CN112347588A - Rotary machine fault diagnosis method based on wavelet packet decomposition - Google Patents

Rotary machine fault diagnosis method based on wavelet packet decomposition Download PDF

Info

Publication number
CN112347588A
CN112347588A CN202011354224.4A CN202011354224A CN112347588A CN 112347588 A CN112347588 A CN 112347588A CN 202011354224 A CN202011354224 A CN 202011354224A CN 112347588 A CN112347588 A CN 112347588A
Authority
CN
China
Prior art keywords
fault
wavelet
fault diagnosis
packet decomposition
wavelet packet
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
CN202011354224.4A
Other languages
Chinese (zh)
Inventor
周涛涛
陈志敏
原宗
张冬
邹大程
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.)
China Ship Development and Design Centre
Original Assignee
China Ship Development and Design Centre
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 China Ship Development and Design Centre filed Critical China Ship Development and Design Centre
Priority to CN202011354224.4A priority Critical patent/CN112347588A/en
Publication of CN112347588A publication Critical patent/CN112347588A/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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Abstract

The invention discloses a rotating machinery fault diagnosis method based on wavelet packet decomposition, which comprises the following steps: 1) collecting vibration signals of the rotating machine in a normal state and a fault state; 2) selecting a wavelet basis function for extracting fault features; 3) acquiring sub-signals of different frequency bands of the vibration signal through wavelet packet decomposition according to the selected wavelet basis function; 4) calculating a fuzzy entropy value of the sub-signals to obtain a fault characteristic vector; 5) sorting the feature importance according to the relevance, and selecting a set number of fault feature vectors with the top sorting result according to the sorting result; 6) constructing a fault diagnosis model by using a classifier, dividing the selected fault characteristic vector and the class label into a training set and a test set together, and training the model by using the training set as the input of the model; 7) and inputting the test set into the fault diagnosis model to obtain a fault diagnosis result. The invention can effectively extract high-quality fault characteristics and improve the accuracy of fault diagnosis.

