CN113792397A - Rotary machine gear box fault diagnosis method based on deep learning - Google Patents

Rotary machine gear box fault diagnosis method based on deep learning Download PDF

Info

Publication number
CN113792397A
CN113792397A CN202111023880.0A CN202111023880A CN113792397A CN 113792397 A CN113792397 A CN 113792397A CN 202111023880 A CN202111023880 A CN 202111023880A CN 113792397 A CN113792397 A CN 113792397A
Authority
CN
China
Prior art keywords
matrix
signal
deep learning
fault diagnosis
dimensional
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
Application number
CN202111023880.0A
Other languages
Chinese (zh)
Inventor
张能文
杨凯铭
秦法涛
何晓琳
吴广
崔飞
蔡昌春
江冰
承敏钢
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Xdg Solutions Technology Co ltd
Original Assignee
Jiangsu Xdg Solutions Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangsu Xdg Solutions Technology Co ltd filed Critical Jiangsu Xdg Solutions Technology Co ltd
Priority to CN202111023880.0A priority Critical patent/CN113792397A/en
Publication of CN113792397A publication Critical patent/CN113792397A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • 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
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/021—Gearings
    • 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
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/028—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/25—Design optimisation, verification or simulation using particle-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/10—Noise analysis or noise optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Geometry (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Optimization (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Pure & Applied Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a rotary machine gear box fault diagnosis method based on deep learning, which belongs to the technical field of rotary machine equipment fault type identification and diagnosis, and comprises the steps of collecting signals such as running vibration acceleration, speed, displacement, frequency and the like of a rotary machine gear box, denoising vibration signals based on singular value decomposition to realize original signal reconstruction, mapping a one-dimensional time sequence form of the vibration signals to a two-dimensional space by using a symmetric dot matrix image analysis method to construct two-dimensional image information, and extracting fault type characteristics based on the density degree of discrete points; and finally, fault training and discrimination are carried out by combining two-dimensional image information and a deep neural network which are constructed by analyzing the symmetrical dot matrix image to obtain a fault diagnosis model.

