CN113420691A - Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient - Google Patents
Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient Download PDFInfo
- Publication number
- CN113420691A CN113420691A CN202110737611.4A CN202110737611A CN113420691A CN 113420691 A CN113420691 A CN 113420691A CN 202110737611 A CN202110737611 A CN 202110737611A CN 113420691 A CN113420691 A CN 113420691A
- Authority
- CN
- China
- Prior art keywords
- value
- signal
- wavelet
- frequency
- domain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000003745 diagnosis Methods 0.000 title claims abstract description 30
- 239000013598 vector Substances 0.000 claims abstract description 57
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 25
- 230000009466 transformation Effects 0.000 claims abstract description 11
- 238000007637 random forest analysis Methods 0.000 claims abstract description 9
- 230000010354 integration Effects 0.000 claims abstract description 5
- 238000004458 analytical method Methods 0.000 claims description 20
- 238000001228 spectrum Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 10
- 230000009467 reduction Effects 0.000 claims description 8
- 238000010219 correlation analysis Methods 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 6
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 claims description 6
- 230000005484 gravity Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000003909 pattern recognition Methods 0.000 claims description 4
- YBJHBAHKTGYVGT-ZKWXMUAHSA-N (+)-Biotin Chemical compound N1C(=O)N[C@@H]2[C@H](CCCCC(=O)O)SC[C@@H]21 YBJHBAHKTGYVGT-ZKWXMUAHSA-N 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 3
- 230000004044 response Effects 0.000 claims description 3
- 230000000630 rising effect Effects 0.000 claims description 3
- FEPMHVLSLDOMQC-UHFFFAOYSA-N virginiamycin-S1 Natural products CC1OC(=O)C(C=2C=CC=CC=2)NC(=O)C2CC(=O)CCN2C(=O)C(CC=2C=CC=CC=2)N(C)C(=O)C2CCCN2C(=O)C(CC)NC(=O)C1NC(=O)C1=NC=CC=C1O FEPMHVLSLDOMQC-UHFFFAOYSA-N 0.000 claims description 3
- 238000011089 mechanical engineering Methods 0.000 abstract description 2
- 238000002474 experimental method Methods 0.000 description 12
- 238000000605 extraction Methods 0.000 description 11
- 238000005096 rolling process Methods 0.000 description 9
- 239000000284 extract Substances 0.000 description 7
- 238000012360 testing method Methods 0.000 description 5
- 230000002596 correlated effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000006378 damage Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000003889 chemical engineering Methods 0.000 description 1
- 230000000875 corresponding effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005272 metallurgy Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- 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
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- General Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a mixed domain characteristic bearing fault diagnosis method based on a Pearson correlation coefficient, and belongs to the technical field of mechanical engineering automation. Firstly, 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transformation feature vectors and 10 complete integration empirical mode decomposition (CEEMDAN) feature vectors of self-adaptive noise are extracted from an original signal, and a bearing fault mixed domain feature set is constructed by combining the extracted feature parameters. Secondly, extracting low-dimensional feature vectors which are easy to identify from a high-dimensional fault feature set by using a Pearson correlation coefficient. And finally, importing the low-dimensional feature set into a random forest for classification and identification. Experimental results show that the classification accuracy of the mixed domain bearing fault diagnosis method based on the Pearson correlation coefficient can reach 97.32%, and the method has obvious advantages compared with other methods.
Description
Technical Field
The invention relates to a mixed domain characteristic bearing fault diagnosis method based on a Pearson correlation coefficient, and belongs to the technical field of mechanical engineering automation.
Background
Rotary machines are widely used as important components of mechanical systems in the fields of electric power, metallurgy, chemical engineering, machine manufacturing, and the like. The health state of the device not only influences the safe and stable operation of the device, but also directly influences the later production. More seriously, equipment failure may cause local damage, significant economic loss and even loss of life and personal injury. Extensive research has shown that bearing failure accounts for 30% of rotary machine failure. Therefore, the deep research on the condition monitoring and fault diagnosis of the bearing has important significance for maintaining the safety of equipment and reducing the maintenance cost. In recent years, with the rapid development of signal processing, data mining and artificial intelligence technologies, fault diagnosis methods play an important role in fault diagnosis of rotary machines, and the steps of the methods mainly comprise three steps of signal processing, feature extraction and pattern recognition.
When the fault diagnosis method is applied to diagnosis of the bearing, the following problems exist, (1) feature extraction and feature set construction are difficult. The training sample and the test sample are typically vibration signals derived under the same operating conditions. The characteristics of the method comprise root mean square, variance, kurtosis, skewness index, information entropy and the like. However, due to coupling between bearings, accurate extraction of features from the collected bearing vibration signals is challenging. In addition, the harsh testing environment increases the difficulty of keeping consistency of the working conditions of the training and testing samples. Mechanical uniqueness reduces the accuracy of the diagnosis; (2) the nonlinear feature set dimension is too high. In order to fully realize the accurate position and fault depth of the bearing fault, various features need to be extracted from the vibration signal to construct a feature set. The resulting feature set contains both insensitive and less sensitive features, with non-linearity and high dimensional numbers. Both the diversity of features and the dimensional characteristics can affect the accuracy of the diagnosis. The high-dimensional feature set increases the calculation time and reduces the diagnosis precision, so that a low-dimensional and easily-recognized main feature vector needs to be acquired from the high-dimensional loose feature set by adopting a dimension reduction method.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: a mixed domain characteristic bearing fault diagnosis method based on a Pearson correlation coefficient is provided, and the problems are solved as follows: (1) feature extraction and feature set construction are difficult; (2) and the dimension of the nonlinear feature set is too high.
The technical scheme adopted by the invention is as follows: a mixed domain characteristic bearing fault diagnosis method based on a Pearson correlation coefficient is disclosed. Firstly, 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transformation feature vectors and 10 complete integration empirical mode decomposition (CEEMDAN) feature vectors of self-adaptive noise are extracted from an original signal, and a bearing fault mixed domain feature set is constructed by combining the extracted feature parameters. Secondly, performing relevance analysis on the extracted mixed domain features by using a Pearson correlation coefficient, and extracting low-dimensional principal feature vectors which are easy to identify from a high-dimensional fault feature set. And finally, importing the low-dimensional feature set into a random forest as the input of pattern recognition. Experimental results show that the model has good distinguishing capability on the characteristic correlation, and the accuracy can reach 97.32%.