Description

Rotary machine fault diagnosis method based on wavelet packet decomposition
Technical Field
The invention relates to a mechanical fault diagnosis technology, in particular to a rotating machine fault diagnosis method based on wavelet packet decomposition.
Background
The rotary machine is used as a key component in a transmission system and is widely applied to industrial production of motors, engines, bearings, gear boxes and the like. The key parts of the rotary machine are easy to break down when running under bad or complex working conditions, and directly affect the mechanical performance, even seriously affect the production safety. Therefore, the fault diagnosis scheme of the rotary machine under the complex working condition is constructed, and the fault diagnosis method has important significance for ensuring the safe operation of equipment and reducing economic loss.
In the field of fault diagnosis of rotary machines, vibration signal acquisition, feature extraction and fault mode identification are three important aspects, and the feature extraction directly influences the final diagnosis result. The fault feature extraction method based on vibration signal analysis comprises time domain analysis, frequency domain analysis and time-frequency domain analysis, and corresponding features are called time domain features, frequency domain features and time-frequency domain features. The time domain characteristics comprise root mean square, mean value, kurtosis and the like, and the frequency domain analysis is mainly based on Fourier transform. However, due to the nonlinearity and non-stationarity of the vibration signal of the rotating machine, these methods are limited by a priori knowledge and expert experience, and it is difficult to efficiently mine the fault information hidden in the vibration signal. In recent years, common time-frequency domain decomposition methods include Empirical Mode Decomposition (EMD), Local Mean Decomposition (LMD), and Variational Mode Decomposition (VMD). However, the above method still has some drawbacks. For example, EMD has the disadvantages of end-point effects, modal aliasing, under-envelope and over-envelope, while LMD has the disadvantages of mode aliasing and computational inefficiency. In addition, these methods are based on "modes" whose sub-signals may lose some of the frequency components of the original signal.
Disclosure of Invention
The invention aims to solve the technical problem of providing a rotating machinery fault diagnosis method based on wavelet packet decomposition aiming at the defects in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a rotating machinery fault diagnosis method based on wavelet packet decomposition comprises the following steps:
1) collecting vibration signals of the rotating machine in a normal state and a fault state to obtain sample points containing the vibration signals in the normal state and the fault state;
2) selecting a wavelet basis function for extracting fault features;
3) acquiring sub-signals of different frequency bands of the vibration signal through wavelet packet decomposition according to the selected wavelet basis function;
4) calculating a fuzzy entropy value of the sub-signals to obtain a fault characteristic vector;
5) selecting fault characteristics; sorting the feature importance according to the relevance, and selecting a set number of fault feature vectors with the top sorting result according to the sorting result;
6) identifying a fault mode, constructing a fault diagnosis model through a classifier, dividing the selected fault characteristic vector and the class label into a training set and a testing set together, and taking the training set as the input of the model to train the model;
7) and inputting the test set into the fault diagnosis model to obtain a fault diagnosis result.
According to the scheme, the wavelet basis functions selected in the step 2) for fault feature extraction are selected based on the maximum energy-shannon entropy ratio and the minimum similarity.
According to the scheme, the wavelet basis functions for extracting the fault features are selected in the step 2), and the method specifically comprises the following steps:
2.1) determining a basis function to be measured by combining rotary mechanical vibration signals and the characteristics of different wavelet families;
2.2) randomly selecting s samples to form a new data set under each fault type, and performing j-layer wavelet packet decomposition on each sample by using the wavelet basis function;
2.3) respectively calculating the total energy and Shannon entropy ratio of the selected samples, and then calculating an average value;
2.4) determining the optimal wavelet basis function of each wavelet family according to the maximum principle of the energy-Shannon entropy ratio;
2.5) calculating the similarity between the reconstructed signal and the original signal according to the optimal wavelet basis functions of different wavelet families, and taking the wavelet basis function with the most similar reconstructed signal and original signal as the wavelet basis function for extracting fault features.
According to the scheme, the basis functions to be measured in the step 2.1) comprise: coif1 to coif5 wavelets, sym2 to sym8 wavelets, and db1 to db10 wavelets.
According to the scheme, the total energy to Shannon entropy ratio of the selected samples is calculated in the step 2.3), and the following formula is adopted:
for j layers of wavelet packet decomposition, total 2jThe energy value e (n) of the node at the nth node is defined as:
Figure BDA0002802121640000041
wherein, i is the serial number of the discrete point in the nth node; m is the total number of discrete points of the nth node; cn,iThe wavelet packet coefficients corresponding to the discrete points.
Entropy S of wavelet coefficient of nth nodeentropy(n) is defined as:
Figure BDA0002802121640000042
wherein
Figure BDA0002802121640000043
Is the energy probability distribution of the wavelet coefficients;
therefore, the ratio of the total energy of a j-layer wavelet packet decomposition to the total shannon entropy is defined as:
Figure BDA0002802121640000044
according to the above scheme, in step 2.5), for the optimal wavelet basis functions of different wavelet families, the similarity between the reconstructed signal and the original signal is calculated as follows:
respectively decomposing wavelet packets by using the optimal wavelet basis functions of different wavelet families, reconstructing the coefficients of the final layer of nodes after decomposition into time sequence signals, and measuring k-dimensional original signals x by standardized Euclidean distancei(i ═ 1,2, …, k) and reconstructed signal yi(i ═ 1,2, …, k) similarity:
Figure BDA0002802121640000051
wherein s isiIs xiAnd yiThe smaller d, the more similar the original signal is to the reconstructed signal.
According to the scheme, before wavelet packet decomposition is carried out in the step 3), Z-Score standardization is carried out on the original vibration signal, the average value of the standardized signal is 0, and the standard deviation is 1.