Description

Rotary machine gear box fault diagnosis method based on deep learning
Technical Field
The invention belongs to the technical field of fault type identification and diagnosis of rotary mechanical equipment, and particularly relates to a fault diagnosis method of a rotary mechanical gearbox based on deep learning.
Background
The rotating machinery is an important device for realizing energy conversion, is used in a large amount in the industrial field, has important relation between the reliability and the safety of operation, directly influences industrial production due to equipment faults, and even threatens the personal safety of workers at the same time, so that the strengthening of the detection of the running state of the rotating machinery equipment and the identification of fault types is very important. With the development of deep learning and big data artificial intelligence technology, a technical support is provided for fault diagnosis of mechanical equipment, how to judge the running state and the development trend of the rotary equipment by using the artificial intelligence technology, analyze the fault reason, the position and the property of the rotary equipment and provide scientific inspection basis for industrial production, thereby effectively reducing the maintenance and the maintenance cost of the equipment, improving the economical efficiency of the equipment operation and ensuring the safe production.
Disclosure of Invention
1. Technical problem to be solved by the invention
The object of the present invention is to solve the above mentioned drawbacks.
2. Technical scheme
In order to achieve the purpose, the technical scheme provided by the invention is as follows:
the invention discloses a rotary machine gearbox fault diagnosis method based on deep learning, which comprises the following steps:
s1, collecting a vibration signal of the operation of the rotary mechanical gear box;
s2, carrying out singular value decomposition on the vibration signal to obtain an original signal;
s3, mapping the one-dimensional time sequence form of the original signal to a two-dimensional space by using a symmetric dot matrix image analysis method to construct two-dimensional image information;
s4, extracting fault type features according to the density degree of discrete points of the two-dimensional image information;
and S5, carrying out fault training and discrimination by using the extracted fault type to carry out deep neural network.
Preferably, the vibration signal includes rotational acceleration, rotational speed, displacement, and frequency.
Preferably, the singular value decomposition method in step S2 is as follows:
s21, constructing a one-dimensional characteristic matrix Hankel matrix of the vibration acceleration time sequence signal, setting an original noise-containing signal sequence as X ═ X (1), X (2),. X (N), and constructing the Hankel matrix as follows:
Figure BDA0003242586780000021
wherein q < N and satisfies N ═ p + q-1, Ap×qIs a true signal feature matrix in the original time sequence signal, Bp×qConstructing Hankel moment for characteristic matrix formed by noise matrix sequenceAnd in the process of array, the number p of rows is equal to the number q of columns, the time sequence of even data points is taken, and even processing is carried out if the sequence is an odd number.
S22, carrying out singular value decomposition on the Hankel matrix, and expressing the Hankel matrix as the product of two orthogonal matrices and an asymmetric diagonal matrix:
Figure BDA0003242586780000022
wherein,
Figure BDA0003242586780000023
and Vq×qP-order and q-order orthogonal matrixes respectively; mp×qIs an asymmetric diagonal matrix;
Figure BDA0003242586780000024
σiis the singular value of the matrix Hankel, and r is the rank of the matrix.
S23, selecting L effective singular values, setting the rest singular values as 0, and reconstructing an inverse matrix in the singular value decomposition process to obtain a reconstructed matrix H ', wherein the H' is a Hankel matrix of the denoised vibration signal, and the method comprises the following steps:
Figure BDA0003242586780000025
s24, carrying out inverse singular decomposition process on the H' matrix to obtain the vibration acceleration signal of the gearbox after noise reduction
X′＝[x′(1)，x′(2)，...x′(N)]。
Preferably, in step S23, the determination of the L value is performed by using an inverse singular value averaging method, where the obtained singular values are arranged in a descending manner:
K＝rank(σ1，σ2，...，σm)
definition of
Figure BDA0003242586780000031
And comparing the b with all singular values, considering the singular value larger than the mean value as an effective singular value, setting other singular values as 0, and performing signal reconstruction.
Preferably, the specific content of step S3 is to map the one-dimensional time-series vibration deceleration signal after noise reduction into a two-dimensional space by using a symmetric image lattice analysis method, form a related symmetric pattern in the two-dimensional space, and perform feature extraction of the gear vibration acceleration signal of the rotating equipment by using the petal signal density, where the vibration acceleration values corresponding to the time-series i and i + l signals are x '(i) and x' (i + l), the signals at these two times are mapped into the two-dimensional space by using a symmetric lattice transformation, and the polar coordinates are expressed as: z (r (i), θ (i), φ (i)), the mapping equation is as follows:
Figure BDA0003242586780000032
wherein r is the radius of the polar coordinate pattern, and theta is the angle of the petals deflected along the mirror image of the symmetrical plane in the counterclockwise direction; phi is a mirror image of the petals in the opposite direction clockwise; x is the number ofmaxIs the maximum amplitude in the signal sequence; x is the number ofminIs the minimum amplitude in the signal sequence; l is a time interval parameter; theta is a corresponding symmetrical angle of the petals when the petals are turned in a mirror image manner,
Figure BDA0003242586780000033
to enlarge factor
Figure BDA0003242586780000034
Preferably, the first and second liquid crystal materials are,
Figure BDA0003242586780000035
the selection of the values l and theta is optimized by improving the particle swarm optimization, an optimization model with the two-dimensional image discrimination index as a target function is constructed, and optimization is carried out
Figure BDA0003242586780000036
l and theta, and the distinguishing indexes comprise: the thickness difference of the petal arms of the two-dimensional patterns and the included angle of the quasi-regression sidelines of the two adjacent petal arms are represented, and the objective function is represented by the functional of the corresponding points as follows:
Figure BDA0003242586780000041
wherein, | | | is functional calculation, and N is the number of sequences.
Preferably, the particle swarm algorithm is improved, and the influence of the average fitness position of the particle swarm on the position of the next generation of particles is increased in the particle swarm position formula, so that the convergence speed of the system is improved, wherein the position formula is as follows:
vi(k)＝ωvi(k-1)+c1r1[pi，best(k-1)-xi(k-1)]
+c2r2[gbest-xi(k-1)]+c3r3[xavg(k-1)-xi(k-1)]
xi(k)＝vi(k)+xi(k-1)
where ω is the inertial weight of the flight velocity, viIs the particle velocity, xiIs the position of the particle, xavgIs the average position of the particles; r is1，r2，r3，c1，c1，c3The values of the random numbers are (0-1), (0-2) and (0-2), and k is the number of iterations.
Preferably, in step S5, a neural network training model is constructed from optimization of four main parameters, i.e., the iteration number, batch size, convolution kernel size, and convolution kernel coefficient of the neural network, according to the vibration signal characteristics of the rotating machine gearbox.
Preferably, the network calculation is used for optimizing parameters such as the iteration number, the batch size, the convolution kernel size, the number of convolution kernels and the like of the neural network, and the network calculation is as follows:
Figure BDA0003242586780000042
setting the actual output O of the network, and the expected outputs Y and E as the difference, wherein k is iteration number, omega is batch size, b is convolution kernel size, beta convolution kernel number, and k, omega, b and beta satisfy the following relations:
Figure BDA0003242586780000043
wherein eta is1，η2，η3，η4The step size adaptive adjustment variables are random numbers of (0, 1).
Preferably, the failure types include, but are not limited to, rotating machine gearbox gear missing teeth, broken teeth, wear, tooth root cracking, bearing inner race failure, outer race, balls, mixed failures, and the like.
3. Advantageous effects
Compared with the prior art, the technical scheme provided by the invention has the following beneficial effects:
(1) according to the rotating machinery gearbox fault diagnosis method based on deep learning, a fault diagnosis model can be obtained, and then fault diagnosis and fault recognition can be accurately and effectively carried out on gears and bearings in a direct gearbox.
Drawings
FIG. 1 is a flow chart of a method for deep learning based fault diagnosis of a rotating machine gearbox in accordance with the present invention;
FIG. 2 is a noise reduction effect diagram of a rotating machine gearbox fault diagnosis method based on deep learning according to the present invention;
FIG. 3 is a diagram of a gearbox fault-free signal collected by a deep learning-based rotating machine gearbox fault diagnosis method of the present invention;
FIG. 4 is a plot of gear box root crack signals collected for a deep learning based rotating machine gear box fault diagnostic method of the present invention;
FIG. 5 is a diagram of collected gearbox gear wear signals for a deep learning based rotating machine gearbox fault diagnosis method of the present invention;
FIG. 6 is a diagram of gear-box gear tooth-breaking signals collected by the rotating machinery gear-box fault diagnosis method based on deep learning of the present invention;
FIG. 7 is a diagram of collected gear missing signals of a gearbox according to a method for diagnosing faults of a rotary mechanical gearbox based on deep learning according to the present invention;
FIG. 8 is a two-dimensional map of a rotating machine gearbox fault diagnosis method based on deep learning according to the present invention under normal conditions of the gearbox;
FIG. 9 is a two-dimensional map of missing gear of a gear box according to a method for diagnosing faults of a rotary mechanical gear box based on deep learning according to the present invention;
FIG. 10 is a two-dimensional map of a gear box gear tooth breakage of a rotary mechanical gear box fault diagnosis method based on deep learning according to the present invention;
FIG. 11 is a two-dimensional map of gear box gear wear for a rotating machine gear box fault diagnosis method based on deep learning of the present invention;
FIG. 12 is a two-dimensional map of gearbox root cracking under a rotary machine gearbox fault diagnosis method based on deep learning of the present invention.
Detailed Description
In order to facilitate an understanding of the invention, the invention will now be described more fully hereinafter with reference to the accompanying drawings, in which several embodiments of the invention are shown, but which may be embodied in many different forms and are not limited to the embodiments described herein, but rather are provided for the purpose of providing a more thorough disclosure of the invention.
It will be understood that when an element is referred to as being "secured to" another element, it can be directly on the other element or intervening elements may also be present; when an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements may also be present; the terms "vertical," "horizontal," "left," "right," and the like as used herein are for illustrative purposes only.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention; as used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
Example 1
Referring to fig. 1, the method for diagnosing faults of a rotary machine gearbox based on deep learning of the embodiment includes the following steps:
s1, collecting a vibration signal of the operation of the rotary mechanical gear box;
s2, carrying out singular value decomposition on the vibration signal to obtain an original signal;
s3, mapping the one-dimensional time sequence form of the original signal to a two-dimensional space by using a symmetric dot matrix image analysis method to construct two-dimensional image information;
s4, extracting fault type features according to the density degree of discrete points of the two-dimensional image information;
and S5, carrying out fault training and discrimination by using the extracted fault type to carry out deep neural network.
The vibration signal of the present embodiment includes rotational acceleration, rotational speed, displacement, and frequency.
The singular value decomposition method in step S2 of the present embodiment is as follows:
s21, constructing a one-dimensional characteristic matrix Hankel matrix of the vibration acceleration time sequence signal, setting an original noise-containing signal sequence as X ═ X (1), X (2),. X (N), and constructing the Hankel matrix as follows:
Figure BDA0003242586780000071
wherein q < N and satisfies N ═ p + q-1, Ap×qIs a true signal feature matrix in the original time sequence signal, Bp×qAnd when constructing a Hankel matrix, the number p of rows is equal to the number q of columns, the time sequence of even data points is taken, and even processing is carried out if the sequence is an odd number.
S22, carrying out singular value decomposition on the Hankel matrix, and expressing the Hankel matrix as the product of two orthogonal matrices and an asymmetric diagonal matrix:
Figure BDA0003242586780000072
wherein,
Figure BDA0003242586780000073
and Vq×qP-order and q-order orthogonal matrixes respectively; mp×qIs an asymmetric diagonal matrix;
Figure BDA0003242586780000074
σiis the singular value of the matrix Hankel, and r is the rank of the matrix.
S23, selecting L effective singular values, setting the rest singular values as 0, and reconstructing an inverse matrix in the singular value decomposition process to obtain a reconstructed matrix H ', wherein the H' is a Hankel matrix of the denoised vibration signal, and the method comprises the following steps:
Figure BDA0003242586780000081
s24, carrying out inverse singular decomposition process on the H' matrix to obtain the vibration acceleration signal of the gearbox after noise reduction
X′＝[x′(1)，x′(2)，...x′(N)]。
In step S23 of this embodiment, an inverse singular value averaging method is used to determine the L value, and the obtained singular values are arranged in a descending manner by the inverse singular value averaging method:
K＝rank(σ1，σ2，...，σm)
definition of
Figure BDA0003242586780000082
And comparing the b with all singular values, considering the singular value larger than the mean value as an effective singular value, setting other singular values as 0, and performing signal reconstruction.
The specific content of step S3 in this embodiment is to map the one-dimensional time-series vibration deceleration signal after noise reduction into a two-dimensional space by using a symmetric image lattice analysis method, form a related symmetric pattern in the two-dimensional space, and perform feature extraction of the gear vibration acceleration signal of the rotating equipment by using the petal signal density, where the vibration acceleration values corresponding to the i-th and i + l-th time-series signals are x '(i) and x' (i + l), the signals at these two times are mapped into the two-dimensional space by using a symmetric lattice transformation, and the polar coordinates are represented as: z (r (i), θ (i), φ (i)), the mapping equation is as follows:
Figure BDA0003242586780000083
wherein r is the radius of the polar coordinate pattern, and theta is the angle of the petals deflected along the mirror image of the symmetrical plane in the counterclockwise direction; phi is a mirror image of the petals in the opposite direction clockwise; x is the number ofmaxIs the maximum amplitude in the signal sequence; x is the number ofminIs the minimum amplitude in the signal sequence; l is a time interval parameter; theta is a corresponding symmetrical angle of the petals when the petals are turned in a mirror image manner,
Figure BDA0003242586780000091
to enlarge factor
Figure BDA0003242586780000092
Of the present embodiment
Figure BDA0003242586780000093
The selection of the values l and theta is optimized by improving the particle swarm optimization, an optimization model with the two-dimensional image discrimination index as a target function is constructed, and optimization is carried out
Figure BDA0003242586780000094
l and theta, and the distinguishing indexes comprise: the thickness difference of the petal arms of the two-dimensional patterns and the included angle of the quasi-regression sidelines of the two adjacent petal arms are represented, and the objective function is represented by the functional of the corresponding points as follows:
Figure BDA0003242586780000095
wherein, | | | is functional calculation, and N is the number of sequences.
In the improved particle swarm algorithm of this embodiment, the influence of the average fitness position of the particle swarm on the position of the next generation of particles is added to the particle swarm position formula, so as to improve the convergence rate of the system, where the position formula is as follows:
vi(k)＝ωvi(k-1)+c1r1[pi，best(k-1)-xi(k-1)]
+c2r2[gbest-xi(k-1)]+c3r3[xavg(k-1)-xi(k-1)]
xi(k)＝vi(k)+xi(k-1)
where ω is the inertial weight of the flight velocity, viIs the particle velocity, xiIs the position of the particle, xavgIs the average position of the particles; r is1，r2，r3，c1，c1，c3The values of the random numbers are (0-1), (0-2) and (0-2), and k is the number of iterations.
In step S5 of this embodiment, a neural network training model is constructed from optimization of four main parameters, such as the number of iterations of the neural network, the batch size, the convolution kernel size, and the convolution kernel coefficient, for the vibration signal characteristics of the rotating machine gearbox.
In this embodiment, parameters such as the number of iterations, the batch size, the size of convolution kernels, and the number of convolution kernels of the neural network are optimized by using network calculation, and the network calculation is as follows:
Figure BDA0003242586780000096
setting the actual output O of the network, and the expected outputs Y and E as the difference, wherein k is iteration number, omega is batch size, b is convolution kernel size, beta convolution kernel number, and k, omega, b and beta satisfy the following relations:
Figure BDA0003242586780000101
wherein eta is1，η2，η3，η4The step size adaptive adjustment variables are random numbers of (0, 1).
The failure types of the present embodiments include, but are not limited to, rotating machine gearbox gear missing teeth, broken teeth, wear, tooth root cracking, bearing inner race failure, outer race, ball, hybrid failure, and the like.
As shown in fig. 2, noise reduction is performed when the singular value is 2, and the effect is as shown in the figure.
Fig. 3-7 show the waveform diagrams of the vibration signals collected under different conditions, which are normal, tooth root crack, gear wear, gear breakage and gear missing in sequence.
Fig. 8-12 show, respectively, two-dimensional maps of vibration signals acquired under different conditions, in order normal, gear missing, gear broken, gear wear, and tooth root cracking.
The vibration signal without noise reduction processing is as follows:
TABLE 1
Test specimen Is normal Failure 1 Failure 2 Failure 3 Failure 4
Sample 1 0.7877 0.7511 0.7466 0.7321 0.7751
Sample 2 0.8332 0.8589 0.7467 0.6533 0.7132
Sample 3 0.7107 0.7531 0.8158 0.7667 0.8042
Sample 4 0.7752 0.8443 0.6514 0.8933 0.8911
Sample 5 0.7423 0.8048 0.8788 0.8689 0.8992
The vibration signals subjected to the noise reduction processing in table 1 are as follows:
TABLE 2
Figure BDA0003242586780000102
Figure BDA0003242586780000111
The above-mentioned embodiments only express a certain implementation mode of the present invention, and the description thereof is specific and detailed, but not construed as limiting the scope of the present invention; it should be noted that, for those skilled in the art, without departing from the concept of the present invention, several variations and modifications can be made, which are within the protection scope of the present invention; therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (10)

1. A rotary machine gearbox fault diagnosis method based on deep learning is characterized by comprising the following steps:
s1, collecting a vibration signal of the operation of the rotary mechanical gear box;
s2, carrying out singular value decomposition on the vibration signal to obtain an original signal;
s3, mapping the one-dimensional time sequence form of the original signal to a two-dimensional space by using a symmetric dot matrix image analysis method to construct two-dimensional image information;
s4, extracting fault type features according to the density degree of discrete points of the two-dimensional image information;
and S5, carrying out fault training and discrimination by using the extracted fault type to carry out deep neural network.
2. The method of claim 1, wherein the vibration signals include rotational acceleration, rotational speed, displacement, and frequency.
3. The rotating machine gearbox fault diagnosis method based on deep learning of claim 1, wherein the singular value decomposition method in the step S2 is as follows:
s21, constructing a one-dimensional characteristic matrix Hankel matrix of the vibration acceleration time sequence signal, setting an original noise-containing signal sequence as X ═ X (1), X (2), … X (N), and constructing the Hankel matrix as follows:
Figure FDA0003242586770000011
wherein q < N and satisfies N ═ p + q-1, Ap×qIs a true signal feature matrix in the original time sequence signal, Bp×qAnd when constructing a Hankel matrix, the number p of rows is equal to the number q of columns, the time sequence of even data points is taken, and even processing is carried out if the sequence is an odd number.
S22, carrying out singular value decomposition on the Hankel matrix, and expressing the Hankel matrix as the product of two orthogonal matrices and an asymmetric diagonal matrix:
Figure FDA0003242586770000012
wherein,
Figure FDA0003242586770000013
and Vq×qP-order and q-order orthogonal matrixes respectively; mp×qIs an asymmetric diagonal matrix;
Figure FDA0003242586770000021
S＝diag(σ1,σ2,…,σr)，
σiis the singular value of the matrix Hankel, and r is the rank of the matrix.
S23, selecting L effective singular values, setting the rest singular values as 0, and reconstructing an inverse matrix in the singular value decomposition process to obtain a reconstructed matrix H ', wherein the H' is a Hankel matrix of the denoised vibration signal, and the method comprises the following steps:
Figure FDA0003242586770000022
s24, carrying out inverse singular decomposition process on the H' matrix to obtain the vibration acceleration signal of the gearbox after noise reduction
X′＝[x′(1),x′(2),…x′(N)]。
4. The rotating machinery gearbox fault diagnosis method based on deep learning of claim 3, wherein the determination of the L value in the step S23 adopts an inverse singular value average spectrum method, and the inverse singular value average spectrum method arranges the obtained singular values in a big-to-small manner:
K＝rank(σ1,σ2,…,σm)
definition of
Figure FDA0003242586770000023
And comparing the b with all singular values, considering the singular value larger than the mean value as an effective singular value, setting other singular values as 0, and performing signal reconstruction.
5. The rotating machine gearbox fault diagnosis method based on deep learning of claim 1, wherein: the specific content of step S3 is to map the one-dimensional time-series vibration deceleration signal after noise reduction into a two-dimensional space by using a symmetric image lattice analysis method, form a related symmetric pattern in the two-dimensional space, and perform feature extraction on the gear vibration acceleration signal of the rotating equipment by using the petal signal density, where the vibration acceleration values corresponding to the i-th and i + l-th signals of the time series are x '(i) and x' (i + l), the signals at these two times are mapped into the two-dimensional space by using a symmetric lattice transformation, and the polar coordinates are represented as: z (r (i), θ (i), φ (i)), the mapping equation is as follows:
Figure FDA0003242586770000031
wherein r is the radius of the polar coordinate pattern, and theta is the angle of the petals deflected along the mirror image of the symmetrical plane in the counterclockwise direction; phi is a mirror image of the petals in the opposite direction clockwise; x is the number ofmaxIs the maximum amplitude in the signal sequence; x is the number ofminIs the minimum amplitude in the signal sequence; l is a time interval parameter; theta is a corresponding symmetrical angle of the petals when the petals are turned in a mirror image manner,
Figure FDA0003242586770000032
to enlarge factor
Figure FDA0003242586770000033
6. The rotating machine gearbox fault diagnosis method based on deep learning of claim 5, wherein: the above-mentioned
Figure FDA0003242586770000034
The selection of the values l and theta is optimized by improving the particle swarm optimization, an optimization model with the two-dimensional image discrimination index as a target function is constructed, and optimization is carried out
Figure FDA0003242586770000035
l and theta, and the distinguishing indexes comprise: the thickness difference of the petal arms of the two-dimensional patterns and the included angle of the quasi-regression sidelines of the two adjacent petal arms are represented, and the objective function is represented by the functional of the corresponding points as follows:
Figure FDA0003242586770000036
wherein, | | | is functional calculation, and N is the number of sequences.
7. The rotating machine gearbox fault diagnosis method based on deep learning of claim 6, wherein: according to the improved particle swarm algorithm, the influence of the average fitness position of the particle swarm on the position of the next generation of particles is increased in a particle swarm position formula, so that the convergence speed of the system is improved, and the position formula is as follows:
vi(k)＝ωvi(k-1)+c1r1[pi,best(k-1)-xi(k-1)]+c2r2[gbest-xi(k-1)]+c3r3[xavg(k-1)-xi(k-1)]
xi(k)＝vi(k)+xi(k-1)
where ω is the inertial weight of the flight velocity, viIs the particle velocity, xiIs the position of the particle, xavgIs the average position of the particles; r is1，r2，r3，c1，c1，c3The values of the random numbers are (0-1), (0-2) and (0-2), and k is the number of iterations.
8. The rotating machine gearbox fault diagnosis method based on deep learning of claim 1, wherein: in step S5, a neural network training model is constructed progressively from the optimization of four main parameters, such as the iteration number, batch size, convolution kernel coefficient, and the like of the neural network, for the vibration signal characteristics of the rotating machine gearbox.
9. The rotating machinery gearbox fault diagnosis method based on deep learning of claim 8, wherein parameters such as iteration number, batch size, convolution kernel size and convolution kernel number of a neural network are optimized by using network calculation, wherein the network calculation is as follows:
Figure FDA0003242586770000041
setting the actual output O of the network, and the expected outputs Y and E as the difference, wherein k is iteration number, omega is batch size, b is convolution kernel size, beta convolution kernel number, and k, omega, b and beta satisfy the following relations:
Figure FDA0003242586770000042
wherein eta is1，η2，η3，η4The step size adaptive adjustment variables are random numbers of (0, 1).
10. The rotating machine gearbox fault diagnosis method based on deep learning of claim 1, wherein: the failure types include, but are not limited to, rotating machine gearbox gear missing teeth, broken teeth, wear, root cracks, bearing inner race failures, outer race, balls, hybrid failures, and the like.
CN202111023880.0A 2021-09-02 2021-09-02 Rotary machine gear box fault diagnosis method based on deep learning Pending CN113792397A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111023880.0A CN113792397A (en) 2021-09-02 2021-09-02 Rotary machine gear box fault diagnosis method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111023880.0A CN113792397A (en) 2021-09-02 2021-09-02 Rotary machine gear box fault diagnosis method based on deep learning

Publications (1)

Publication Number Publication Date
CN113792397A true CN113792397A (en) 2021-12-14

Family

ID=79182733

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111023880.0A Pending CN113792397A (en) 2021-09-02 2021-09-02 Rotary machine gear box fault diagnosis method based on deep learning

Country Status (1)

Country Link
CN (1) CN113792397A (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040001095A1 (en) * 2002-07-01 2004-01-01 Todd Marques Method and apparatus for universal device management
US20080151263A1 (en) * 2006-08-25 2008-06-26 Gerhard Randers-Pehrson Systems and methods for focusing optics
CN103810394A (en) * 2014-02-28 2014-05-21 东北电力大学 Design method for singular value decomposition and de-noising of rotating device fault signal
US9432298B1 (en) * 2011-12-09 2016-08-30 P4tents1, LLC System, method, and computer program product for improving memory systems
US20180012092A1 (en) * 2016-07-05 2018-01-11 Nauto, Inc. System and method for automatic driver identification
CN110509186A (en) * 2019-08-29 2019-11-29 华中科技大学 A kind of robot grinding and polishing quality characterization method based on processing vibration performance
CN110657985A (en) * 2019-10-11 2020-01-07 重庆邮电大学 Gearbox fault diagnosis method and system based on singular value spectrum manifold analysis

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040001095A1 (en) * 2002-07-01 2004-01-01 Todd Marques Method and apparatus for universal device management
US20080151263A1 (en) * 2006-08-25 2008-06-26 Gerhard Randers-Pehrson Systems and methods for focusing optics
US9432298B1 (en) * 2011-12-09 2016-08-30 P4tents1, LLC System, method, and computer program product for improving memory systems
CN103810394A (en) * 2014-02-28 2014-05-21 东北电力大学 Design method for singular value decomposition and de-noising of rotating device fault signal
US20180012092A1 (en) * 2016-07-05 2018-01-11 Nauto, Inc. System and method for automatic driver identification
CN110509186A (en) * 2019-08-29 2019-11-29 华中科技大学 A kind of robot grinding and polishing quality characterization method based on processing vibration performance
CN110657985A (en) * 2019-10-11 2020-01-07 重庆邮电大学 Gearbox fault diagnosis method and system based on singular value spectrum manifold analysis

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
修嘉芸等: "基于LSTM与迁移学习的滚动轴承故障诊断", 《重庆理工大学学报(自然科学)》 *

Similar Documents

Publication Publication Date Title
CN112161784B (en) Mechanical fault diagnosis method based on multi-sensor information fusion migration network
Hu et al. Data-driven fault diagnosis method based on compressed sensing and improved multiscale network
Xia et al. Fault diagnosis for rotating machinery using multiple sensors and convolutional neural networks
Han et al. Multi-level wavelet packet fusion in dynamic ensemble convolutional neural network for fault diagnosis
CN108171263B (en) Based on the Fault Diagnosis of Roller Bearings for improving variation mode decomposition and extreme learning machine
CN109765053B (en) Rolling bearing fault diagnosis method using convolutional neural network and kurtosis index
CN106980822B (en) A kind of rotary machinery fault diagnosis method based on selective ensemble study
Gecgel et al. Gearbox fault diagnostics using deep learning with simulated data
CN106447039A (en) Non-supervision feature extraction method based on self-coding neural network
CN105973594A (en) Rolling bearing fault prediction method based on continuous deep belief network
CN109186964B (en) rotary machine fault diagnosis method based on angle resampling and ROC-SVM
CN111220373A (en) Method for diagnosing faults of centrifugal pump rotor system
CN111753891B (en) Rolling bearing fault diagnosis method based on unsupervised feature learning
CN112257530A (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN110487547A (en) Fault Diagnosis of Roller Bearings under variable working condition based on vibrorecord and transfer learning
CN110455530A (en) Compose the gear case of blower combined failure diagnostic method of kurtosis combination convolutional neural networks
CN112461543B (en) Rotary machine fault diagnosis method based on multi-classification support vector data description
CN115127806A (en) Gear box fault diagnosis method and device based on multi-sensor vibration signals
CN112116029A (en) Intelligent fault diagnosis method for gearbox with multi-scale structure and characteristic fusion
Feng et al. A fault diagnosis for rolling bearing based on multilevel denoising method and improved deep residual network
Liu et al. Intelligent cross-condition fault recognition of rolling bearings based on normalized resampled characteristic power and self-organizing map
CN115587290A (en) Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network
Kong et al. Untrained compound fault diagnosis for planetary gearbox based on adaptive learning VMD and DSSECNN
CN113916535A (en) Bearing diagnosis method, system, equipment and medium based on time frequency and CNN
CN114186602A (en) Sparse filtering domain-based mechanical fault diagnosis method for anti-neural network

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