The method comprises the following specific steps:
(1) the method comprises the steps of utilizing fan end bearing data with a data set of 12K sampling frequency to respectively collect 4 fault data of different states including normal state, inner ring fault, rolling element fault and outer ring fault (6 o' clock direction), wherein each state except the normal data has 3 fault depth types, the diameters of the fault depth types are 0.1778MM, 0.3556MM and 0.5334MM respectively, the load of a bearing motor is 0, the rotating speed of the bearing is 1797r/min, and 10 types of fault types are used as data sources of the experiment. Each class of data is divided into 115 classification samples for a total of 1150 samples of 10 classes. The training set size is 700 parts per class, 70 parts per class, the test set 450 parts per class 45 parts. Wherein RF, IF and OF are faults OF the rolling body, the inner ring and the outer ring (6 o' clock direction), respectively.
(2) The collected original signals are subjected to feature extraction, and a time domain, a frequency domain and a time-frequency domain (wavelet analysis and CEEMDAN) are respectively extracted. And extracting time domain features and numbering. 1-10 are dimensional time domain features, which are respectively a maximum value, a minimum value, a peak-to-peak value, an average value, an absolute average value, a square root amplitude value, a variance, a standard deviation and an effective value. 11-17 are dimensionless time domain features, which are kurtosis, skewness, form factor, peak factor, pulse factor, margin factor, and clearance factor, respectively, and are recorded as follows. The frequency domain features are numbered 18-22, which are the average frequency, the center of gravity frequency, the root mean square of the frequency, and the standard deviation of the frequency, respectively. The time-frequency characteristics mainly extract the wavelet transform and CEEMDAN related characteristics. In the wavelet transformation, three-layer decomposition is carried out on an original vibration signal, and the original vibration signal is divided into 8 sub-frequency bands. It can be seen that the first four frequency bands contain most of the energy of the original signal, so that the wavelet scale entropy of the first four sub-bands is extracted, and then the wavelet energy spectrum entropy and the wavelet singular entropy of the signal are extracted to form one time-frequency feature subset. And respectively carrying out CEEMDAN decomposition on the 10 types of original signals, and taking four types of original signals to obtain 9 inherent modal components. The 1 st component signal has a significantly smaller amplitude and faster vibration frequency than other components, and therefore, the component can be judged to be random noise and not selected as a component for feature extraction. The 2 nd to 6 th components have slow fluctuation, most of time domain waveform diagrams are composed of higher harmonics, the frequency spectrum energy is concentrated, most of time domain waveforms are close to sine waves, and the time domain waveforms are ideal signal analysis time sequences. The residual component contains less characteristic information, has small correlation with the original signal, lacks the physical analysis significance, belongs to a false component and is removed. Thus extracting the permutation entropy and instantaneous energy of the 2-6 component signals respectively to form another subset of the time-frequency characteristics.
(3) Normalizing the extracted data: normalization is a general step of data processing, and has the advantages of accelerating the optimization speed of gradient descent solution after feature processing and improving the algorithm classification precision. Therefore, firstly, the extracted time-frequency domain characteristics are normalized, and in the experiment, 0,1 is selected as the value range of the normalization.
(4) Performing correlation analysis on the extracted time-frequency characteristics, selecting a characteristic confusion matrix diagram larger than a threshold value through a Pearson Correlation Coefficient (PCC), and indicating that the correlation coefficient is from weak to strong from the top down, wherein the characteristics are as follows: the first component permutation entropy value of CEEMDAN, the second component permutation entropy value of CEEMDAN, the third component permutation entropy value of CEEMDAN, the fourth component permutation entropy value of CEEMDAN, the fifth component permutation entropy value of CEEMDAN, the barycentric frequency, the frequency root mean square, the instantaneous energy of the first component of CEEMDAN, the frequency standard deviation, the wavelet singular spectrum entropy, the valley factor, the instantaneous energy of the second component of CEEMDAN, the wavelet energy entropy, the wavelet scale entropy of the first subband of wavelet decomposition, the wavelet scale entropy of the second subband of wavelet decomposition, the form factor, the pulse factor, the peak factor, the instantaneous energy of the fourth component of CEEMDAN, the minimum value, and the kurtosis. The Pearson correlation coefficient of the permutation entropy and other features is the weakest in the extracted features, which shows that the features have smaller linear correlation with other features and is beneficial to final classification.
Specifically, 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transform feature vectors and 10 complete integrated empirical mode decomposition (CEEMDAN) feature vectors of adaptive noise are extracted from an original signal, and a bearing fault mixed domain feature set is constructed by combining the extracted feature parameters, and the method specifically comprises the following steps:
step 1: the time domain signal is a visual embodiment of the vibration signal, and some useful features can be extracted from the time domain. By xiTo represent a time series of the acquired vibration signals (i ═ 1,2, … N), where N is xrThe sampling points of (a). The time domain features are calculated separately using the following equation:
(1) standard deviation:
s represents the standard deviation of the measured signal,
is a mean value
(2) The degree of effectiveness:
Xrmssignificance value X representing significance, vibration signal of bearingrmsTend to correlate well with the wave shape of random vibrations produced by the irregularity of the bearing surface.
(3) Skewness:
the skewness a reflects the asymmetry to the ordinate, which is more asymmetric if a is larger.
(4) Kurtosis:
kurtosis coefficient kvAnd taking the 4 th power of the pulse response amplitude as a judgment basis, and enlarging the difference between the pulse signal and the background noise to improve the signal-to-noise ratio so as to represent the occurrence probability of the large-amplitude pulse signal. The method is sensitive to early faults of the bearing and is an important basis for simple diagnosis of the faults of the bearing.
(5) Peak value:
Xpeak＝max(xi)
Xpeakindicating a peak value and max indicating a large value for the sampling point. Peak value XpeakThe signal is obtained by averaging a series of maximum instantaneous amplitude values of the signal, the signal strength can be reflected, and the method has better applicability to faults which can generate instantaneous impact, such as surface pitting and the like.