According to the scheme, the vibration signal is decomposed into 2 according to the wavelet packet decomposition of the j layers in the step 4)jAnd calculating fuzzy entropy values of each sub-signal to form a fault characteristic vector.
According to the scheme, the fault characteristic vector obtaining mode in the step 4) is as follows:
for each subsignal, the subsignal corresponds to an N-dimensional time series { x (i) ═ 1,2, …, N }, and the similarity is defined as:
Figure BDA0002802121640000052
wherein r is a similarity tolerance,
Figure BDA0002802121640000053
to represent
Figure BDA0002802121640000054
And
Figure BDA0002802121640000055
the distance between, t represents the gradient of the similarity tolerance;
defining functions
Figure BDA0002802121640000056
Obtaining a fuzzy entropy value of the sub-signal, namely a fault characteristic value is as follows:
Figure BDA0002802121640000057
for j-layer wavelet packet decomposition, the vibration signal is decomposed into 2jSub-signals having different frequency bands, so that the dimension of the fault feature vector is 2j。
According to the scheme, the feature importance ranking is carried out through the minimum redundancy-maximum correlation criterion in the step 5).
According to the scheme, a fault diagnosis model is constructed through a Catboost classifier in the step 6).
The invention has the following beneficial effects:
1. the wavelet packet decomposition is used for extracting the frequency components of the signals on the basis of selecting the optimal wavelet basis function by combining the wavelet packet decomposition with the fuzzy entropy. In addition, the fuzzy entropy is utilized to extract the implicit fault information in the decomposed sub-signals, so that the advantages of the information entropy method are retained, and the dynamic information of the time sequence can be effectively extracted. Meanwhile, the fuzzy entropy has the advantages of insensitivity to background noise and good robustness. The combination of the two can effectively extract high-quality fault characteristics, and is particularly suitable for extracting the fault characteristics under complex working conditions;
2. the invention adopts a characteristic selection method based on the minimum redundancy-maximum correlation criterion, can effectively remove redundant characteristics and simplify the modeling of a classifier;
3. according to the method, the vibration signals are decomposed into the sub-signals with different frequency bands through wavelet packet decomposition, the fuzzy entropy values of the sub-signals are calculated, high-quality fault feature vectors are obtained, and on the basis, a fault diagnosis model based on a Catboost classifier is constructed and trained according to the feature vectors selected according to the minimum redundancy-maximum correlation criterion, so that the health or fault state is effectively identified.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method of an embodiment of the present invention;
FIG. 2 is a schematic diagram of computing a fault signature vector according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a visualization of a fault feature vector of an embodiment of the invention;
FIG. 4 is a schematic diagram of the relationship between the selection of different numbers of features and the model training time and the fault diagnosis result in the feature selection process according to the embodiment of the present invention;
FIG. 5 is a diagram of a fault diagnosis multi-level confusion matrix according to an embodiment of the invention;
FIG. 6 is a schematic diagram of classification accuracy of a rotating machine fault diagnosis method under a complex working condition data set according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, a method for diagnosing faults of a rotating machine based on wavelet packet decomposition includes the following steps:
s1: acquiring original vibration signals of a rotary machine in a normal state and a fault state;
the method comprises the following steps that the method comprises nine fault types and a health state, wherein the fault types comprise a rotating shaft center bending, a rotor warping, a coupler bending, an eccentric rotor, a rotor unbalance, a rolling element fault of a rolling bearing, an inner ring fault, an outer ring fault and a mixed fault of an inner ring and an outer ring;
specifically, vibration signals of the rotating machine in a normal state and a fault state are collected through an acceleration sensor and a data collection system, and sample points containing the vibration signals in the normal state and the fault state are obtained;
s2: selecting an optimal wavelet basis function together according to an energy-shannon entropy ratio maximum principle and a similarity minimum principle;
in general, the main features to be considered in selecting wavelet basis functions include orthogonality, tight support, symmetry, and vanishing moments. In combination with the characteristics of the rotary machine vibration signal and different wavelet families, consider the coif1 to coif5 wavelets, sym2 to sym8 wavelets, and db1 to db10 wavelets as basis functions to be measured.
Specifically, in order to reduce the calculation time, 10 samples are randomly selected to form a new data set under each fault type, and 5-layer wavelet packet decomposition is performed on each sample by using the wavelet basis functions. The total energy to shannon entropy ratio ζ of the selected samples was calculated separately as follows, and then the average was calculated:
the energy value e (n) of the wavelet packet decomposition at the nth node is defined as:
Figure BDA0002802121640000081
wherein, i is the serial number of the discrete point in the nth node; m is the total number of discrete points of the nth node; cn,iThe wavelet packet coefficients corresponding to the discrete points.
Entropy S of wavelet coefficient of nth nodeentropy(n) is defined as:
Figure BDA0002802121640000091
wherein
Figure BDA0002802121640000092
As a wavelet systemEnergy probability distribution of numbers.
The ratio of the total energy of a j-layer wavelet packet decomposition to the shannon entropy is defined as:
Figure BDA0002802121640000093
and reconstructing the signal coefficient of the last layer of nodes after the decomposition of the corresponding wavelet packet aiming at the optimal wavelet basis functions of different wavelet families. K-dimensional raw signal x by normalized Euclidean distance metrici(i ═ 1,2, …, k) and reconstructed signal yiSimilarity between (i ═ 1,2, …, k):
Figure BDA0002802121640000094
wherein s isiIs xiAnd yiThe smaller d is, the more similar the original signal and the reconstructed signal is, and the more suitable the corresponding wavelet basis function is for signal analysis;
s3: on the basis of selecting the optimal wavelet basis function in the step S2, obtaining sub-signals of the vibration signal on different frequency bands, which are obtained in the step S1, through wavelet packet decomposition;
specifically, given a scale function φ (n) and a wavelet function ψ (n), let
