithm of the single-class support vector machine, and establishing a new single-class support vector machine;
and 5, adopting a new single-type support vector machine to carry out fault identification on the bearing vibration signal.
2. The rolling bearing fault detection method based on the measured signal according to claim 1, wherein the specific process of the step 1 is as follows:
step 1.1, sampling a vibration signal and a rotating speed signal at equal time intervals delta t in two paths by a constant sampling rate, wherein the vibration signal is set as x (n), and the rotating speed signal is set as s (n);
step 1.2, calculating a time sequence t corresponding to the equal angle increment delta theta through a rotating speed signal s (n)ik(ii) a Firstly, searching the points of s (n) which satisfy the condition that s (n) is less than or equal to 0 and s (n +1) is greater than or equal to O in s (n), and assuming the corresponding time value tsnAnd tsn+1The waveform in between is linear;
obtaining s from equation (1)(n) exact zero crossing time tziThe time value corresponding to each rotation of the reference shaft;
for time series tziThe difference is calculated to obtain the time T required by each rotation of the reference shaftiAssuming that the reference axis does a uniform variable speed motion within one revolution, and assuming that the angular accelerations of the first two revolutions are equal; and (3) calculating the equiangular sampling time t in the cycle by recursion in combination with a kinematic lawikAnd instantaneous speed of rotation vik；
Step 1.3, according to the equal-angle sampling time value tik(i 1, 2..) interpolating the vibration signal x (n) to obtain the value t of the vibration signal at the equal angular sampling time tikCorresponding amplitude value, obtaining an equiangular resampling signal x (t) of the vibration signalik) I.e. the angular domain stationary signal x (t)ik)。
3. The rolling bearing fault detection method based on the measured signal according to claim 2, wherein the i-th-3, 4-th-in-week equiangular sampling time t is calculated in a recursive manner in combination with the kinematics law in step 1.2ikAnd instantaneous speed of rotation vikThe specific process is as follows:
the initial velocity v of the first two weeks is estimated by equation set (2)10And angular acceleration a1Comprises the following steps:
setting equal-angle sampling M points of each rotation of a reference shaft, namely equal-angle increment delta theta is 2 pi/M; therefore, the angle of the k-th sampling point rotating corresponding to the reference axis is 2k pi/M, and the time value t corresponding to each equiangular sampling point in the first two weeks is obtained by the formula (3) and the formula (4)ikInstantaneous rotational speed vikAnd initial velocity v of week 330；
v30＝v10+a1(T1+T2) (4)
Wherein, the angle mark i represents the number of revolutions; the corner mark k represents the kth sample point in each revolution;
the angular acceleration a of the cycle of i 3,4, … is obtained by recursion of the formula (5)iAnd an initial velocity vi+1,0；
The equiangular sampling time t in the week can be recursively calculated according to equation (6), wherein the i-th is 3,4ikAnd instantaneous speed of rotation vik；
4. The rolling bearing fault detection method based on the measured signal according to claim 1, wherein the step 2 comprises the following specific processes: by initializing the variational modal decomposition number K and the penalty factor alpha, setting a set of the decomposition number K and the penalty factor alpha as a search space of a search algorithm of a Tianniu searching place, setting a network error function MSF as a target optimization function, and iterating to obtain the optimal variational modal decomposition number K and the optimal penalty factor alpha; and carrying out self-adaptive decomposition on the angular domain stationary signals through a variational modal decomposition algorithm with the optimal variational modal decomposition number K and the optimal penalty factor alpha to obtain an inherent modal function component.