(6) Peak-to-peak value:
Xvpp＝max(xi)-min(xi)
Xvppand represents the peak-to-peak value, which is the difference between the highest value and the lowest value of the signal in a period, i.e. the range between the maximum value and the minimum value.
(7) Form factor:
the form factor is significance (X)rms) And absolute mean value
The ratio of.
(8) Pulse factor:
the pulse factor I being the peak value XpeakAnd absolute mean value
The ratio of. As with the peak, the pulse factor has a greater correlation with the instantaneous impact produced by the fault. Research shows that the pulse factor can rise along with the increase of the fault in the early stage of the fault, and can be weakened after rising to a certain degree.
(9) Crest factor:
the crest factor C is the crest value XrmsDivided by the degree of effectiveness Xrms. Compared with the peak value, the peak factor is not influenced by the magnitude of the signal amplitude, so that the dependence on the sensitivity of the sensor is relatively low, and even if the sensitivity of the sensor changes, the peak factor does not change greatly.
(10) Allowance is as follows:
Xpeakdenotes the peak value, XrRepresenting the square root amplitude, XpIndicating a peak. When the information contained in the time signal is not from one part or component but belongs to a plurality of elements, information from components such as a high-speed gear, a low-speed gear, and a bearing is often contained in the vibration signal of the multi-stage gear, and in this case, the fault diagnosis or analysis can be performed using the form factor S, the pulse factor I, the margin L, and the like.
(11) Maximum value:
TF＝max{xi}
(12) minimum value:
TF＝min{xi}
(13) mean value:
(14) absolute mean value
(15) Square root amplitude:
(16) variance:
step 2: when the bearing fault characteristics are extracted, the information contained in the time domain characteristics is not enough, and the frequency domain characteristics of the original vibration signals can be extracted to provide more information for people. The invention extracts 4 commonly used frequency domain feature vectors to form a frequency domain feature set, processes an original signal by Fourier analysis before constructing the frequency domain set, expresses the frequency in u (i) of the following four formulas,
representing the average frequency.
(1) Frequency of center of gravity
(2) Mean frequency
(3) Root mean square frequency
(4) Standard deviation of frequency
Wherein FFCRepresenting frequency of center of gravity
And step 3: wavelet transform feature set
Wavelet analysis is the use of a set of wavelet functions to describe or approximate a signal. Transient signals or mutation signals are key point information of the rolling bearing, but the signals are not only subjected to overall analysis, but also subjected to local analysis, and a single time domain or frequency domain cannot meet the requirement, so that the method introduces a wavelet transformation characteristic set, wherein the wavelet transformation characteristic set comprises 1 wavelet energy entropy, 4 wavelet scale entropies and 1 wavelet singular spectrum entropy to comprehensively depict the time-frequency domain characteristics of the signals.
(1) Wavelet energy entropy
Firstly, decomposing the original signal x (t) into j layers, and decomposing into 2jA node, each frequency band signal being represented by a node coefficient, wherein Cm,k(m＝0,1,2......j；k＝0,1,2....2k-1) Representing the kth node of the mth layer. Secondly, the small packet wave coefficient D of each frequency bandm,kExtracting from high to low and calculating energy value E of each frequency band of m-th layerm,kAnd energy entropy Hm,k。
Energy entropy Hm,kCalculating the formula:
wherein
Em,k＝∫|Dm,k(t)|2dt
In the formula, pm,kThe k frequency band of the mth layer accounts for the proportion of the total energy; e is the total energy.
(2) Entropy of wavelet scale
The invention utilizes the noise reduction characteristic of wavelet correlation filtering method to carry out wavelet correlation filtering noise reduction processing on the original signal x (t) to obtain the scale domain wavelet coefficient D with higher signal-to-noise ratioj＝{dj(k) N, j 1,2.. j, and a scaling factor CmEntropy of wavelet scale WCFSEjCalculating the formula:
wherein
dF(j)(k) Is represented by dj(k) Fourier transform of
Defined wavelet dependent feature scale entropy WCFSEjThe method is characterized in that the uniformity of the signal energy distribution on the scale j describes the complexity of the signal on the scale j, so that the fault characteristics are quantized.
(3) Entropy of wavelet singular spectrum
The wavelet packet singular spectrum entropy is a spectrum analysis method which integrates wavelet packet transformation, singular value decomposition and information entropy theory, and the basic idea is to transform a coefficient matrix of a vibration signal subjected to wavelet packet decomposition and reconstruction into a series of singular values capable of reflecting basic characteristics of an original signal, and then solve information entropy for the singular value set, thereby providing a certain measure for the complexity of the vibration signal. The wavelet singular spectrum entropy calculation formula is as follows:
Si＝-gilog2gi
and 4, step 4: integrated empirical mode decomposition (CEEMDAN) of adaptive noise
The CEEMDAN method is based on EMD. By adding white noise in a self-adaptive manner, the problem of mode aliasing of the EEMD method is solved, a better mode separation spectrum is obtained, and meanwhile, the operation efficiency is improved.
Let Ek () be the generation operator of the kth IMF from EMD, and let w (i) be the realization of zero mean unit variance white noise. For a given signal x, the main steps of CEEMDAN are as follows:
(1) for signal xi＝x+β0ω(i)I1, 2, …, I, the first obtained by EMD decomposition
The first IMF of CEEMDAN is defined as:
(2) the first residual is calculated as:
r1＝x-d1
(3) for the signal r1+β1E1(ω(i)) I 1,2, … … i, the first IMF was obtained by EMD. The second IMF of CEEMDAN is defined as:
(4) calculate the KTH residual as:
rk＝r(k-1)-dk
where K is 2, … K,
(5) for the signal rk+βkEk(ω(i)) I ═ 1,2, … … i. definitions CEEMDAN (k +1) times IMF:
(6) next k to step 4
And (4) repeating the steps from 4 to 6 until the obtained residual error cannot be decomposed by EMD, or the IMF condition is met, or the number of local extreme values of the residual error is less than 3. The final residuals are:
k is the total number of IMFs, then the signal x can be expressed as:
the coefficient betak＝εkstd(rk) Allows the selection of the signal-to-noise ratio in each iteration, where std (.) is the standard deviation operator.