Figure BDA0002802121640000095
Wavelet packet
Figure BDA0002802121640000096
Is defined as:
Figure BDA0002802121640000097
wherein i is 0,1,2 … …, n represents the value of a time sequence corresponding to a certain time, k represents time or position parameter, Z is integer set, hkDenotes a low-pass filter, gkDenotes a high-pass filter, hkAnd gkIs a pair of conjugate filters and satisfies gk＝(-1)kh1-k；
The vibration signal x (n) being decomposed into sub-signals using wavelet packets
Figure BDA0002802121640000101
Wherein j represents the number of decomposition layers, p represents the number of signals and p is 0,1, …,2j-1，
Figure BDA0002802121640000102
And represents the p-th wavelet packet coefficient of the j-th layer.
Before wavelet packet decomposition, the original vibration signal needs to be Z-Score normalized, and the normalized signal has a mean value of 0 and a standard deviation of 1.
S4: calculating the fuzzy entropy value of the sub-signal obtained in the step S3 to obtain a fault feature vector;
specifically, the sub-signals correspond to an N-dimensional time series { x (i) ═ 1,2, …, N }, and the similarity is defined as:
Figure BDA0002802121640000103
wherein r is a similarity tolerance,
Figure BDA0002802121640000104
to represent
Figure BDA0002802121640000105
And
Figure BDA0002802121640000106
the distance between, t is the gradient of the similar tolerance, define
Figure BDA0002802121640000107
The function is:
Figure BDA0002802121640000108
obtaining a fault characteristic value as follows:
Figure BDA0002802121640000109
it should be noted that the vibration signal is decomposed into 2 by j-layer wavelet packet decompositionjCalculating fuzzy entropy values of each sub-signal to form fault characteristic vectors;
s5: obtaining the importance ranking of the fault feature vectors obtained in the step S4 according to the minimum redundancy-maximum correlation criterion, and selecting different numbers of feature vectors according to the ranking result;
specifically, given two random variables X and Y, their mutual information is defined as follows:
Figure BDA0002802121640000111
where p (X) and p (Y) represent the probabilities of X and Y, respectively, and p (X, Y) is the joint probability density function of X and Y. Let x beiRepresenting a single feature, c represents a category, and the maximum correlation criterion is defined as:
Figure BDA0002802121640000112
where | S | is the dimension of the feature space S, I (x)i(ii) a c) Is a single feature xiAnd class c.
Supplementing a minimum redundancy condition to select mutually exclusive features, defined as:
Figure BDA0002802121640000113
the minimum redundancy-maximum correlation criterion is from the set X-Sm-1The mth feature is selected. For the selection of m-1 features, it is expressed as follows:
Figure BDA0002802121640000114
s6: constructing a fault diagnosis model through a Catboost classifier, dividing the selected fault feature vector and the class label into a training set and a testing set together, and taking the training set as the input of the model to train the model;
specifically, the data set after the feature extraction in step S4 is divided into a training set and a test set according to a ratio of 3:2, in order to eliminate the influence of contingency in sample division, ten-fold cross validation is performed on the training set, and the trained model is used for classification prediction of the test set. The parameters of the Catboost model are set, and the main parameters are shown in Table 1.
TABLE 1
Figure BDA0002802121640000121
S7: and inputting the test set into the fault diagnosis model trained in the step S6, so as to obtain a fault diagnosis result.
The utility of the intellectual achievement is further verified below.
Example 1: validation using data of rotor and bearing hybrid faults
To verify the effectiveness of the fault diagnosis method proposed by the present intellectual development, the present embodiment uses a data set of mechanical fault simulator platform rotor and bearing hybrid faults to verify the effectiveness of the proposed method. The experimental platform consists of an alternating current motor, a coupler, an acceleration sensor, a rotor, a rolling bearing, a centering adjusting disk, a data acquisition box and an inverter. The data set used contained nine fault types, including shaft center bending, rotor warping, coupling bending, eccentric rotor, rotor imbalance, and rolling bearing rolling element failure, inner ring failure, outer ring failure, mixed inner and outer ring failure, and a health status, as shown in table 2. The data sampling frequency was 6kHz and the motor speed was 2100 rpm.
The method comprises the following specific steps:
(1) data acquisition
And (4) forming an experimental data set of 10 types of faults by adopting an MFS data set under a single working condition. There are 160 samples per fault type, a total of 1600 samples, and each sample is 1000 data points in length.
TABLE 2
Figure BDA0002802121640000131
Figure BDA0002802121640000141
(2) Wavelet packet decomposition and selection of optimal wavelet basis function
The number of decomposition layers of the wavelet packet decomposition is set to 5, and the frequency band of each sample is uniformly divided into 32 parts. The optimal wavelet basis functions are selected according to the analysis of step S2. In order to reduce the calculation time, 10 samples are randomly selected under each fault type to form a new data set, the total energy to Shannon entropy ratio Zeta of the selected samples is calculated respectively, and then the average value is calculated. The results are shown in Table 3.
TABLE 3
Wavelet base Mean value of Wavelet base Mean value of Wavelet base Mean value of Wavelet base Mean value of
db1 4.538 db7 6.311 sym4 5.854 coif2 5.857
db2 5.349 db8 5.986 sym5 6.044 coif3 6.319
db3 5.660 db9 6.285 sym6 5.968 coif4 6.057
db4 5.960 db10 6.237 sym7 6.310 coif5 6.241
db5 5.986 sym2 5.349 sym8 6.151 / /
db6 5.951 sym3 5.660 coif1 5.362 / /
As can be seen from Table 3, the wavelet bases having the largest average values in the same wavelet family are db7, sym7 and coif3, respectively. The mean value of the similarity coefficients d for the selected samples is then calculated. The results are shown in Table 4. As can be seen from table 4, the original signal is most similar to the reconstructed signal when the coif3 wavelet basis is chosen. The Coif wavelet has orthogonality and tight support. Furthermore, it has better symmetry than the db wavelet. Therefore, the coif3 wavelet basis is selected to extract the fault characteristic information of the pulse signal more reasonably and effectively.
TABLE 4
Wavelet base db7 sym7 coif3
Similarity coefficient d (10)-12) 20.036 9.213 7.632
(3) Computing fuzzy entropy of sub-signals
And respectively calculating fuzzy entropy values of the sub-signals after wavelet packet decomposition, wherein each sample corresponds to 32 fault characteristics. The four parameter settings for blur entropy are shown in table 5, where STD is the standard deviation of the signal. The time consumption for feature extraction was 1.29 s/sample. In order to visualize the calculation results, a sample is randomly selected from each fault type, wavelet packet decomposition is performed on the sample, a fuzzy entropy value is calculated to obtain a feature vector, and the result is shown in fig. 2.
