CN113536226B - Blind deconvolution algorithm for enhancing fault signal characteristics of rotary machine - Google Patents
Blind deconvolution algorithm for enhancing fault signal characteristics of rotary machine Download PDFInfo
- Publication number
- CN113536226B CN113536226B CN202110798951.8A CN202110798951A CN113536226B CN 113536226 B CN113536226 B CN 113536226B CN 202110798951 A CN202110798951 A CN 202110798951A CN 113536226 B CN113536226 B CN 113536226B
- Authority
- CN
- China
- Prior art keywords
- filters
- objective function
- blind deconvolution
- signal
- filter
- 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.)
- Active
Links
Classifications
-
- 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
- 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
-
- 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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Acoustics & Sound (AREA)
- Health & Medical Sciences (AREA)
- Algebra (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a blind deconvolution algorithm for enhancing fault signal characteristics of rotary machinery, which comprises the following steps: s1: constructing a plurality of cascaded FIR filters, determining a maximization criterion of a blind deconvolution algorithm according to the characteristics of an original vibration signal, and taking the maximization criterion as an objective function; s2: sequentially carrying out convolution operation on the original vibration signals by using a cascaded FIR filter to obtain filtered signals, and calculating objective function values of the filtered signals; s3: calculating the gradient of the objective function value to the filter under the current iteration times by adopting a backward automatic differential algorithm; s4: updating values of all filters; s5: and repeating S2-S4, and outputting a final filtered signal after the maximum iteration number is reached. The method solves the technical problem that the fault diagnosis precision of the rotating machinery is low due to the fact that iterative algorithms of different blind deconvolution algorithms cannot be used and the blind deconvolution algorithms are poor in performance.
Description
Technical Field
The invention relates to the technical field of rotary machine fault diagnosis and calculation based on deep learning, in particular to a blind deconvolution algorithm for enhancing rotary machine fault signal characteristics.
Background
Gearboxes and bearings are important components of rotating machinery, and their failure is the most common cause of mechanical failure. In order to reduce potential safety hazards and ensure normal operation of equipment, it is particularly important to monitor states of components such as a gear box and a bearing of a rotating machine. Upon the formation of a gearbox or bearing failure, a transient pulse will be generated that will be periodic. However, due to the influence of the transmission path and the environmental noise, the signal fault pulses collected by the vibration sensor are greatly attenuated, which brings a great impediment to fault analysis.
Blind deconvolution aims at solving the problem that an inverse FIR filter filters an original signal to recover pulse characteristics excited by faults to the maximum extent. The filter coefficients are solved by maximizing some measure of the filtered signal, which is commonly referred to as the maximization criterion. The main difference between different blind deconvolution algorithms is the different maximization criteria. For example, a kurtosis maximization-based blind deconvolution algorithm-MED; a blind deconvolution algorithm-MCKD based on the maximization of the correlation kurtosis; blind deconvolution algorithm based on D-norm maximization-ome, etc. These algorithms require artificial derivation of the partial derivatives of the maximization criteria to the filter coefficients, followed by further derivation of an iterative update formula for the filter coefficients, and finally multiple iterations of solving the filter coefficients and the filtered signal. The solving process needs to manually deduce an iterative updating formula of the filter coefficient, the calculating process is complex, the iterative updating formula of the blind deconvolution algorithm based on different maximization criteria cannot be used generally, and some maximization criteria cannot even deduce the iterative formula, so that the selection and design of the maximization criteria under different scenes are limited to a great extent; on the other hand, the blind deconvolution is essentially the solution of a non-convex optimization problem, and the conventional blind deconvolution algorithm is terminated when searching a maximum value in a solution space and cannot continue to search for a better solution, so that the conventional blind deconvolution algorithm has limited performance.
The blind deconvolution general solution based on different maximization criteria is researched, so that the flexibility of design of the maximization criteria can be ensured, and the requirements of fault diagnosis of rotary machinery in various occasions can be met. The iterative solving process is optimized, the performance of the blind deconvolution algorithm is improved, and the method has great significance for improving the accuracy of fault diagnosis of the rotary machine.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a blind deconvolution algorithm for enhancing the fault signal characteristics of rotary machinery, which solves the problems of low precision and small application range of the existing blind deconvolution algorithm.