Specifically, the specific steps of performing correlation analysis on the extracted mixed domain specific diagnosis by using a pearson correlation coefficient, extracting a low-dimensional main feature vector which is easy to identify from a high-dimensional fault feature set, and finally introducing the low-dimensional feature set into a random forest as an input of pattern identification are as follows:
the pearson correlation coefficient is based on a concept proposed by karl pearson. The pearson correlation coefficient is defined as the pearson correlation coefficient between the rank variables. For a sample with a sample capacity of n, n original data are converted into grade data, and a correlation coefficient r is
xi，yiTaking the feature vectors of two different mixed domain signals respectively, (i ═ 1,2.. N),
r represents the degree of linear correlation between two variables, which is the mean of the feature vectors. The value of r is between-1 and + 1. If r>0, which means that two variables are positively correlated, i.e., the larger the value of one variable, the larger the value of the other variable; if r is<0, then two variables are represented. The variables are inversely related, i.e., the larger the value of one variable, the smaller the value of the other variable. The larger the absolute value of r, the stronger the correlation. The larger the absolute value of the correlation coefficient is, the stronger the correlation is, the closer the correlation coefficient is to 1 or-1, the stronger the correlation is, the closer the correlation coefficient is to 0, and the weaker the correlation is.
Generally, the correlation strength of a variable is judged by the following value range: 0.8-1.0 are very strongly correlated; 0.6-0.8 strong correlation; 0.4-0.6 moderate correlation; 0.2-0.4 weakly correlated; 0.0-0.2 are very weakly or not correlated. Meanwhile, when r is larger than 0 and smaller than 1, the positive correlation between x and y is shown, when r is larger than 1 and smaller than 0, the negative correlation between x and y is shown, r is 1, which means that x and y are completely positive correlations, and r is 1, which means that x and y are completely negative correlations. When r is 0, it means that x and y are irrelevant.
The invention has the beneficial effects that:
1. the fault diagnosis model and the optimization method of the rolling bearing are provided, so that the fault diagnosis process is expressed clearly and accurately, and the diagnosis accuracy is improved.
2. The construction method of the mixed domain feature set is provided, and the defect that the single domain feature cannot reflect the bearing fault comprehensively is overcome.
3. And performing correlation analysis on the acquired mixed domain by adopting a Pearson feature selection method, extracting features with weak correlation and low-dimensional features, inputting the features and the low-dimensional features into a random forest classifier, reducing the dimension of the extracted data on one hand, removing redundant features, and improving the diagnosis precision of fault diagnosis on the other hand.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is an exploded view of a wavelet under normal operating conditions;
FIG. 3 is a wavelet exploded view under a rolling element fault condition;
FIG. 4 is an exploded view of wavelets under inner ring fault conditions;
FIG. 5 is a wavelet exploded view under the outer ring fault condition;
FIG. 6 is an exploded view of CEEMDAN under normal operating conditions;
FIG. 7 is an exploded view of CEEMDAN under a rolling element fault condition;
FIG. 8 is an exploded view of CEEMDAN under an inner ring fault condition;
FIG. 9 is an exploded view of CEEMDAN under an outer ring fault condition;
fig. 10 pearson correlation coefficient confusion matrix diagram.
Detailed Description
The invention is further described with reference to the following figures and specific examples.
Example 1: as shown in fig. 1 to 10, a mixed domain characteristic bearing fault diagnosis method based on pearson correlation coefficient includes the following steps: firstly, extracting 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transformation feature vectors and 10 complete integration empirical mode decomposition CEEMDAN feature vectors of self-adaptive noise from an original signal, and constructing a bearing fault mixed domain feature set by combining the extracted feature parameters; secondly, performing relevance analysis on the extracted mixed domain features by using a Pearson correlation coefficient, and extracting low-dimensional principal feature vectors which are easy to identify from a high-dimensional fault feature set; and finally, importing the low-dimensional feature set into a random forest as the input of pattern recognition.
Further, extracting 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transformation feature vectors and 10 complete integrated empirical mode decomposition CEEMDAN feature vectors of self-adaptive noise from the original signal, and combining the extracted feature parameters to construct a bearing fault mixed domain feature set, wherein the specific steps of the feature set are as follows:
step 1: by xiTo represent a time series of acquired vibration signals (i ═ 1,2.. N), N being xiRespectively calculating time domain characteristics by using the following formula:
(1) standard deviation:
s represents the standard deviation of the measured signal,
is an average value;
(2) the degree of effectiveness:
Xrmsindicates the degree of effectiveness;
(3) skewness:
the skewness a reflects the asymmetry of the ordinate, and if a is larger, the asymmetry is more serious;
(4) kurtosis:
kurtosis coefficient kvTaking the 4 th power of the pulse response amplitude as a judgment basis, and enlarging the difference between the pulse signal and the background noise to improve the signal-to-noise ratio so as to represent the occurrence probability of the large-amplitude pulse signal;
(5) peak value:
Xpeak＝max(xi)
Xpeakdenotes the peak value, max denotes the large value of the sampling point, peak value XpeakAveraging a series of maximum instantaneous amplitudes of the signal;
(6) peak-to-peak value:
Xvpp＝max(xi)-min(xi)
Xvppthe peak value is the difference value between the highest value and the lowest value of the signal in a period, namely the range between the maximum value and the minimum value;
(7) form factor:
the form factor is significance (X)rms) And absolute mean value
The ratio of (A) to (B);
(8) pulse factor:
the pulse factor I being the peak value XpeakAnd absolute mean value
In comparison, the pulse factor rises with the increase of the fault in the early stage of the fault, and is weakened when rising to a certain degree;
(9) crest factor:
the crest factor C is the crest value XrmsDivided by the degree of effectiveness Xrms；
(10) Allowance is as follows:
Xpeakdenotes the peak value, XrRepresenting the square root amplitude, and performing fault diagnosis or analysis by using the form factor S, the pulse factor I and the margin L when the information contained in the time signal is not from one part or component but belongs to a plurality of elements;
(11) maximum value:
TF＝max{xi}
(12) minimum value:
TF＝min{xi}
(13) mean value:
(14) absolute mean value
(15) Square root amplitude:
(16) variance:
step 2: the invention extracts 4 commonly used frequency domain feature vectors to form a frequency domain feature set, processes an original signal by Fourier analysis before constructing the frequency domain set, expresses the frequency in u (i) of the following four formulas,
represents the average frequency;
(1) frequency of center of gravity
(2) Mean frequency
(3) Root mean square frequency
(4) Standard deviation of frequency
And step 3: wavelet transform feature set
(1) Wavelet energy entropy
First, the original signal x (t) is decomposed intoj layer, decomposed to 2jA node, each frequency band signal being represented by a node coefficient, wherein Cm,k(m＝0,1,2......j；k＝0,1,2....2k-1) The kth node of the mth layer is represented, and the wavelet coefficient D of each frequency band is then determinedm,kExtracting from high to low and calculating energy value E of each frequency band of m-th layerm,kAnd energy entropy Hm,k；
Energy entropy Hm,kCalculating the formula:
wherein
Em,k＝∫|Dm,k(t)|2dt
In the formula, pm,kThe k frequency band of the mth layer accounts for the proportion of the total energy; e is the total energy of the whole body,
(2) entropy of wavelet scale
The invention utilizes the noise reduction characteristic of wavelet correlation filtering method to carry out wavelet correlation filtering noise reduction processing on the original signal x (t) to obtain the scale domain wavelet coefficient D with higher signal-to-noise ratioj＝{dj(k) K 1,2, … N, j 1,2 … m, and a scaling factor CmEntropy of wavelet scale WCFSEjCalculating the formula:
wherein
dF(j)(k) Is represented by dj(k) Fourier transform of
(3) Entropy of wavelet singular spectrum
The wavelet singular spectrum entropy calculation formula is as follows:
Si＝-gilog2gi
and 4, step 4: integrated empirical mode decomposition (CEEMDAN) of adaptive noise
Let Ek (.) be the generation operator of the kth IMF from EMD and w (i) be the realization of zero mean unit variance white noise, the main steps of CEEMDAN for a given signal x are as follows:
(1) for signal xi＝x+β0ω(i)I1, 2, …, I, the first obtained by EMD decomposition
The first IMF of CEEMDAN is defined as:
(2) the first residual is calculated as:
r1＝x-d1
(3) for the signal r1+β1E1(ω(i)) 1,2, … … i, the first IMF was obtained by EMD, the second IMF of CEEMDAN was defined as:
(4) calculate the KTH residual as:
rk＝r(k-1)-dk
where K is 2, … K,
(5) for the signal rk+βkEk(ω(i)) I ═ 1,2, … … i. definitions CEEMDAN (k +1) times IMF:
(6) next k to step 4
Repeating the steps 4 to 6 until the obtained residual error cannot be decomposed by EMD, or the obtained residual error meets the IMF condition, or the number of local extreme values of the residual error is less than 3, and the final residual error is:
k is the total number of IMFs, then the signal x can be expressed as:
the coefficient betak＝εkstd(rk) Allows the selection of the signal-to-noise ratio in each iteration, where std (.) is the standard deviation operator.
Further, performing relevance analysis on the extracted mixed domain special diagnosis by using a Pearson correlation coefficient, extracting a low-dimensional main feature vector which is easy to identify from a high-dimensional fault feature set, and finally introducing the low-dimensional feature set into a random forest as input of pattern identification;
the method comprises the following specific steps:
for a sample with a sample capacity of n, n original data are converted into grade data, and a correlation coefficient r is
xi，yiTaking the feature vectors of two different mixed domain signals respectively, (i ═ 1,2.. N),
r is the mean of the feature vectors and represents the degree of linear correlation between the two variables, with r having a value between-1 and + 1.
Example 2: in this example, the method shown in embodiment 1 is used for fault diagnosis of the bearing, and the specific implementation steps are as follows:
(1) the experiment adopts a bearing fault data set collected by a bearing data center of Kaiser university. The data set is fan end bearing data under 12K sampling frequency, failure data of 4 different states including normal, inner ring failure, rolling element failure and outer ring failure (6 o' clock direction) are respectively collected, each state except the normal data has 3 failure depth types, the diameters are 0.1778MM, 0.3556MM and 0.5334MM respectively, the load of a bearing motor is 0, the rotating speed of the bearing is 1797r/min, and 10 types of failure types are taken as data sources of the experiment. Each class of data is divided into 115 classification samples for a total of 1150 samples of 10 classes. The training set size is 700 parts per class, 70 parts per class, the test set 450 parts per class 45 parts. The classification is shown in table 1, wherein RF, IF, and OF are faults OF the rolling element, the inner ring, and the outer ring (at 6 o' clock direction), and the specific data is shown in table 1:
TABLE 1 experimental number of rolling bearings
(2) Feature extraction
The experiment characteristic extraction extracts a series of new characteristics from an original time signal sequence through function mapping, the total length of each type of original time sequence in the experiment is 117760, the original time sequence is divided into 1150 parts, the length of each part is 1024, and the time domain, the frequency domain and the time-frequency domain of each time sequence are respectively extracted to obtain 37 characteristics. The original feature vector is marked as a1, a2, …, An, n is 1024, the new feature vector is extracted and is shown as B1, B2, …, Bm, m is 37(m < n), fi is the mapping function corresponding to the new feature vector, and can be shown as
Bi＝fi(A1,A2,...,Aj),i∈[1,m]
And extracting time domain features and numbering. 1-10 are dimensional time domain features, which are respectively a maximum value, a minimum value, a peak-to-peak value, an average value, an absolute average value, a square root amplitude value, a variance, a standard deviation and an effective value. 11-17 are dimensionless time domain features, which are kurtosis, skewness, form factor, peak factor, pulse factor, margin factor, and clearance factor, respectively, and are recorded as follows. The frequency domain features are numbered 18-22, which are the average frequency, the center of gravity frequency, the root mean square of the frequency, and the standard deviation of the frequency, respectively.
The time-frequency characteristics mainly extract the wavelet transform and CEEMDAN related characteristics. In the wavelet transformation, three-layer decomposition is carried out on an original vibration signal, and the original vibration signal is divided into 8 sub-frequency bands. The specific decomposition effect is shown in fig. 2-5, and it can be seen that the first four frequency bands contain most of the energy of the original signal, so we extract the wavelet scale entropy of the first four sub-bands, and then extract the wavelet energy spectral entropy and the wavelet singular entropy of the signal, to form one time-frequency feature subset.
Secondly, the original signals of the 10 classes of signals are respectively subjected to CEEMDAN decomposition, four classes of the original signals are taken out, and 9 natural modal component decomposition effects are obtained and are shown in FIGS. 6 to 9. The 1 st component signal has a significantly smaller amplitude and faster vibration frequency than other components, and therefore, the component can be judged to be random noise and not selected as a component for feature extraction. The 2 nd to 6 th components have slow fluctuation, most of time domain waveform diagrams are composed of higher harmonics, the frequency spectrum energy is concentrated, most of time domain waveforms are close to sine waves, and the time domain waveforms are ideal signal analysis time sequences. The residual component contains less characteristic information, has small correlation with the original signal, lacks the physical analysis significance, belongs to a false component and is removed. Thus extracting the permutation entropy and instantaneous energy of the 2-6 component signals respectively to form another subset of the time-frequency characteristics.
(3) Analysis of experiments
Normalization is a general step of data processing, and has the advantages of accelerating the optimization speed of gradient descent solution after feature processing and improving the algorithm classification precision. Therefore, firstly, the extracted time-frequency domain characteristics are normalized, and in the experiment, 0,1 is selected as the value range of the normalization.
Secondly, performing correlation analysis on the extracted time-frequency features, respectively calculating a Pearson coefficient value of each feature vector, setting a Pearson correlation coefficient threshold value to be 0.45, respectively counting the strength of the correlation between each type of features and other features, if the correlation is more than 0.45, indicating that the correlation has strong correlation, setting the specific relation between the features to be 1, if the correlation is less than 0.45, indicating that the correlation has weak correlation between the features, setting the correlation to be 0, counting the number of the features of each type to be 1, sequencing according to the number, and extracting a plurality of features with the largest number as classification data sources of the experiment. In this experiment, since the feature vectors were normalized to between [0,1] and were all positive values, the extracted features were also positive values in the range of [0,1] after correlation analysis.
As shown in fig. 10, a characteristic confusion matrix with a Pearson Correlation Coefficient (PCC) greater than a threshold is selected, and from top to bottom, it is shown that the correlation coefficient is from weak to strong, and the characteristics are: the first component permutation entropy value of CEEMDAN, the second component permutation entropy value of CEEMDAN, the third component permutation entropy value of CEEMDAN, the fourth component permutation entropy value of CEEMDAN, the fifth component permutation entropy value of CEEMDAN, the barycentric frequency, the frequency root mean square, the instantaneous energy of the first component of CEEMDAN, the frequency standard deviation, the wavelet singular spectrum entropy, the valley factor, the instantaneous energy of the second component of CEEMDAN, the wavelet energy entropy, the wavelet scale entropy of the first subband of wavelet decomposition, the wavelet scale entropy of the second subband of wavelet decomposition, the form factor, the pulse factor, the peak factor, the instantaneous energy of the fourth component of CEEMDAN, the minimum value, and the kurtosis. The Pearson correlation coefficient of the permutation entropy and other features is the weakest in the extracted features, which shows that the features have smaller linear correlation with other features and is beneficial to final classification. In order to prove the effectiveness of the feature extraction method, the invention adopts 7 feature processing methods of time domain, frequency domain, time-frequency domain, MIC (maximum information coefficient), PCA and mixed domain (MF) to carry out comparison experiments, in order to adopt a random forest classifier (rf) to carry out classification, each comparison experiment is operated for 5 times, the average value is taken, and the specific classification conditions are shown in Table 2:
TABLE 27 model Performance comparisons
From the analysis of the table above, under the same condition and with the same classifier, the mixed domain feature extraction method based on the pearson correlation coefficient provided by the invention has average accuracy rates higher than those of other models and has shorter running time than those of other models, wherein the average accuracy rates exceed 97%. Although the operation time of the classification method based on the frequency domain is short, the model has low accuracy and does not have practical engineering application value, and the MIC-rf model has the longest operation time and relatively low accuracy in all models due to the long calculation time of the MIC approximation algorithm. Although the PCA-rf has short running time, the data after dimensionality reduction cannot well represent the rule of the original time sequence, and the classification accuracy is low. The model of the invention is based on the mixed domain-rf model for feature selection, and as can be seen from the table, after the feature selection is carried out by the Pearson correlation coefficient, the accuracy rate is not reduced but is slightly improved, which shows that in the mixed domain formed by the 41 original features extracted by the invention, individual features have stronger correlation with other features, thereby influencing the classification accuracy rate, but after the correlation analysis, the data with higher correlation can be eliminated, the feature rule of the data is well fitted, thereby achieving a good classification effect.
Claims (3)
1. A mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficients is characterized in that: the method comprises the following steps: firstly, extracting 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transformation feature vectors and 10 complete integration empirical mode decomposition CEEMDAN feature vectors of self-adaptive noise from an original signal, and constructing a bearing fault mixed domain feature set by combining the extracted feature parameters; secondly, performing relevance analysis on the extracted mixed domain features by using a Pearson correlation coefficient, and extracting low-dimensional principal feature vectors which are easy to identify from a high-dimensional fault feature set; and finally, importing the low-dimensional feature set into a random forest as the input of pattern recognition.