TABLE 5
Figure BDA0002802121640000151
Figure BDA0002802121640000161
(4) Feature visualization
t-SNE is a commonly used method in data reduction and feature visualization. The extracted features were compressed into two dimensions by t-SNE, the main parameters of which are shown in Table 6. For each fault type, 20 groups of data after t-SNE dimensionality reduction are selected to draw a scatter diagram, and the result is shown in FIG. 3. It can be seen that ten fault types have obvious differences, which indicates that the feature extraction method can effectively extract fault information of the rotary machine.
TABLE 6
Parameter(s) Value of
Algorithm exact
Principal component analysis quantity 20
Penalty term 40
Dimension of compression 2
Standardization false
Learning rate 1000
(5) Feature selection using minimum redundancy-maximum correlation
If too many features are extracted, the classification accuracy of the classifier may be affected, and the time consumption is increased. The minimum redundancy-maximum correlation criterion described in step S5 is optimized using an incremental search method. For 32 extracted fault features, the method can obtain the importance degree of each feature, and feature selection is carried out according to the importance degree. Table 7 gives the ranking results of the first 16 failure signatures of the signature selection.
TABLE 7
Serial number 1 2 3 4 5 6 7 8
Characteristic serial number 23 6 16 18 15 5 21 11
Value of 1.00 0.65 0.57 0.55 0.51 0.48 0.46 0.45
Serial number 9 10 11 12 13 14 15 16
Characteristic serial number 19 24 14 1 22 20 27 13
Value of 0.42 0.41 0.39 0.38 0.38 0.37 0.36 0.36
(6) Diagnostic results and analysis
Firstly, a data set consisting of the extracted features and the class labels is divided into a training set and a testing set according to the proportion of 3:2, namely, 96 samples are used for training a model for each fault type, and the rest 64 samples are used for testing. Setting model parameters of a Catboost classifier to construct a model, taking a training set as the input of the model, sorting according to the characteristics in the step (5) and the table 7, sequentially adding different numbers of characteristic training models, and then testing in a test set to obtain a relation curve among the characteristic number, the model training time and the classification precision. The results are shown in FIG. 4.
As can be seen from fig. 4, the model training time is positively correlated with the feature number, which is consistent with engineering experience. When 22 features are selected, the classification accuracy reaches the maximum of 99.17%. However, when all 32 features were used, the accuracy dropped by 0.21%, only 98.96%. Therefore, the classification accuracy using 22 features is taken as the final classification result, and the time consumption of model training is 19.80s at this time, which is reduced by 6.07s compared with the case of using all the features. Therefore, by using the feature selection method, better classification results can be obtained by using fewer features, and meanwhile, the redundancy of the features and the training time of the model can be effectively reduced.
Fig. 5 shows a multi-level confusion matrix in the fault diagnosis results when 22 features are used. It can be seen that the diagnosis results of the fault diagnosis method provided by the intellectual achievement on all fault categories are more than 98%. The diagnosis accuracy rate for the 2 nd, 3 rd, 6 th, 7 th, 9 th and 10 th fault types is 100%, the diagnosis accuracy rate for the 1 st, 4 th, 5 th and 8 th fault state types is 98%, and the 3 rd, 5 th, 4 th and 6 th fault state types can be misdiagnosed with a probability of 2%. It can be seen that the fault diagnosis method provided by the intellectual achievement has better diagnosis performance.
Example 2: validation using gearbox data under complex conditions
Since the mixed data set of the rotor and the bearing in the embodiment 1 has a single working condition, the gearbox data set with more complex working conditions is used for further verifying the effectiveness of the method provided by the intellectual achievement. The experimental platform consists of an electromagnetic brake, a torque sensor, a single-stage reducer, a brake controller and a servo motor. Gears with different crack lengths (including 0, 5, 10, 15 mm) were used and the sampling frequency was 5 kHz. In the intellectual achievement, ten data sets consisting of data from twenty working conditions are adopted for experimental verification, and detailed information of the twenty working conditions and the ten data sets is shown in tables 8 and 9 respectively.
TABLE 8
Figure BDA0002802121640000191
TABLE 9
Figure BDA0002802121640000192
Figure BDA0002802121640000201
The data sets from A1 to A4 correspond to single speed and multiple load conditions, the data sets from A5 to A9 correspond to single load and multiple speeds, and the data set A10 corresponds to the most complex conditions, i.e., multiple speeds and multiple loads. Each sample was of length that did not overlap 1500 data points, for a total of 3200 samples of a 10.
The classification result of the experiment is shown in fig. 6, and the accuracy of the training set of ten data sets reaches 100%. The accuracy of the test sets from A1 to A4 is higher than 97.08%, and the accuracy of the test sets from A5 to A9 is higher than 96.67%. Under the most complex conditions, the test set accuracy of A10 was 98.96%. Among them, the classification accuracy of a6 is the highest, reaching 100.00%. The validity of the proposed method was verified on the actual data set of the single-stage transmission.
The present embodiment also provides a fault diagnosis system for a rotary machine, including:
the signal acquisition unit is used for acquiring vibration signals of the rotary machine in a normal state and a fault state;
the signal decomposition unit is used for decomposing the acquired vibration signals by using a wavelet packet decomposition method to obtain sub-signals of different frequency bands;
the characteristic vector calculating unit is used for calculating the fuzzy entropy value of the sub-signals to obtain fault characteristic vectors;
the characteristic selection unit is used for calculating and obtaining importance ranking of the fault characteristic vectors;
the model training unit is used for constructing a fault diagnosis model by using a Catboost classifier, dividing the selected fault characteristic vector and the class label into a training set and a test set together, and taking the training set as the input of the model to train the model;
and the fault diagnosis unit is used for inputting the test set into the fault diagnosis model so as to obtain a fault diagnosis result.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.

Claims (10)

1. A rotating machinery fault diagnosis method based on wavelet packet decomposition is characterized by comprising the following steps:
1) collecting vibration signals of the rotating machine in a normal state and a fault state to obtain sample points containing the vibration signals in the normal state and the fault state;
2) selecting a wavelet basis function for extracting fault features;
3) acquiring sub-signals of different frequency bands of the vibration signal through wavelet packet decomposition according to the selected wavelet basis function;
4) calculating a fuzzy entropy value of the sub-signals to obtain a fault characteristic vector;
5) selecting fault characteristics;
sorting the feature importance according to the relevance, and selecting a set number of fault feature vectors with the top sorting result according to the sorting result;
6) identifying a fault mode;
constructing a fault diagnosis model by using a classifier, dividing the selected fault characteristic vector and the class label into a training set and a test set together, and training the model by using the training set as the input of the model;
7) and inputting the test set into the fault diagnosis model to obtain a fault diagnosis result.
2. The rotating machine fault diagnosis method based on wavelet packet decomposition according to claim 1, wherein the wavelet basis functions selected in step 2) for fault feature extraction are selected based on maximum energy-shannon entropy ratio and minimum similarity.
3. The rotating machine fault diagnosis method based on wavelet packet decomposition according to claim 1, wherein the wavelet basis functions for fault feature extraction are selected in step 2), specifically as follows:
2.1) determining a basis function to be measured by combining rotary mechanical vibration signals and the characteristics of different wavelet families;
2.2) randomly selecting s samples to form a new data set under each fault type, and performing j-layer wavelet packet decomposition on each sample by using the wavelet basis function;
2.3) respectively calculating the total energy and Shannon entropy ratio of the selected samples, and then calculating an average value;
2.4) determining the optimal wavelet basis function of each wavelet family according to the maximum principle of the energy-Shannon entropy ratio;
2.5) calculating the similarity between the reconstructed signal and the original signal according to the optimal wavelet basis functions of different wavelet families, and taking the wavelet basis function with the most similar reconstructed signal and original signal as the wavelet basis function for extracting fault features.
4. The method for fault diagnosis of rotating machinery based on wavelet packet decomposition according to claim 3, wherein the basis functions to be measured in step 2.1) comprise: coif1 to coif5 wavelets, sym2 to sym8 wavelets, and db1 to db10 wavelets.
5. The method for fault diagnosis of rotating machinery based on wavelet packet decomposition according to claim 3, wherein the total energy to Shannon entropy ratio of the selected samples is calculated in step 2.3), using the following formula:
for j layers of wavelet packet decomposition, total 2jThe energy value e (n) of the node at the nth node is defined as:
Figure FDA0002802121630000021
wherein, i is the serial number of the discrete point in the nth node; m is the total number of discrete points of the nth node; cn,iThe wavelet packet coefficients corresponding to the discrete points.
Entropy S of wavelet coefficient of nth nodeentropy(n) is defined as:
Figure FDA0002802121630000022
wherein
Figure FDA0002802121630000023
Is the energy probability distribution of the wavelet coefficients;
thus, the ratio of the total energy of a j-layer wavelet packet decomposition to the total shannon entropy can be defined as:
Figure FDA0002802121630000024
6. the fault diagnosis method for rotating machinery based on wavelet packet decomposition according to claim 3, wherein the similarity between the reconstructed signal and the original signal constructed by the method in step 2.5) is calculated for the optimal wavelet basis functions of different wavelet families as follows:
respectively decomposing the wavelet basis functions of j layers by using the optimal wavelet basis functions of different wavelet families, reconstructing the coefficient of the node of the last layer after decomposition into a time sequence signal, and measuring a k-dimensional original signal x by a standardized Euclidean distanceiI ═ 1,2, …, k; and reconstructing the signal yiI ═ 1,2, …, k; similarity between:
Figure FDA0002802121630000031
wherein s isiIs xiAnd yiThe smaller d, the more similar the original signal is to the reconstructed signal.
7. The method for fault diagnosis of rotating machinery based on wavelet packet decomposition according to claim 1, wherein in step 4), the vibration signal is decomposed into 2 according to j layers of wavelet packet decompositionjAnd calculating fuzzy entropy values of each sub-signal to form a fault characteristic vector.
8. The rotating machinery fault diagnosis method based on wavelet packet decomposition according to claim 1, wherein the fault feature vector in step 4) is obtained as follows:
for each subsignal, the subsignal corresponds to an N-dimensional time series { x (i) ═ 1,2, …, N }, and the similarity is defined as:
Figure FDA0002802121630000032
wherein r is a similarity tolerance,
Figure FDA0002802121630000033
to represent
Figure FDA0002802121630000034
And
Figure FDA0002802121630000035
the distance between, t represents the gradient of the similarity tolerance;
defining functions
Figure FDA0002802121630000036
Obtaining a fuzzy entropy value of the sub-signal, namely a fault characteristic value is as follows:
Figure FDA0002802121630000041
for j-layer wavelet packet decomposition, the vibration signal is decomposed into 2jSub-signals having different frequency bands, so that the dimension of the fault feature vector is 2j。
9. The method according to claim 1, wherein the step 5) is performed by sorting the importance of features according to the minimum redundancy-maximum correlation criterion.
10. The rotating machine fault diagnosis method based on wavelet packet decomposition according to claim 1, wherein in the step 6), a fault diagnosis model is constructed through a Catboost classifier.
CN202011354224.4A 2020-11-26 2020-11-26 Rotary machine fault diagnosis method based on wavelet packet decomposition Pending CN112347588A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011354224.4A CN112347588A (en) 2020-11-26 2020-11-26 Rotary machine fault diagnosis method based on wavelet packet decomposition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011354224.4A CN112347588A (en) 2020-11-26 2020-11-26 Rotary machine fault diagnosis method based on wavelet packet decomposition

Publications (1)

Publication Number Publication Date
CN112347588A true CN112347588A (en) 2021-02-09

Family

ID=74364994

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011354224.4A Pending CN112347588A (en) 2020-11-26 2020-11-26 Rotary machine fault diagnosis method based on wavelet packet decomposition

Country Status (1)

Country Link
CN (1) CN112347588A (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113076844A (en) * 2021-03-26 2021-07-06 华中科技大学 Method for constructing fault diagnosis model of rotating part and application
CN113095170A (en) * 2021-03-29 2021-07-09 天地（常州）自动化股份有限公司 Motor fault diagnosis method based on adjustable Q wavelet
CN113191232A (en) * 2021-04-21 2021-07-30 西安交通大学 Electro-hydrostatic actuator fault identification method based on multi-mode homologous features and XGboost model
CN113283292A (en) * 2021-04-13 2021-08-20 江苏大学镇江流体工程装备技术研究院 Method and device for fault diagnosis of underwater micro propeller
CN113378943A (en) * 2021-06-16 2021-09-10 西北工业大学 Engine rotor rubbing fault diagnosis method based on wavelet-gray level co-occurrence matrix
CN113434973A (en) * 2021-06-21 2021-09-24 西安科技大学 Mechanical fault diagnosis method based on improved PSO-SOM-BPNN
CN114034381A (en) * 2021-11-12 2022-02-11 广东电网有限责任公司江门供电局 Distribution transformer vibration extraction method and system based on wavelet information entropy
CN114414242A (en) * 2021-12-17 2022-04-29 哈尔滨工业大学 Large-scale rotating equipment spindle state signal feature extraction method based on multi-domain analysis and Relieff
CN115355162A (en) * 2022-08-08 2022-11-18 西安交通大学 Diaphragm compressor diaphragm fault diagnosis method based on oil pressure
CN115510925A (en) * 2022-11-17 2022-12-23 中国人民解放军国防科技大学 Mechanical fault diagnosis method, device and medium
CN116659863A (en) * 2023-05-19 2023-08-29 云南中广核能源服务有限公司 Bearing vibration signal processing method based on wavelet packet
CN116736091A (en) * 2023-08-10 2023-09-12 湖南遥光科技有限公司 Electronic system test point expansion method and system, and fault diagnosis method and system

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020131497A1 (en) * 2001-02-07 2002-09-19 Samsung Electronics Co., Ltd. Apparatus and method for image coding using tree-structured quantization based on wavelet transform
CN105701470A (en) * 2016-01-13 2016-06-22 合肥工业大学 Analog circuit fault characteristic extraction method based on optimal wavelet packet decomposition
CN106092574A (en) * 2016-05-30 2016-11-09 西安工业大学 The Method for Bearing Fault Diagnosis selected with sensitive features is decomposed based on improving EMD
CN106441896A (en) * 2016-10-14 2017-02-22 石家庄铁道大学 Characteristic vector extraction method for rolling bearing fault mode identification and state monitoring
CN107228766A (en) * 2017-05-22 2017-10-03 上海理工大学 Based on the Fault Diagnosis of Roller Bearings for improving multiple dimensioned fuzzy entropy
CN109521301A (en) * 2018-11-30 2019-03-26 北京航空航天大学 A kind of fault electric arc generation device and its detection method
US20190205778A1 (en) * 2017-12-28 2019-07-04 Tata Consultancy Services Limited Systems and methods for obtaining optimal mother wavelets for facilitating machine learning tasks
CN111027259A (en) * 2019-12-20 2020-04-17 昆明理工大学 Rolling bearing fault detection method for optimizing BP neural network based on dragonfly algorithm
CN111397896A (en) * 2020-03-08 2020-07-10 华中科技大学 Fault diagnosis method and system for rotary machine and storage medium

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020131497A1 (en) * 2001-02-07 2002-09-19 Samsung Electronics Co., Ltd. Apparatus and method for image coding using tree-structured quantization based on wavelet transform
CN105701470A (en) * 2016-01-13 2016-06-22 合肥工业大学 Analog circuit fault characteristic extraction method based on optimal wavelet packet decomposition
CN106092574A (en) * 2016-05-30 2016-11-09 西安工业大学 The Method for Bearing Fault Diagnosis selected with sensitive features is decomposed based on improving EMD
CN106441896A (en) * 2016-10-14 2017-02-22 石家庄铁道大学 Characteristic vector extraction method for rolling bearing fault mode identification and state monitoring
CN107228766A (en) * 2017-05-22 2017-10-03 上海理工大学 Based on the Fault Diagnosis of Roller Bearings for improving multiple dimensioned fuzzy entropy
US20190205778A1 (en) * 2017-12-28 2019-07-04 Tata Consultancy Services Limited Systems and methods for obtaining optimal mother wavelets for facilitating machine learning tasks
CN109521301A (en) * 2018-11-30 2019-03-26 北京航空航天大学 A kind of fault electric arc generation device and its detection method
CN111027259A (en) * 2019-12-20 2020-04-17 昆明理工大学 Rolling bearing fault detection method for optimizing BP neural network based on dragonfly algorithm
CN111397896A (en) * 2020-03-08 2020-07-10 华中科技大学 Fault diagnosis method and system for rotary machine and storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
杨咸启: "《接触力学理论与滚动轴承设计分析》", 30 April 2018, pages: 326 - 329 *

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113076844B (en) * 2021-03-26 2022-05-20 华中科技大学 Method for constructing fault diagnosis model of rotating part and application
CN113076844A (en) * 2021-03-26 2021-07-06 华中科技大学 Method for constructing fault diagnosis model of rotating part and application
CN113095170A (en) * 2021-03-29 2021-07-09 天地（常州）自动化股份有限公司 Motor fault diagnosis method based on adjustable Q wavelet
CN113095170B (en) * 2021-03-29 2024-04-02 天地（常州）自动化股份有限公司 Fault diagnosis method based on adjustable Q wavelet motor
CN113283292B (en) * 2021-04-13 2023-08-08 江苏大学镇江流体工程装备技术研究院 Method and device for diagnosing faults of underwater micro-propeller
CN113283292A (en) * 2021-04-13 2021-08-20 江苏大学镇江流体工程装备技术研究院 Method and device for fault diagnosis of underwater micro propeller
CN113191232A (en) * 2021-04-21 2021-07-30 西安交通大学 Electro-hydrostatic actuator fault identification method based on multi-mode homologous features and XGboost model
CN113378943A (en) * 2021-06-16 2021-09-10 西北工业大学 Engine rotor rubbing fault diagnosis method based on wavelet-gray level co-occurrence matrix
CN113434973B (en) * 2021-06-21 2023-04-07 西安科技大学 Mechanical fault diagnosis method based on improved PSO-SOM-BPNN
CN113434973A (en) * 2021-06-21 2021-09-24 西安科技大学 Mechanical fault diagnosis method based on improved PSO-SOM-BPNN
CN114034381B (en) * 2021-11-12 2023-10-10 广东电网有限责任公司江门供电局 Distribution transformer vibration extraction method and system based on wavelet information entropy
CN114034381A (en) * 2021-11-12 2022-02-11 广东电网有限责任公司江门供电局 Distribution transformer vibration extraction method and system based on wavelet information entropy
CN114414242A (en) * 2021-12-17 2022-04-29 哈尔滨工业大学 Large-scale rotating equipment spindle state signal feature extraction method based on multi-domain analysis and Relieff
CN115355162A (en) * 2022-08-08 2022-11-18 西安交通大学 Diaphragm compressor diaphragm fault diagnosis method based on oil pressure
CN115355162B (en) * 2022-08-08 2023-12-19 西安交通大学 Diaphragm fault diagnosis method of diaphragm compressor based on oil pressure
CN115510925A (en) * 2022-11-17 2022-12-23 中国人民解放军国防科技大学 Mechanical fault diagnosis method, device and medium
CN116659863A (en) * 2023-05-19 2023-08-29 云南中广核能源服务有限公司 Bearing vibration signal processing method based on wavelet packet
CN116659863B (en) * 2023-05-19 2024-04-19 云南中广核能源服务有限公司 Bearing vibration signal processing method based on wavelet packet
CN116736091A (en) * 2023-08-10 2023-09-12 湖南遥光科技有限公司 Electronic system test point expansion method and system, and fault diagnosis method and system

Similar Documents

Publication Publication Date Title
CN112347588A (en) Rotary machine fault diagnosis method based on wavelet packet decomposition
CN111397896B (en) Fault diagnosis method and system for rotary machine and storage medium
CN107702922B (en) Rolling bearing fault diagnosis method based on LCD and stacking automatic encoder
CN109187021B (en) Multi-source Wind turbines Method for Bearing Fault Diagnosis based on entropy
CN111413089A (en) Gear fault diagnosis method based on combination of VMD entropy method and VPMCD
CN110617960A (en) Wind turbine generator gearbox fault diagnosis method and system
CN107543722A (en) The Rolling Bearing Fault Character extracting method of dictionary learning is stacked based on depth
CN113177537B (en) Fault diagnosis method and system for rotary mechanical equipment
CN115730199B (en) Rolling bearing vibration signal noise reduction and fault feature extraction method and system
CN113112000B (en) Construction method and application of multilayer convolutional neural network model
Chen et al. A visualized classification method via t-distributed stochastic neighbor embedding and various diagnostic parameters for planetary gearbox fault identification from raw mechanical data
Al Tobi et al. Using MLP‐GABP and SVM with wavelet packet transform‐based feature extraction for fault diagnosis of a centrifugal pump
CN112116029A (en) Intelligent fault diagnosis method for gearbox with multi-scale structure and characteristic fusion
Zhang et al. Fault diagnosis based on optimized node entropy using lifting wavelet packet transform and genetic algorithms
CN114048787A (en) Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model
Fa-jun et al. Compound fault diagnosis of gearbox based on wavelet packet transform and sparse representation classification
CN112163630A (en) Compound fault diagnosis method and device based on unbalanced learning
Ou et al. Research on rolling bearing fault diagnosis using improved majorization-minimization-based total variation and empirical wavelet transform
CN114136604B (en) Rotary equipment fault diagnosis method and system based on improved sparse dictionary
CN115048960A (en) Equipment state detection method
CN112485001A (en) Fault diagnosis method for intelligent elevator
Chakraborty et al. Bearing Fault Detection based on Internet of Things using Convolutional Neural Network
Xiang et al. Virtual simulation analysis and experimental study on gear fault diagnosis based on wavelet neural network
Guo et al. PGA-MSM-HI: A Health Indicator for Initial Degeneration Point Detection of Machines Based on Deep Metric Learning
Wang et al. Mathematical Problems in Engineering Research on Composite Fault Separation of Rolling Bearing Based on Functional Mixing Decoupling Model

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

Application publication date: 20210209

RJ01 Rejection of invention patent application after publication