The technical scheme adopted by the invention is as follows:
a blind deconvolution algorithm for enhancing the fault signal characteristics of a rotating machine, comprising the steps of:
s1: constructing a plurality of cascaded FIR filters, determining a maximization criterion of a blind deconvolution algorithm according to the characteristics of an original vibration signal, and taking the maximization criterion as an objective function;
s2: sequentially carrying out convolution operation on the original vibration signals by using the cascaded FIR filters to obtain filtered signals, and calculating objective function values of the filtered signals;
s3: calculating the gradient of the objective function on the filter under the current iteration round by adopting a backward automatic differential algorithm;
s4: updating the value of each filter;
s5: and repeating S2-S4, and outputting a final filtered signal after the maximum iteration number is reached.
The further technical scheme is as follows:
using n cascaded FIR filters { f 1 ,f 2 ,…,f n-1 ,f n The original vibration signals are subjected to convolution operation in sequence, and the formula is as follows:
s 1 ＝f 1 *x,s 2 ＝f 2 *s 1 ,…,s n-1 ＝f n-1 *s n-2 ,s n ＝f n *s n-1
wherein s is i Representing the output of the ith filter, i E [1, n]And x represents convolution operation, s n A filtered signal of the original vibration signal x, which is sequentially filtered by n filters;
the objective function determined according to the determined maximization criterion is J, s is calculated n Is J(s) n )。
The updating of the values of the filters comprises the following steps:
calculating a first order momentum and a second order momentum:
m t ＝β 1 ·m t-1 +(1-β 1 )·g t ,v t ＝β 2 ·v t-1 +(1-β 2 )·g t 2
wherein beta is 1 Is a first order momentum attenuation coefficient, beta 2 G is the second order momentum attenuation coefficient t For the objective function value J(s) at the current iteration number calculated by the backward automatic differentiation algorithm n ) For the gradient of the filter, the subscript t denotesIteration t; subscript t-1 represents the t-1 th iteration;
updating values of all filters:
wherein θ is t ＝{f 1 ,f 2 ,…,f n-1 ,f n The values of all filters at iteration t, α the learning rate, epsilon=10 -8 。
Gradient g t The calculation formula of (2) is as follows:
wherein J represents the group J (s n ) T represents the current iteration round, and n is the number of filters.
The beneficial effects of the invention are as follows:
the invention constructs a plurality of cascaded FIR filters, adopts a backward automatic differential algorithm, can search a better solution in a solution space, can select a maximization criterion according to specific conditions, furthest recovers transient pulses of fault signals of the rotary machine, and improves the precision and accuracy of fault diagnosis of the rotary machine.
The invention adopts the backward automatic differential algorithm to replace the manual derivation of the gradient formula, so that blind deconvolution iterative algorithm based on different maximization criteria can be universal, and the flexibility requirement of the design of the maximization criteria is met.
The invention filters the original signal by adopting a filter cascade method, so that the performance of the blind deconvolution algorithm is greatly improved.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
FIG. 2 is a graph showing the fault signals collected from the outer race of a rolling bearing for a rotary machine according to an embodiment of the present invention.
Fig. 3 is a graph showing the maximization criterion of the filtered signal in the iterative process according to the embodiment of the present invention.
Fig. 4 shows the filtered signal calculated in accordance with an embodiment of the present invention.
Fig. 5 is an envelope spectrum of a filtered signal calculated in accordance with an embodiment of the present invention.
Detailed Description
The following describes specific embodiments of the present invention with reference to the drawings.
A blind deconvolution algorithm for enhancing the fault characteristics of a rotating mechanical signal according to the present application may refer to fig. 1, and specifically includes the following steps:
s1: constructing a plurality of cascaded FIR filters, determining a maximization criterion of a blind deconvolution algorithm according to the characteristics of an original vibration signal x, and taking the maximization criterion as an objective function J;
the method specifically comprises the following steps:
s11: the original vibration signal x of the rotary machine is acquired by an acceleration sensor as an original input of a blind deconvolution algorithm,
x＝[x 1 ,x 2 ,…,x N-1 ,x N ] (1)
wherein x is i Each data value representing the original vibration signal, N representing the length of the signal.