2. The method for diagnosing the fault of the mixed domain characteristic bearing based on the Pearson correlation coefficient as claimed in claim 1, wherein: the method comprises the following specific steps of extracting 6 time domain dimensionless vectors, 10 time domain dimensionless vectors, 4 frequency domain feature vectors, 6 wavelet transformation feature vectors and 10 complete integration empirical mode decomposition CEEMDAN feature vectors of self-adaptive noise from original signals, and constructing a bearing fault mixed domain feature set by combining the extracted feature parameters:
step 1: by xiTo represent a time series of acquired vibration signals (i ═ 1,2.. N), N being xiRespectively calculating time domain characteristics by using the following formula:
(1) standard deviation:
s represents the standard deviation of the measured signal,
is an average value;
(2) the degree of effectiveness:
Xrmsindicates the degree of effectiveness;
(3) skewness:
the skewness a reflects the asymmetry of the ordinate, and if a is larger, the asymmetry is more serious;
(4) kurtosis:
kurtosis coefficient kvTaking the 4 th power of the pulse response amplitude as a judgment basis, and enlarging the difference between the pulse signal and the background noise to improve the signal-to-noise ratio so as to represent the occurrence probability of the large-amplitude pulse signal;
(5) peak value:
Xpeak＝max(xi)
Xpeakrepresents a peakThe value, max, represents the maximum value, peak X, of the sampling pointpeakAveraging a series of maximum instantaneous amplitudes of the signal;
(6) peak-to-peak value:
Xvpp＝max(xi)-min(xi)
Xvppthe peak value is the difference value between the highest value and the lowest value of the signal in a period, namely the range between the maximum value and the minimum value;
(7) form factor:
the form factor is significance (X)rms) And absolute mean value
The ratio of (A) to (B);
(8) pulse factor:
the pulse factor I being the peak value XpeakAnd absolute mean value
In comparison, the pulse factor rises with the increase of the fault in the early stage of the fault, and is weakened when rising to a certain degree;
(9) crest factor:
the crest factor C is the crest value XrmsDivided by the degree of effectiveness Xrms；
(10) Allowance is as follows:
Xpeakdenotes the peak value, XrRepresenting the square root amplitude, and performing fault diagnosis or analysis by using the form factor S, the pulse factor I and the margin L when the information contained in the time signal is not from one part or component but belongs to a plurality of elements;
(11) maximum value:
TF＝max{xi}
(12) minimum value:
TF＝min{xi}
(13) mean value:
(14) absolute mean value
(15) Square root amplitude:
(16) variance:
step 2: the method comprises the steps of extracting 4 common frequency domain feature vectors to form a frequency domain feature set, processing an original signal by using Fourier analysis before constructing the frequency domain feature set, and expressing the frequency in u (i) in the following four formulas,
represents the average frequency;
(1) frequency of center of gravity
(2) Mean frequency
(3) Root mean square frequency
(4) Standard deviation of frequency
And step 3: wavelet transform feature set
(1) Wavelet energy entropy
Firstly, decomposing the original signal x (t) into j layers, and decomposing into 2jA node, each frequency band signal being represented by a node coefficient, wherein Cm，k(m＝0，1，2……j；k＝0，1，2…2k-1) The kth node of the mth layer is represented, and the wavelet coefficient D of each frequency band is then determinedm，kExtracting from high to low and calculating energy value E of each frequency band of m-th layerm，kAnd energy entropy Hm，k；
Energy entropy Hm，kCalculating the formula:
wherein
Em，k＝∫|Dm，k(t)|2dt
In the formula, pm，kThe k frequency band of the mth layer accounts for the proportion of the total energy; e is total energy, (2) wavelet scale entropy
The method utilizes the noise reduction characteristic of a wavelet correlation filtering method to perform wavelet correlation filtering noise reduction processing on an original signal x (t) to obtain a scale domain wavelet coefficient D with higher signal-to-noise ratioj＝{dj(k) N, j 1,2 … m, and a scaling factor CmEntropy of wavelet scale WCFSEjCalculating the formula:
wherein
dF(j)(k) Is represented by dj(k) Fourier transform of
(3) Entropy of wavelet singular spectrum
The wavelet singular spectrum entropy calculation formula is as follows:
Si＝-gilog2gi
and 4, step 4: complete integrated empirical mode decomposition (CEEMDAN) of adaptive noise
Let Ek (.) be the generation operator of the kth IMF from EMD and w (i) be the realization of zero mean unit variance white noise, the main steps of CEEMDAN for a given signal x are as follows:
(1) for signal xi＝x+β0ω(i)I1, 2, …, I, the first obtained by EMD decomposition
The first IMF of CEEMDAN is defined as:
(2) the first residual is calculated as:
r1＝x-d1
(3) for the signal r1+β1E1(ω(i)) 1,2, … … i, the first IMF was obtained by EMD, the second IMF of CEEMDAN was defined as:
(4) calculate the KTH residual as:
rk＝r(k-1)-dk
where K is 2, … K,
(5) for the signal rk+βkEk(ω(i)) I ═ 1,2, … … i. definitions CEEMDAN (k +1) times IMF:
(6) next k to step 4
Repeating the steps 4 to 6 until the obtained residual error cannot be decomposed by EMD, or the obtained residual error meets the IMF condition, or the number of local extreme values of the residual error is less than 3, and the final residual error is:
k is the total number of IMFs, then the signal x can be expressed as:
the coefficient betak＝εkstd(rk) Allows the selection of the signal-to-noise ratio in each iteration, where std (.) is the standard deviation operator.
3. The method for diagnosing the fault of the mixed domain characteristic bearing based on the Pearson correlation coefficient as claimed in claim 1, wherein: performing correlation analysis on the extracted mixed domain specific diagnosis by using a Pearson correlation coefficient, extracting a low-dimensional main feature vector which is easy to identify from a high-dimensional fault feature set, and finally introducing the low-dimensional feature set into a random forest as input of mode identification;
the method comprises the following specific steps:
for a sample with a sample capacity of n, n original data are converted into grade data, and a correlation coefficient r is
xi，yiTaking the feature vectors of two different mixed domain signals respectively, (i ═ 1,2.. N),
r is the mean of the feature vectors and represents the degree of linear correlation between the two variables, with r having a value between-1 and + 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110737611.4A CN113420691A (en) | 2021-06-30 | 2021-06-30 | Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110737611.4A CN113420691A (en) | 2021-06-30 | 2021-06-30 | Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113420691A true CN113420691A (en) | 2021-09-21 |
Family
ID=77717341
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110737611.4A Pending CN113420691A (en) | 2021-06-30 | 2021-06-30 | Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113420691A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114354188A (en) * | 2022-01-10 | 2022-04-15 | 郑州大学 | Rotary machine fault diagnosis method based on fully-adaptive noise set empirical mode decomposition |
| CN114403831A (en) * | 2022-03-25 | 2022-04-29 | 广东玖智科技有限公司 | PPG waveform pulse extraction method and device |
| CN114580828A (en) * | 2021-12-29 | 2022-06-03 | 吉林大学 | Safety evaluation method for numerical control machine during transportation process |
| CN115046764A (en) * | 2022-05-05 | 2022-09-13 | 燕山大学 | Early fault diagnosis method for rolling bearing |
| CN115438035A (en) * | 2022-10-27 | 2022-12-06 | 江西师范大学 | Data exception handling method based on KPCA and mixed similarity |
| CN115508703A (en) * | 2022-09-19 | 2022-12-23 | 中国人民解放军陆军装甲兵学院 | Multi-source information fusion motor fault diagnosis method and system |
| CN116148656A (en) * | 2023-04-20 | 2023-05-23 | 国网山东省电力公司莱州市供电公司 | Portable analog breaker fault detection method |
| CN116399589A (en) * | 2023-03-29 | 2023-07-07 | 哈尔滨理工大学 | Rolling bearing weak signal feature extraction method based on RCMWE |
| CN116955117A (en) * | 2023-09-18 | 2023-10-27 | 深圳市艺高智慧科技有限公司 | Computer radiator performance analysis system based on data visualization enhancement |
-
2021
- 2021-06-30 CN CN202110737611.4A patent/CN113420691A/en active Pending
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114580828A (en) * | 2021-12-29 | 2022-06-03 | 吉林大学 | Safety evaluation method for numerical control machine during transportation process |
| CN114354188A (en) * | 2022-01-10 | 2022-04-15 | 郑州大学 | Rotary machine fault diagnosis method based on fully-adaptive noise set empirical mode decomposition |
| CN114403831A (en) * | 2022-03-25 | 2022-04-29 | 广东玖智科技有限公司 | PPG waveform pulse extraction method and device |
| CN115046764A (en) * | 2022-05-05 | 2022-09-13 | 燕山大学 | Early fault diagnosis method for rolling bearing |
| CN115508703A (en) * | 2022-09-19 | 2022-12-23 | 中国人民解放军陆军装甲兵学院 | Multi-source information fusion motor fault diagnosis method and system |
| CN115438035A (en) * | 2022-10-27 | 2022-12-06 | 江西师范大学 | Data exception handling method based on KPCA and mixed similarity |
| CN116399589A (en) * | 2023-03-29 | 2023-07-07 | 哈尔滨理工大学 | Rolling bearing weak signal feature extraction method based on RCMWE |
| CN116399589B (en) * | 2023-03-29 | 2024-01-12 | 哈尔滨理工大学 | Rolling bearing weak signal feature extraction method based on RCMWE |
| CN116148656A (en) * | 2023-04-20 | 2023-05-23 | 国网山东省电力公司莱州市供电公司 | Portable analog breaker fault detection method |
| CN116955117A (en) * | 2023-09-18 | 2023-10-27 | 深圳市艺高智慧科技有限公司 | Computer radiator performance analysis system based on data visualization enhancement |
| CN116955117B (en) * | 2023-09-18 | 2023-12-22 | 深圳市艺高智慧科技有限公司 | Computer radiator performance analysis system based on data visualization enhancement |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113420691A (en) | Mixed domain characteristic bearing fault diagnosis method based on Pearson correlation coefficient | |
| Qin et al. | The optimized deep belief networks with improved logistic sigmoid units and their application in fault diagnosis for planetary gearboxes of wind turbines | |
| Li et al. | The entropy algorithm and its variants in the fault diagnosis of rotating machinery: A review | |
| CN110470477B (en) | Rolling bearing fault diagnosis method based on SSAE and BA-ELM | |
| CN105275833B (en) | CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump | |
| CN109975013B (en) | IVMD-SE-based wind turbine generator gearbox fault feature extraction method | |
| CN110674892A (en) | Fault feature screening method based on weighted multi-feature fusion and SVM classification | |
| CN103968937B (en) | A kind of distribution switch mechanical state diagnostic method based on EMD Sample Entropies and FCM | |
| Li et al. | Bearing fault feature selection method based on weighted multidimensional feature fusion | |
| CN111504635B (en) | Planetary gear fault diagnosis method based on differential evolution probability neural network | |
| Li et al. | Multiscale symbolic diversity entropy: a novel measurement approach for time-series analysis and its application in fault diagnosis of planetary gearboxes | |
| CN111413089A (en) | Gear fault diagnosis method based on combination of VMD entropy method and VPMCD | |
| CN113780055B (en) | MOMEDA and compression sensing rolling bearing fault diagnosis method | |
| CN116226646B (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
| Chen et al. | Fault identification method for planetary gear based on DT-CWT threshold denoising and LE | |
| CN112461547A (en) | Bearing fault diagnosis method based on multi-island inheritance and variation modal decomposition | |
| CN115130495A (en) | Rolling bearing fault prediction method and system | |
| CN112101142A (en) | Slurry pump running state evaluation method, monitoring terminal and computer readable storage medium | |
| CN112182912B (en) | Manufacturing equipment spindle bearing health assessment method based on probability description and spectrum analysis | |
| CN107966287B (en) | Weak fault feature extraction method for self-adaptive electromechanical equipment | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| CN116186590A (en) | Bearing fault diagnosis method based on data layer feature fusion and convolutional neural network | |
| Liu et al. | Composite multi-scale basic scale Entropy based on CEEMDAN and its application in hydraulic pump fault diagnosis | |
| Luo et al. | A novel fault detection scheme using improved inherent multiscale fuzzy entropy with partly ensemble local characteristic-scale decomposition | |
| Han et al. | Fault diagnosis of rolling bearings using dual-tree complex wavelet packet transform and time-shifted multiscale range entropy |
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 |