5. The rolling bearing fault detection method based on the measured signal according to claim 4, wherein the specific process of obtaining the optimal variational modal decomposition number K and the optimal penalty factor α is as follows:
step A, setting x as the central position of the longicorn, xleftAnd xrightThe two whiskers serving as the longicorn are respectively positioned on the left side and the right side of a central position x at a distance d; wherein, xleft，xrightAre all n-dimensional vectors;
step B, setting the direction of each previous step of the longicorn to be a random direction, namely the direction of the right beard pointing to the left beard has uncertainty; each n-dimensional unit random vector is generated, and dir represents the direction of the right whisker pointing to the left whisker:
in equation (7), rands (n,1) represents that an n-dimensional random vector is generated, and the positions of the left and right whiskers are:
xleft＝x(k)+dir·d(k)/2 (8)
xright＝x(k)-dir·d(k)/2 (9)
in formulas (8) and (9): x (k) represents the center position at the k-th iteration, and d (k) represents the two-whisker distance at the k-th iteration;
c, calculating the corresponding fitness values f of the two whiskers according to the current position and the target optimization function fleft，frightAnd determining the advancing direction of the next step according to the magnitude relation:
x(k+1)＝x(k)-S(k)·dir·sgn(fleft-fright) (10)
in equation (10), s (k) represents the advance step at the k-th iteration;
updating the decomposition number K and the penalty factor alpha according to the advancing direction;
and D, judging whether the optimization precision is met or is greater than the maximum iteration number, finishing the optimization if the optimization precision is met, outputting the optimal variational modal decomposition number K and the optimal penalty factor alpha, and otherwise, returning to the step B.
6. The rolling bearing fault detection method based on the measured signal according to claim 4, wherein the formula of Renyi entropy is as follows:
in the formula, a is the order of entropy, and when a is 1, the order Renyi entropy is Shannon entropy; pi is the probability density (i ═ 1,2, …).
7. The rolling bearing fault detection method based on the measured signal according to claim 1, wherein the specific process of the step 4 is as follows:
4.1, randomly selecting one group of feature subsets as initial samples and the other group as newly added samples from two groups of feature subsets obtained by the rolling bearing time domain vibration signals through the steps 2 and 3;
step 4.2, inputting the initial sample into a classifier of a support vector machine to obtain a classifier phi1And a set of support vectors a 1;
inputting the newly added sample into a classifier of a support vector machine to obtain a misclassified sample set B;
step 4.3, inputting the support vector set A1 and the misclassification sample set B into a classifier of the support vector machine to obtain a new classifier phi2And a set of support vectors a 2;
step 4.4, taking the union of the support vector set A1 and the support vector set A2 as a new training set, comparing the new training set with the misclassification sample set B, taking the number set difference as a new misclassification sample set B, returning to the step 4.3, judging whether the new misclassification sample set B is empty, and outputting a new classifier phi if the new misclassification sample set B is empty2And supporting vector set A2, otherwise, returning to step 4.3;
step 4.5, passing a new classifier phi2And the support vector set A2 constructs a new single-class support vector machine.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011217106.9A CN112414713A (en) | 2020-11-04 | 2020-11-04 | Rolling bearing fault detection method based on measured signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011217106.9A CN112414713A (en) | 2020-11-04 | 2020-11-04 | Rolling bearing fault detection method based on measured signals |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112414713A true CN112414713A (en) | 2021-02-26 |
Family
ID=74828530
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011217106.9A Pending CN112414713A (en) | 2020-11-04 | 2020-11-04 | Rolling bearing fault detection method based on measured signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112414713A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113418685A (en) * | 2021-06-01 | 2021-09-21 | 中广核工程有限公司 | Vibration monitoring device and fault analysis system and method based on vibration monitoring |
| CN113625164A (en) * | 2021-08-02 | 2021-11-09 | 南京航空航天大学 | Aviation generator fault feature extraction method, system, medium and computing device |
| CN113670615A (en) * | 2021-08-23 | 2021-11-19 | 洛阳轴承研究所有限公司 | Bearing unit vibration testing method and system |
| CN114238038A (en) * | 2022-03-01 | 2022-03-25 | 湖南云箭智能科技有限公司 | Board card temperature monitoring method, device, equipment and readable storage medium |
| CN114722876A (en) * | 2022-04-21 | 2022-07-08 | 沈阳顺义科技有限公司 | State prediction and mode classification method and system for rotary machine |
| CN114897032A (en) * | 2022-07-12 | 2022-08-12 | 武汉格蓝若智能技术有限公司 | Current transformer fault diagnosis method and device based on width learning |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2226621A1 (en) * | 2009-03-05 | 2010-09-08 | Tetra Laval Holdings & Finance SA | Predictive rolling bearing maintenance |
| CN103868694A (en) * | 2014-03-26 | 2014-06-18 | 东南大学 | Embedded variable-rotation-speed bearing fault diagnosis device |
| CN105758644A (en) * | 2016-05-16 | 2016-07-13 | 上海电力学院 | Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy |
| CN105784366A (en) * | 2016-03-30 | 2016-07-20 | 华北电力大学（保定） | Wind turbine generator bearing fault diagnosis method under variable speed |
| CN107941513A (en) * | 2017-12-25 | 2018-04-20 | 北京建筑大学 | A kind of time-frequency Computed order tracking method of train EEF bogie bearing non-stationary O＆M |
| CN108375472A (en) * | 2018-02-12 | 2018-08-07 | 武汉科技大学 | Based on the Method for Bearing Fault Diagnosis and system and device for improving experience wavelet transformation |
| US20180335366A1 (en) * | 2017-05-19 | 2018-11-22 | Nutech Ventures | Detecting Faults in Wind Turbines |
| CN111008648A (en) * | 2019-11-12 | 2020-04-14 | 国网湖南省电力有限公司 | Fault identification method, system and medium for pure optical fiber electronic current transformer |
| CN111307460A (en) * | 2020-03-14 | 2020-06-19 | 中国石化销售股份有限公司华南分公司 | Rolling bearing fault diagnosis method based on calculation order tracking and spectral kurtosis |
| CN111855208A (en) * | 2020-07-02 | 2020-10-30 | 南京航空航天大学 | Rotary machine weak fault signal extraction method based on order analysis and sparse coding |
-
2020
- 2020-11-04 CN CN202011217106.9A patent/CN112414713A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2226621A1 (en) * | 2009-03-05 | 2010-09-08 | Tetra Laval Holdings & Finance SA | Predictive rolling bearing maintenance |
| CN103868694A (en) * | 2014-03-26 | 2014-06-18 | 东南大学 | Embedded variable-rotation-speed bearing fault diagnosis device |
| CN105784366A (en) * | 2016-03-30 | 2016-07-20 | 华北电力大学（保定） | Wind turbine generator bearing fault diagnosis method under variable speed |
| CN105758644A (en) * | 2016-05-16 | 2016-07-13 | 上海电力学院 | Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy |
| US20180335366A1 (en) * | 2017-05-19 | 2018-11-22 | Nutech Ventures | Detecting Faults in Wind Turbines |
| CN107941513A (en) * | 2017-12-25 | 2018-04-20 | 北京建筑大学 | A kind of time-frequency Computed order tracking method of train EEF bogie bearing non-stationary O＆M |
| CN108375472A (en) * | 2018-02-12 | 2018-08-07 | 武汉科技大学 | Based on the Method for Bearing Fault Diagnosis and system and device for improving experience wavelet transformation |
| CN111008648A (en) * | 2019-11-12 | 2020-04-14 | 国网湖南省电力有限公司 | Fault identification method, system and medium for pure optical fiber electronic current transformer |
| CN111307460A (en) * | 2020-03-14 | 2020-06-19 | 中国石化销售股份有限公司华南分公司 | Rolling bearing fault diagnosis method based on calculation order tracking and spectral kurtosis |
| CN111855208A (en) * | 2020-07-02 | 2020-10-30 | 南京航空航天大学 | Rotary machine weak fault signal extraction method based on order analysis and sparse coding |
Non-Patent Citations (5)
| Title |
|---|
| HENG-DI WANG等: "Parameter-Adaptive VMD Method Based on BAS Optimization Algorithm for Incipient Bearing Fault Diagnosis", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
| MING ZENG等: "One-class classification based on the convex hull for bearing fault detection", 《MECHANICAL SYSTEMS AND SIGNAL PROCESSING》 * |
| 张亢等: "基于LMD和阶次跟踪分析的滚动轴承故障诊断", 《振动.测试与诊断》 * |
| 耿蒲龙等: "基于SVM增量学习算法的煤矿高压断路器故障模式识别方法", 《煤炭学报》 * |
| 黄建校等: "一种改进的SVM增量学习算法研究", 《无线互联科技》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113418685A (en) * | 2021-06-01 | 2021-09-21 | 中广核工程有限公司 | Vibration monitoring device and fault analysis system and method based on vibration monitoring |
| CN113625164A (en) * | 2021-08-02 | 2021-11-09 | 南京航空航天大学 | Aviation generator fault feature extraction method, system, medium and computing device |
| CN113670615A (en) * | 2021-08-23 | 2021-11-19 | 洛阳轴承研究所有限公司 | Bearing unit vibration testing method and system |
| CN114238038A (en) * | 2022-03-01 | 2022-03-25 | 湖南云箭智能科技有限公司 | Board card temperature monitoring method, device, equipment and readable storage medium |
| CN114722876A (en) * | 2022-04-21 | 2022-07-08 | 沈阳顺义科技有限公司 | State prediction and mode classification method and system for rotary machine |
| CN114897032A (en) * | 2022-07-12 | 2022-08-12 | 武汉格蓝若智能技术有限公司 | Current transformer fault diagnosis method and device based on width learning |
| CN114897032B (en) * | 2022-07-12 | 2022-10-11 | 武汉格蓝若智能技术有限公司 | Current transformer fault diagnosis method and device based on width learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112414713A (en) | Rolling bearing fault detection method based on measured signals | |
| Huo et al. | Incipient fault diagnosis of roller bearing using optimized wavelet transform based multi-speed vibration signatures | |
| Kang et al. | Time-varying and multiresolution envelope analysis and discriminative feature analysis for bearing fault diagnosis | |
| Chen et al. | Multi-fault diagnosis study on roller bearing based on multi-kernel support vector machine with chaotic particle swarm optimization | |
| Liu et al. | Flexible generalized demodulation for intelligent bearing fault diagnosis under nonstationary conditions | |
| Gharavian et al. | Comparison of FDA-based and PCA-based features in fault diagnosis of automobile gearboxes | |
| CN108051189A (en) | A kind of rotary machine fault characteristic extraction method and device | |
| CN110146279B (en) | Ship shafting unbalance fault diagnosis method based on vector evidence reasoning | |
| Deng et al. | Fault feature extraction of a rotor system based on local mean decomposition and Teager energy kurtosis | |
| Sun et al. | Composite-graph-based sparse subspace clustering for machine fault diagnosis | |
| CN111307452A (en) | Intelligent fault diagnosis method for rotating machinery at time-varying rotating speed | |
| Prosvirin et al. | An improved algorithm for selecting IMF components in ensemble empirical mode decomposition for domain of rub-impact fault diagnosis | |
| Dong et al. | Research on diagnosis method of centrifugal pump rotor faults based on IPSO-VMD and RVM | |
| Küppers et al. | Random forest on an embedded device for real-time machine state classification | |
| CN114298110B (en) | Rolling bearing fault diagnosis method and system based on interpretable 1DCNN model | |
| Zemouri et al. | Recent research and applications in variational autoencoders for industrial prognosis and health management: A survey | |
| Fritzen et al. | Vibration-based damage detection under changing environmental and operational conditions | |
| CN112818819B (en) | AUV state monitoring method based on dynamic model and complex network theory | |
| Zhao et al. | Research on constructing a degradation index and predicting the remaining useful life for rolling element bearings of complex equipment | |
| Jeon et al. | Statistical approach to diagnostic rules for various malfunctions of journal bearing system using fisher discriminant analysis | |
| Geng et al. | Fault identification of rolling bearing with variable speed based on generalized broadband mode decomposition and DET | |
| Ali et al. | Experimental investigation of machine learning based fault diagnosis for induction motors | |
| Ali et al. | Machine learning based fault diagnosis for single-and multi-faults for induction motors fed by variable frequency drives | |
| Zhi et al. | Anti-noise motor fault diagnosis method based on decision tree and the feature extraction methods in the time domain and frequency domain | |
| CN115795341A (en) | Two-dimensional piston pump health state assessment method based on variable rotating speed |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210226 |