S12: initializing a filter by adopting random numbers conforming to standard normal distribution:
θ 0 ＝{f 1 ,f 2 ,…,f n-1 ,f n } (2)
where n is the number of filters, n.gtoreq.2, θ represents all filters, and the subscript "0" represents the initialization operation prior to performing the iteration, where each filter f has L coefficients, i.e.,
l represents the length of the filter;
s13: specifying the maximum iteration number M, and initializing first-order momentum M and second-order momentum v:
m 0 ＝0,v 0 ＝0 (4)
s14: determining a maximization criterion function expression, and taking the maximization criterion function expression as an objective function J;
s2: sequentially carrying out convolution operation on the original vibration signal x by using the cascaded FIR filters to obtain a filtered signal, and calculating an objective function value of the filtered signal;
the method specifically comprises the following steps:
s21: using n cascaded FIR filters { f 1 ,f 2 ,…,f n-1 ,f n The original vibration signal x is filtered in turn, with the following formula:
s 1 ＝f 1 *x,s 2 ＝f 2 *s 1 ,…,s n-1 ＝f n-1 *s n-2 ,s n ＝f n *s n-1 (5)
s i representing the output of the ith filter, i E [1, n]And x represents convolution operation, s n The original vibration signal x is a signal which is filtered by n filters in turn;
the convolution operation is expressed in matrix form as follows:
y＝f*x＝X T f (6)
wherein X is a matrix constructed by the original vibration signal X, and the expression is as follows:
f represents a certain filter, and the expression is formula (3);
the expression of the original vibration signal x is formula (1);
s22: calculation s n Is J(s) n )；
S3: calculating gradient g of objective function value to filter under current iteration round t by adopting backward automatic differential algorithm t ；
Wherein g t For the gradient, J represents the above J(s) n ) T represents the current iteration round, f represents the filter, and n is the number of filters as described above.
S4: updating values of all filters;
the method comprises the following steps:
calculating a first order momentum and a second order momentum:
m t ＝β 1 ·m t-1 +(1-β 1 )·g t ,v t ＝β 2 ·v t-1 +(1-β 2 )·g t 2 (8)
wherein g t Gradient as described above; beta 1 The value of the first-order momentum attenuation coefficient is 0.9, beta 2 The value of the second-order momentum attenuation coefficient is 0.999; subscript t represents the t-th iteration; subscript t-1 represents the t-1 th iteration;
updating values of all filters:
wherein θ is t ＝{f 1 ,f 2 ,…,f n-1 ,f n The values of all filters in the t-th iteration are represented, and alpha represents the learning rate and is set to 0.01; to prevent the divisor from being 0, epsilon=10 is set -8 。
S5: performing iteration S2-S4, adding 1 to the iteration round t once each iteration, ending the iteration when the iteration round t reaches the maximum iteration times, and outputting a final filtered signal S n . Fault diagnosis technology is used to analyze the fault type and fault degree of the rotating machine.
The following further illustrates the aspects of the present application by way of specific examples:
the fault signals of the outer ring of the rolling bearing are collected through an acceleration sensor, the rotating speed of the shaft is 1649r/min, and the corresponding fault characteristic frequencies of the inner ring and the outer ring are 148.8Hz and 98.5Hz respectively. The sampling frequency is 10240Hz, the signal lasts for 0.6 seconds, and the original vibration signal is collected as shown in figure 2. From fig. 2 it can be seen that the sequence of faulty pulses in the original signal is drowned out by the noise component.
The maximum number of iterations m=100 is set. Setting parameters of an algorithm, wherein the filter length L=30, the filter number n=2, and initializing a filter f by adopting random numbers conforming to standard normal distribution 1 ,f 2 The kurtosis index is used as a maximization criterion.
Wherein, the expression of kurtosis is:
y n representing each data value of signal y, N is the number of data values of signal y, subscript N ε [1, N]。
The objective function is:
steps S2-S4 represent an iterative process, step S2 calculating the filtered signal S of the current iteration 2 And objective function value J(s) 2 ) S3 and S4 are used to update the values of the filter. S as the iterative process proceeds 2 The kurtosis of (c) gradually increases and eventually converges as shown in fig. 3.
When the maximum iteration number M is reached, the iteration is exited, and a filtered signal s is output 2 As shown in fig. 4. The amplitude of the signal noise after filtering is smaller, and the fault pulse sequence is clear.
Subsequently, the fault type and the fault degree of the rolling bearing are analyzed through an envelope analysis technology, and the envelope spectrum of the filtered signal is shown in fig. 5. The characteristic frequency of the outer ring fault and the frequency multiplication component thereof are clearly visible, which indicates that the rolling bearing of the rotary machine has serious outer ring fault.
Claims (1)
1. A blind deconvolution algorithm for enhancing the fault signal characteristics of a rotating machine, comprising the steps of:
s1: constructing a plurality of cascaded FIR filters, determining a maximization criterion of a blind deconvolution algorithm according to the characteristics of an original vibration signal, and taking the maximization criterion as an objective function;
s2: sequentially performing convolution operation on the original vibration signals by using the cascaded FIR filters to obtain filtered signals, and calculating objective function values of the filtered signals;
s3: calculating the gradient of the objective function value to the filter under the current iteration round by adopting a backward automatic differential algorithm;
s4: updating the value of each filter;
s5: repeating S2-S4, and outputting a final filtered signal after the maximum iteration number is reached;
using n cascaded FIR filters { f 1 ,f 2 ,…,f n-1 ,f n The original vibration signals are subjected to convolution operation in sequence, and the formula is as follows:
s 1 ＝f 1 *x,s 2 ＝f 2 *s 1 ,…,s n-1 ＝f n-1 *s n-2 ,s n ＝f n *s n-1
wherein s is i Representing the output of the ith filter, i E [1, n]And x represents convolution operation, s n A filtered signal of the original vibration signal x, which is sequentially filtered by n filters;
the objective function determined according to the determined maximization criterion is J, s is calculated n Is J(s) n )；
The updating of the values of the filters comprises the following steps:
calculating a first order momentum and a second order momentum:
m t ＝β 1 ·m t-1 +(1-β 1 )·g t ,v t ＝β 2 ·v t-1 +(1-β 2 )·g t 2
wherein beta is 1 Is a first order momentum attenuation coefficient, beta 2 G is the second order momentum attenuation coefficient t The objective function value J(s) under the current iteration round calculated by adopting the backward automatic differential algorithm n ) For the gradient of the filter, the subscript t represents the t-th iteration; subscript t-1 represents the firstt-1 times of iteration;
updating values of all filters:
wherein θ is t ＝{f 1 ,f 2 ,…,f n-1 ,f n The values of all filters at iteration t, α the learning rate, epsilon=10 -8 ；
Gradient g t The calculation formula of (2) is as follows:
wherein J represents the group J (s n ) T represents the current iteration round, and n is the number of filters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110798951.8A CN113536226B (en) | 2021-07-14 | 2021-07-14 | Blind deconvolution algorithm for enhancing fault signal characteristics of rotary machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110798951.8A CN113536226B (en) | 2021-07-14 | 2021-07-14 | Blind deconvolution algorithm for enhancing fault signal characteristics of rotary machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113536226A CN113536226A (en) | 2021-10-22 |
| CN113536226B true CN113536226B (en) | 2024-04-12 |
Family
ID=78099345
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110798951.8A Active CN113536226B (en) | 2021-07-14 | 2021-07-14 | Blind deconvolution algorithm for enhancing fault signal characteristics of rotary machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113536226B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114528525B (en) * | 2022-01-11 | 2023-03-28 | 西南交通大学 | Mechanical fault diagnosis method based on maximum weighted kurtosis blind deconvolution |
| CN115169417B (en) * | 2022-08-02 | 2023-11-03 | 江苏利核仪控技术有限公司 | Rolling bearing fault feature extraction method based on deflection maximization |
| CN115859091B (en) * | 2022-11-01 | 2023-05-26 | 哈尔滨工业大学 | A method for extracting bearing fault features, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104198187A (en) * | 2014-09-04 | 2014-12-10 | 昆明理工大学 | Mechanical vibration fault characteristic time domain blind extraction method |
| CN105699082A (en) * | 2016-01-25 | 2016-06-22 | 西安交通大学 | Sparse maximum signal-to-noise ratio deconvolution method |
| CN112284727A (en) * | 2020-09-30 | 2021-01-29 | 华南理工大学 | Rotary machine fault diagnosis method based on convolution maximum and minimum penalty algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102903368B (en) * | 2011-07-29 | 2017-04-12 | 杜比实验室特许公司 | Method and equipment for separating convoluted blind sources |
-
2021
- 2021-07-14 CN CN202110798951.8A patent/CN113536226B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104198187A (en) * | 2014-09-04 | 2014-12-10 | 昆明理工大学 | Mechanical vibration fault characteristic time domain blind extraction method |
| CN105699082A (en) * | 2016-01-25 | 2016-06-22 | 西安交通大学 | Sparse maximum signal-to-noise ratio deconvolution method |
| CN112284727A (en) * | 2020-09-30 | 2021-01-29 | 华南理工大学 | Rotary machine fault diagnosis method based on convolution maximum and minimum penalty algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 基于微粒群算法的盲解卷积及其在故障轴承声信号中的应用;沈沂 等;中国机械工程;20100625(第12期);全文 * |
| 基于最小熵解卷积-变分模态分解和优化支持向量机的滚动轴承故障诊断方法;姚成玉 等;中国机械工程;20171219;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113536226A (en) | 2021-10-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113536226B (en) | Blind deconvolution algorithm for enhancing fault signal characteristics of rotary machine | |
| McDonald et al. | Multipoint optimal minimum entropy deconvolution and convolution fix: application to vibration fault detection | |
| Randall et al. | A comparison of methods for separation of deterministic and random signals | |
| CN104677632B (en) | Utilize particle filter and the Fault Diagnosis of Roller Bearings of spectrum kurtosis | |
| CN112183259B (en) | A rolling bearing fault diagnosis method based on CEEMD and kurtosis weighted average threshold denoising | |
| CN109063672A (en) | A kind of early stage bearing outer ring method for diagnosing faults based on adaptive M CKD | |
| CN112098105B (en) | Gas turbine blade fault monitoring and early warning method based on casing broadband vibration signal | |
| CN110940522A (en) | Sparse separation and diagnosis method of periodic pulses for bearing faults under strong background noise | |
| CN113569990B (en) | Strong noise interference environment-oriented performance equipment fault diagnosis model construction method | |
| CN106502850B (en) | A kind of the multiple faults discrimination method and device of sparse tight frame dictionary learning model | |
| CN117851872B (en) | High-frequency vibration treatment diagnosis method and system for aero-engine | |
| CN114441172A (en) | Rolling bearing fault vibration signal analysis method | |
| CN115034263A (en) | Variable-scale evolution self-adaptive noise elimination method, noise elimination system and fault diagnosis system | |
| CN117828447B (en) | Aero-engine bearing fault diagnosis method based on spectral wavelet transform and multi-scale permutation entropy | |
| Yang et al. | Research on Fault Feature Extraction Method Based on FDM‐RobustICA and MOMEDA | |
| CN118332373A (en) | Rolling bearing fault diagnosis method based on parameter self-adaptive characteristic modal decomposition | |
| CN115169417B (en) | Rolling bearing fault feature extraction method based on deflection maximization | |
| CN115014765B (en) | Method for extracting fault characteristics of rolling bearing retainer through acoustic signals | |
| CN116361632A (en) | Feature Extraction Method of Bearing Minor Fault Based on CEEMDAN and MCKD | |
| CN110569478B (en) | An Improved Variational Mode Decomposition Method for Encoder Signal Analysis | |
| Ye et al. | Bearing fault diagnosis under non-stationary conditions based on a speed signal resonance component demodulation algorithm | |
| CN119202678B (en) | Rolling bearing fault diagnosis method and system based on controllable sparsity morphological decomposition | |
| CN114564986A (en) | Deconvolution feature extraction method for minimizing periodic noise amplitude ratio | |
| CN114528525A (en) | Mechanical fault diagnosis method based on maximum weighted kurtosis blind deconvolution | |
| Elasha et al. | Effectiveness of adaptive filter algorithms and spectral kurtosis in bearing faults detection in a gearbox |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |