CN108955855B - Vibration signal feature extraction method, monitoring method and device for rotary machine - Google Patents
Vibration signal feature extraction method, monitoring method and device for rotary machine Download PDFInfo
- Publication number
- CN108955855B CN108955855B CN201810586819.9A CN201810586819A CN108955855B CN 108955855 B CN108955855 B CN 108955855B CN 201810586819 A CN201810586819 A CN 201810586819A CN 108955855 B CN108955855 B CN 108955855B
- Authority
- CN
- China
- Prior art keywords
- vibration signal
- vibration
- period
- rotary machine
- feature extraction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H1/00—Measuring characteristics of vibrations in solids by using direct conduction to the detector
- G01H1/003—Measuring characteristics of vibrations in solids by using direct conduction to the detector of rotating machines
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H1/00—Measuring characteristics of vibrations in solids by using direct conduction to the detector
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01P—MEASURING LINEAR OR ANGULAR SPEED, ACCELERATION, DECELERATION, OR SHOCK; INDICATING PRESENCE, ABSENCE, OR DIRECTION, OF MOVEMENT
- G01P3/00—Measuring linear or angular speed; Measuring differences of linear or angular speeds
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a vibration signal feature extraction method, a vibration signal feature monitoring method and a vibration signal feature monitoring device for rotary machinery. The method for extracting the vibration signal features comprises the following steps of (1) constructing a codebook: using the collected vibration signals as training samples, clustering the training samples, and forming a group of code words by the clustering mass center and the clustering variance of each type so as to construct a codebook; step (2) dividing the cycle: carrying out periodic division on the vibration signal monitored in real time; step (3) calculating a correlation diagram; calculating an entropy value: and performing dimensionality reduction treatment on the obtained correlation diagram sequence by using the entropy value to obtain the characteristic reflecting the mechanical state operation trend. The correlation diagram is applied to the characteristic extraction process, and in the process of characteristic extraction depending on the amplitude of the vibration signal, the time sequence information of the original vibration signal is also considered, so that the extracted characteristic contains more effective information.
Description
Technical Field
The invention belongs to the field of rotary machinery, and particularly relates to a vibration signal feature extraction method, a vibration signal feature monitoring method and a vibration signal feature monitoring device for rotary machinery.
Background
In recent years, rotating machines have become more complex and larger, and have become more widely used, and any small problem may have serious consequences, so it is necessary to monitor the operating state of the rotating machine. The monitoring of the rotating speed of the rotating machine is an important problem, and the monitoring of the operating state of the rotating machine can be realized by monitoring the rotating speed of the rotating machine, and the monitoring device is widely applied to the fields of machining, digital control and the like.
The following problems exist in the rotating speed monitoring of the rotating machinery:
(1) in practical applications, it is difficult to monitor the rotation speed of the rotating machine directly by using a speedometer due to the limitations of the size, installation location and environment of the rotating machine.
(2) The rotating speed monitoring of the rotating machinery can be realized by analyzing the vibration signals generated by the speed change, and the current common vibration signal analysis methods comprise frequency domain analysis and time domain analysis.
(2.1) although the frequency domain analysis method based on fourier transform is widely used, the frequency domain analysis is not applicable when the amplitude of the acquired data is small or the amount of the acquired data is small.
(2.2) time domain analysis generally consists of two parts: and (4) feature extraction and decision making. The feature extraction plays an important role in the design of the detection algorithm and has an important influence on the final detection result. The existing time domain feature extraction methods have root mean square values, variances, kurtosis, skewness, peak values and the like, and all the feature extraction methods rely on the amplitude of the vibration signal to extract features, and neglect the time sequence information of the vibration signal.
Disclosure of Invention
In order to solve the defects of the prior art, a first object of the present invention is to provide a vibration signal feature extraction method for a rotating machine, which takes into account timing information of an original vibration signal and makes the extracted feature contain more effective information.
The invention relates to a vibration signal feature extraction method for rotary machinery, which comprises the following steps:
step (1), codebook construction: using the collected vibration signals as training samples, clustering the training samples, and forming a group of code words by the clustering mass center and the clustering variance of each type so as to construct a codebook;
step (2) dividing the cycle: carrying out periodic division on the vibration signal monitored in real time;
and (3) calculating a correlation diagram, which specifically comprises the following steps:
step (3.1): forming a group of vibration data by two vibration data with a preset time interval in a period;
step (3.2): randomly selecting two groups of code words in a codebook;
step (3.3): calculating a mapping value from one vibration data in each group of vibration data to a selected group of coding words and a mapping value from the other vibration data to another selected group of coding words by using a Gaussian kernel function, multiplying the two mapping values, and finally accumulating the product results of the mapping values of all the groups of vibration data to obtain a correlation diagram of the current period of the vibration signal;
step (3.4): repeating the steps (3.2) to (3.3) on the vibration signal of each period obtained by continuous monitoring to obtain a correlation diagram sequence with the same number as the period number;
calculating an entropy value: and performing dimensionality reduction treatment on the obtained correlation diagram sequence by using the entropy value to obtain the characteristic reflecting the mechanical state operation trend.
Further, in the step (1) of constructing the codebook, clustering the training samples by using a K-means algorithm.
The K-means algorithm is an algorithm that inputs the number K of clusters, and a database containing n data objects, and outputs K clusters satisfying the minimum variance criterion. The K-means algorithm accepts an input K; the n data objects are then divided into k clusters so that the obtained clusters satisfy: the similarity of objects in the same cluster is higher; while the object similarity in different clusters is smaller.
It should be noted that, in addition to clustering the training samples by using the K-means algorithm, the K-medoid algorithm or the CLARANS algorithm may also be used to cluster the training samples.
Further, in the step (2) of dividing the period, the vibration signal monitored in real time is divided periodically by using time series analysis, and the specific process includes:
step (2.1): determining a set of candidate vibration signal segments a based on the newly monitored vibration signal;
step (2.2): randomly selecting a candidate segment B with the end point being the same as the starting point of the selected vibration signal segment, and calculating the matching loss between the candidate segment B and the newly monitored vibration signal by using a dynamic time warping method;
step (2.3): the time period of the candidate segment B corresponding to the minimum matching loss is the optimal period;
step (2.4): the vibration signal is periodically divided according to the optimum period.
The method and the device utilize the optimal period to carry out period division on the vibration signal, and can accurately and quickly carry out feature extraction on the vibration signal.
It should be noted that, the preset period may also be used to perform periodic division on the vibration signal monitored in real time.
A second object of the present invention is to provide a vibration signal feature extraction apparatus for a rotary machine, which takes into account timing information of an original vibration signal and makes extracted features contain more effective information.
The invention relates to a vibration signal characteristic extraction device for rotary machinery, which comprises a vibration signal characteristic extraction processor, wherein the vibration signal characteristic extraction processor comprises:
a build codebook module configured to: using the collected vibration signals as training samples, clustering the training samples, and forming a group of code words by the clustering mass center and the clustering variance of each type so as to construct a codebook;
a split period module configured to: carrying out periodic division on the vibration signal monitored in real time;
a compute dependency graph module configured to:
forming a group of vibration data by two vibration data with a preset time interval in a period;
randomly selecting two groups of code words in a codebook;
calculating a mapping value from one vibration data in each group of vibration data to a selected group of coding words and a mapping value from the other vibration data to another selected group of coding words by using a Gaussian kernel function, multiplying the two mapping values, and finally accumulating the product results of the mapping values of all the groups of vibration data to obtain a correlation diagram of the current period of the vibration signal;
and obtaining a correlation diagram sequence with the same number as the vibration signal period number.
A compute entropy module configured to: and performing dimensionality reduction treatment on the obtained correlation diagram sequence by using the entropy value to obtain the characteristic reflecting the mechanical state operation trend.
Further, in the codebook building module, a K-means algorithm is used for clustering training samples.
The K-means algorithm is an algorithm that inputs the number K of clusters, and a database containing n data objects, and outputs K clusters satisfying the minimum variance criterion. The K-means algorithm accepts an input K; the n data objects are then divided into k clusters so that the obtained clusters satisfy: the similarity of objects in the same cluster is higher; while the object similarity in different clusters is smaller.
It should be noted that, in addition to clustering the training samples by using the K-means algorithm, the K-medoid algorithm or the CLARANS algorithm may also be used to cluster the training samples.
Further, in the split period module, the real-time monitored vibration signal is periodically divided by time-series analysis, and the split period module is further configured to:
determining a set of candidate vibration signal segments a based on the newly monitored vibration signal;
randomly selecting a candidate segment B with the end point being the same as the starting point of the selected vibration signal segment, and calculating the matching loss between the candidate segment B and the newly monitored vibration signal by using a dynamic time warping method;
the time period of the candidate segment B corresponding to the minimum matching loss is the optimal period;
the vibration signal is periodically divided according to the optimum period.
The method and the device utilize the optimal period to carry out period division on the vibration signal, and can accurately and quickly carry out feature extraction on the vibration signal.
It should be noted that, the preset period may also be used to perform periodic division on the vibration signal monitored in real time.
A third object of the present invention is to provide a method for monitoring an operating state of a rotary machine, in which a vibration signal of the rotary machine is analyzed by using a time domain analysis method, the obtained characteristic reflecting an operating trend of the machine state is obtained, an abnormality degree measurement is performed on the characteristic by using a euclidean distance, and finally, a decision is made on the obtained abnormality degree by using a hypothesis test, so that the operating state of the rotary machine can be quickly and accurately monitored.
The invention relates to a method for monitoring the running state of a rotating machine, which comprises the following steps:
and (3) extracting characteristics: the vibration signal characteristic extraction method facing the rotary machine is adopted to obtain the characteristic reflecting the running trend of the machine state;
calculating the degree of abnormality: based on the acquired characteristics of the reaction mechanical state operation trend, carrying out anomaly measurement on the reaction mechanical state operation trend by using the Euclidean distance;
a decision analysis step: and (4) adopting hypothesis testing to make a decision on the obtained abnormality degree.
Further, in the decision analysis step, it is checked whether a change in the operation state of the rotary machine has occurred using the 3 σ principle in the hypothesis test.
The hypothesis test is a method for deducing a population from a sample according to certain hypothesis in mathematical statistics. The specific method comprises the following steps: making some assumption about the population studied as required by the problem, denoted H0; selecting a suitable statistic chosen such that its distribution is known, assuming H0 holds; from the measured samples, the value of the statistic is calculated and tested against a predetermined level of significance, making a decision to reject or accept the hypothesis H0.
Hypothesis testing is an important element in sample inference. The method is a test method which makes an assumption whether a total index is equal to a certain numerical value or not and whether a certain random variable obeys a certain probability distribution or not according to raw data, then calculates statistic related to test by using a certain statistical method according to sample data, judges whether an estimated value and the total numerical value (or the estimated distribution and the actual distribution) have significant difference or not according to a certain probability principle and judges whether the raw data should be selected.
A fourth object of the present invention is to provide an operation state monitoring apparatus for a rotary machine, which analyzes a vibration signal of the rotary machine by using a time domain analysis method, obtains a characteristic reflecting an operation trend of a machine state, measures an abnormality degree of the characteristic by using a euclidean distance, and finally decides the obtained abnormality degree by using a hypothesis test, thereby rapidly and accurately monitoring an operation state of the rotary machine.
The invention relates to a device for monitoring the running state of a rotating machine, which comprises:
the vibration signal feature extraction device for the rotary machine;
vibration signal feature extraction element still links to each other with the state monitoring treater, the state monitoring treater includes:
a calculate outliers module configured to: based on the acquired characteristics of the reaction mechanical state operation trend, carrying out anomaly measurement on the reaction mechanical state operation trend by using the Euclidean distance;
a decision analysis module configured to: and (4) adopting hypothesis testing to make a decision on the obtained abnormality degree.
Further, in the decision analysis module, a 3 σ rule in hypothesis testing is employed to check whether a change in the operating state of the rotary machine has occurred.
The hypothesis test is a method for deducing a population from a sample according to certain hypothesis in mathematical statistics. The specific method comprises the following steps: making some assumption about the population studied as required by the problem, denoted H0; selecting a suitable statistic chosen such that its distribution is known, assuming H0 holds; from the measured samples, the value of the statistic is calculated and tested against a predetermined level of significance, making a decision to reject or accept the hypothesis H0.
Hypothesis testing is an important element in sample inference. The method is a test method which makes an assumption whether a total index is equal to a certain numerical value or not and whether a certain random variable obeys a certain probability distribution or not according to raw data, then calculates statistic related to test by using a certain statistical method according to sample data, judges whether an estimated value and the total numerical value (or the estimated distribution and the actual distribution) have significant difference or not according to a certain probability principle and judges whether the raw data should be selected.
Compared with the prior art, the invention has the beneficial effects that:
(1) in the invention, the correlation diagram is applied to the characteristic extraction process, and in the process of characteristic extraction depending on the amplitude of the vibration signal, the time sequence information of the original vibration signal is also considered, so that the extracted characteristic contains more effective information.
(2) In the invention, a vibration signal of the rotating machine is analyzed by using a time domain analysis method, the acquired characteristic reflecting the running trend of the machine state is measured by using the Euclidean distance, and finally, the obtained abnormal degree is decided by adopting hypothesis test, so that the running state of the rotating machine can be rapidly and accurately monitored.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
Fig. 1 is a flowchart of a vibration signal feature extraction method for a rotating machine according to the present invention.
Fig. 2 is a schematic diagram of the collected signals of the invention practically applied to the rotating speed monitoring of the rotating machinery.
Fig. 3 is a schematic diagram of feature extraction applied to rotating speed monitoring of a rotating machine in practice.
Fig. 4 is a schematic diagram of a vibration signal feature extraction processor according to the present invention.
Fig. 5 is a flowchart of an operation state monitoring method of a rotary machine of the present invention.
Fig. 6 is a schematic diagram of the degree of abnormality in the rotational speed monitoring of the rotary machine to which the present invention is actually applied.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Fig. 1 is a flowchart of a vibration signal feature extraction method for a rotating machine according to the present invention.
As shown in fig. 1, the method for extracting a vibration signal feature for a rotary machine according to the present invention includes steps (1) to (4). Specifically, the method comprises the following steps:
step (1), codebook construction: and clustering the training samples by using the acquired vibration signals as training samples, and forming a group of code words by using the clustering mass center and the clustering variance of each type so as to construct a codebook.
And (2) in the step (1) of constructing the codebook, clustering the training samples by using a K-means algorithm.
The K-means algorithm is an algorithm that inputs the number K of clusters, and a database containing n data objects, and outputs K clusters satisfying the minimum variance criterion. The K-means algorithm accepts an input K; the n data objects are then divided into k clusters so that the obtained clusters satisfy: the similarity of objects in the same cluster is higher; while the object similarity in different clusters is smaller.
It should be noted that, in addition to clustering the training samples by using the K-means algorithm, the K-medoid algorithm or the CLARANS algorithm may also be used to cluster the training samples.
Specifically, the step (1) of constructing the codebook includes:
step (1.1): the vibration signal x collected on the rotating machinei(i-1, 2 …, m) as a training sample set D-x1,x2,…,xm}; wherein m is a positive integer;
step (1.2): assuming that the number of clusters is k, randomly selecting k samples from a training sample set D as initial k centroids: { mu. }1,μ2,…,μkEach centroid corresponds to a class, and then there is k in totalClass, denoted as { Q1,Q2,…,Qk}；
Step (1.3): for i ═ 1,2, …, m, sample x is calculatediAnd each centroid muj(j ═ 1,2, …, k) and distance di,j：
di,j＝||xi-μj||2 2
X is to beiDivision into minimum di,jCorresponding class QjIn, and update Qj＝Qj∪{xi}；
Step (1.4): to QjRecalculate the centroid for all sample points:
step (1.5): iterating the two steps of the step (1.3) and the step (1.4) until the new centroid and the original centroid are equal to or smaller than a set threshold value, and calculating each class CjVariance of all samples in (1):
the centroid and the variance of each class can be obtained through the calculation, the centroid and the variance of each class form a group of coded words, and k groups of coded words are shared in the k classes, and the k groups of coded words form a codebook.
Step (2) dividing the cycle: and periodically dividing the vibration signal monitored in real time.
Specifically, in the step (2) of dividing the period, the vibration signal monitored in real time is periodically divided by using time series analysis, and the specific process includes:
step (2.1): based on newly monitored vibration signal X ═ { X1,x2,…,xZDetermining a set of candidate vibration signal segments a-Y1,…,yl}; wherein Z is a positive integer; is e [ l ∈ [ ]a,lb]. Wherein laAnd lbAre empirically estimated and are all positive integers.
Step (2.2): randomly selecting a candidate segment B with the end point being the same as the starting point of the selected vibration signal segment, and calculating the matching loss between the candidate segment B and the newly monitored vibration signal by using a dynamic time warping method;
for a candidate segment B-Y1,…,ylv}，lv∈[la,lb]，v∈[a,b]Using Dynamic Time Warping (DTW), it is calculated with the newly monitored vibration signal X ═ X1,x2,…,xZMatching loss of the filter element, the formula of the matching loss is:
Dv(Y,X)＝d(yi,xj)
wherein:
d(y1,x1)＝|y1-x1|；
d(y1,x2)＝|y1-x2|；
d(y2,x1)＝|y2-x1|；
wherein yi,xjL is yiAnd xjThe difference of (a).
For v ═ a, …, B, each candidate segment B ═ Y1,…,ylvRepeatedly calculating the matching loss to obtain { D }vA, …, b, finding the smallest DvThe corresponding l is the optimal period T.
Step (2.3): the time period of the candidate segment B corresponding to the minimum matching loss is the optimal period;
step (2.4): the vibration signal is periodically divided according to the optimum period.
The method and the device utilize the optimal period to carry out period division on the vibration signal, and can accurately and quickly carry out feature extraction on the vibration signal.
It should be noted that, the preset period may also be used to perform periodic division on the vibration signal monitored in real time.
And (3) calculating a correlation diagram, which specifically comprises the following steps:
step (3.1): forming a group of vibration data by two vibration data with a preset time interval in a period;
step (3.2): randomly selecting two groups of code words in a codebook;
step (3.3): calculating a mapping value from one vibration data in each group of vibration data to a selected group of coding words and a mapping value from the other vibration data to another selected group of coding words by using a Gaussian kernel function, multiplying the two mapping values, and finally accumulating the product results of the mapping values of all the groups of vibration data to obtain a correlation diagram of the current period of the vibration signal;
step (3.4): repeating the steps (3.2) to (3.3) on the vibration signal of each period obtained by continuous monitoring to obtain a phase diagram sequence with the same number as the period number;
for example: for a well-divided number x of time intervals Δ t in one periodtAnd xt+ΔtRespectively calculating the code words (mu) to the ith group by using Gaussian kernel functioni,σi 2) And j (th) group of code words (mu)j,σj 2) To obtain Wi,tAnd Wj,t+ΔtThe gaussian kernel function calculation formula is as follows:
and calculate Wi,tAnd Wj,t+ΔtThe product of (a);
for each pair of times within the nth period of length T, x is a number Δ TtAnd xt+ΔtAnd T belongs to (1,2, …, T-delta T), and performing Gaussian kernel function calculation to obtain W of T-delta Ti,tAnd Wj,t+ΔtAre added up to obtain a correlation diagram Cn(i,j；Δt)，Cn(i, j; Δ t) is calculated as follows:
assuming that there are k sets of code words, i is 1,2, …, k, j is 1,2, …, k, then CnIs a k × k matrix;
correlating k × knEach element in (1) is connected end to end in a row and written as a single element with k2Column vector C 'of rows'n；
Assuming that N periods are provided, repeating the above steps for each period to obtain N correlation graphs, arranging the N correlation graphs in columns according to the period sequence to obtain k2Correlation graph representation of row N column, { C'1,C'2,…,C'n,…C'N}。
Calculating an entropy value: using entropy value to carry out dimensionality reduction treatment on the obtained correlation diagram sequence to obtain the characteristic of reflecting the mechanical state operation trend, wherein the specific process comprises the following steps:
step (4.1): for calculated k2N column C 'in the correlation graph of row N columns'nCalculating its entropy value, the formula is as follows
Wherein
Wherein N is the number of the relevant graphs, namely the number of the cycles; n is 1,2, …, N represents a column of the correlation diagram, i.e. a period, m is 1,2, …, k2A row representing a correlation graph; q is a constant, pmnCorrelation map value c representing the m-th number in the n-th cyclemnIn the nth cycleThe occupied specific gravity;
step (4.2): for each cycle, repeat step (4.1) to get { e1,…,en,…eNAnd f, obtaining the extracted features.
An example of the vibration signal characteristic extraction of the change of the rotating speed of the engine gearbox is shown below, and the experimental result is shown in fig. 3. Examples are briefly described below:
the original signal is a gearbox vibration signal, the initial rotating speed of an engine is 300rpm, then the rotating speed is changed into 400rpm, the vibration signal collected by the gearbox in real time is shown in figure 2, the collected vibration signal is subjected to feature extraction according to the steps, and the extracted features are shown in figure 3.
In the invention, the correlation diagram is applied to the characteristic extraction process, and in the process of characteristic extraction depending on the amplitude of the vibration signal, the time sequence information of the original vibration signal is also considered, so that the extracted characteristic contains more effective information.
The invention also provides a vibration signal characteristic extraction device for the rotary machine, which takes the time sequence information of the original vibration signal into consideration and enables the extracted characteristics to contain more effective information.
The invention relates to a vibration signal feature extraction device for rotary machinery, which comprises a vibration signal feature extraction processor. As shown in fig. 4, the vibration signal feature extraction processor includes:
(1) a build codebook module configured to: and clustering the training samples by using the acquired vibration signals as training samples, and forming a group of code words by using the clustering mass center and the clustering variance of each type so as to construct a codebook.
Specifically, in the codebook building module, training samples are clustered by using a K-means algorithm.
The K-means algorithm is an algorithm that inputs the number K of clusters, and a database containing n data objects, and outputs K clusters satisfying the minimum variance criterion. The K-means algorithm accepts an input K; the n data objects are then divided into k clusters so that the obtained clusters satisfy: the similarity of objects in the same cluster is higher; while the object similarity in different clusters is smaller.
It should be noted that, in addition to clustering the training samples by using the K-means algorithm, the K-medoid algorithm or the CLARANS algorithm may also be used to cluster the training samples.
(2) A split period module configured to: and periodically dividing the vibration signal monitored in real time.
In particular, in the split period module, the real-time monitored vibration signal is periodically divided using time-series analysis, the split period module being further configured to:
determining a set of candidate vibration signal segments a based on the newly monitored vibration signal;
randomly selecting a candidate segment B with the end point being the same as the starting point of the selected vibration signal segment, and calculating the matching loss between the candidate segment B and the newly monitored vibration signal by using a dynamic time warping method;
the time period of the candidate segment B corresponding to the minimum matching loss is the optimal period;
the vibration signal is periodically divided according to the optimum period.
The method and the device utilize the optimal period to carry out period division on the vibration signal, and can accurately and quickly carry out feature extraction on the vibration signal.
It should be noted that, the preset period may also be used to perform periodic division on the vibration signal monitored in real time.
(3) A compute dependency graph module configured to:
forming a group of vibration data by two vibration data with a preset time interval in a period;
randomly selecting two groups of code words in a codebook;
calculating a mapping value from one vibration data in each group of vibration data to a selected group of coding words and a mapping value from the other vibration data to another selected group of coding words by using a Gaussian kernel function, multiplying the two mapping values, and finally accumulating the product results of the mapping values of all the groups of vibration data to obtain a correlation diagram of the current period of the vibration signal;
and obtaining a correlation diagram sequence with the same number as the vibration signal period number.
(4) A compute entropy module configured to: and performing dimensionality reduction treatment on the obtained correlation diagram sequence by using the entropy value to obtain the characteristic reflecting the mechanical state operation trend.
In the invention, the correlation diagram is applied to the characteristic extraction process, and in the process of characteristic extraction depending on the amplitude of the vibration signal, the time sequence information of the original vibration signal is also considered, so that the extracted characteristic contains more effective information.
Fig. 5 is a flowchart of an operation state monitoring method of a rotary machine of the present invention.
As shown in fig. 5, an operation state monitoring method of a rotary machine of the present invention includes:
(1) and (3) extracting characteristics: the vibration signal feature extraction method facing the rotating machine as shown in fig. 1 is adopted to obtain features reflecting the running trend of the machine state.
(2) Calculating the degree of abnormality: and measuring the abnormality degree of the reaction mechanical state by using the Euclidean distance based on the acquired characteristics reflecting the mechanical state operation trend.
Features e extracted based on the method shown in FIG. 11,…,en,…eNE, calculating en-1And enThe absolute value of the difference therebetween being taken as the degree of abnormality, i.e.
sn＝|en-en-1|
Are sequentially paired with { e1,…,en,…eNCalculating two adjacent numbers in the image as above to obtain the abnormal degree s2,…,sn,…sN}。
(3) A decision analysis step: and (4) adopting hypothesis testing to make a decision on the obtained abnormality degree.
In the decision analysis step, it is checked whether a change in the operating state of the rotary machine has occurred using the 3 σ principle in the hypothesis test.
For the obtained anomaly value s2,…,sn,…sNUsing the 3 sigma principle in hypothesis test to test whether changes occur, the hypothesis test is as follows
H0：|sn-μ'j|＜3σ'jNo change occurs;
H1：|sn-μ'j|≥3σ'j'the change occurs.
Wherein, mu'jAnd σ'jAre respectively data { sjMean and standard deviation of 2,3, …, n-1; h0Indicating no anomaly in the nth cycle, H1Indicating that a change has occurred during the nth period.
The hypothesis test is a method for deducing a population from a sample according to certain hypothesis in mathematical statistics. The specific method comprises the following steps: making some assumption about the population studied as required by the problem, denoted H0; selecting a suitable statistic chosen such that its distribution is known, assuming H0 holds; from the measured samples, the value of the statistic is calculated and tested against a predetermined level of significance, making a decision to reject or accept the hypothesis H0.
Hypothesis testing is an important element in sample inference. The method is a test method which makes an assumption whether a total index is equal to a certain numerical value or not and whether a certain random variable obeys a certain probability distribution or not according to raw data, then calculates statistic related to test by using a certain statistical method according to sample data, judges whether an estimated value and the total numerical value (or the estimated distribution and the actual distribution) have significant difference or not according to a certain probability principle and judges whether the raw data should be selected.
The following shows an example of extracting the vibration signal characteristic of the change of the rotating speed of the engine gearbox:
the original signal is a gearbox vibration signal, the initial rotating speed of an engine is 300rpm, then the rotating speed is changed into 400rpm, the vibration signal collected by the gearbox in real time is shown in figure 2, the collected vibration signal is subjected to feature extraction and statistical analysis decision according to the steps, the extracted features are shown in figure 3, figure 6 is an abnormal value of the extracted features, and the position of the rotating speed change point detected by the method is marked in figure 6.
As shown in fig. 5, the method includes two parts, feature extraction and statistical analysis decision. Wherein, the feature extraction is divided into a learning stage and a testing stage. The learning stage is a process of clustering the collected rotating mechanical vibration signals by using (K-means) to obtain a codebook; in the testing stage, the newly acquired vibration signals are periodically divided by utilizing dynamic time warping; and calculating a correlation diagram by using the vibration signals with the divided Gaussian kernel periods, and performing dimensionality reduction on the obtained correlation diagram by using entropy values to obtain a characteristic capable of reflecting the trend of the mechanical state. Finally, the obtained characteristics are subjected to statistical analysis and decision making, and the running state of the rotary machine is monitored.
In the invention, a vibration signal of the rotating machine is analyzed by using a time domain analysis method, the acquired characteristic reflecting the running trend of the machine state is measured by using the Euclidean distance, and finally, the obtained abnormal degree is decided by adopting hypothesis test, so that the running state of the rotating machine can be rapidly and accurately monitored.
The invention also provides a device for monitoring the running state of the rotary machine.
The invention relates to a device for monitoring the running state of a rotating machine, which comprises:
a vibration signal feature extraction device facing the rotary machine;
vibration signal feature extraction element still links to each other with the state monitoring treater, the state monitoring treater includes:
a calculate outliers module configured to: based on the acquired characteristics of the reaction mechanical state operation trend, carrying out anomaly measurement on the reaction mechanical state operation trend by using the Euclidean distance;
a decision analysis module configured to: and (4) adopting hypothesis testing to make a decision on the obtained abnormality degree.
Specifically, in the decision analysis module, the 3 σ principle in hypothesis testing is employed to check whether a change in the operating state of the rotary machine has occurred.
The hypothesis test is a method for deducing a population from a sample according to certain hypothesis in mathematical statistics. The specific method comprises the following steps: making some assumption about the population studied as required by the problem, denoted H0; selecting a suitable statistic chosen such that its distribution is known, assuming H0 holds; from the measured samples, the value of the statistic is calculated and tested against a predetermined level of significance, making a decision to reject or accept the hypothesis H0.
Hypothesis testing is an important element in sample inference. The method is a test method which makes an assumption whether a total index is equal to a certain numerical value or not and whether a certain random variable obeys a certain probability distribution or not according to raw data, then calculates statistic related to test by using a certain statistical method according to sample data, judges whether an estimated value and the total numerical value (or the estimated distribution and the actual distribution) have significant difference or not according to a certain probability principle and judges whether the raw data should be selected.
In the invention, a vibration signal of the rotating machine is analyzed by using a time domain analysis method, the acquired characteristic reflecting the running trend of the machine state is measured by using the Euclidean distance, and finally, the obtained abnormal degree is decided by adopting hypothesis test, so that the running state of the rotating machine can be rapidly and accurately monitored.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (10)
1. A vibration signal feature extraction method for a rotary machine, comprising:
step (1), codebook construction: using the collected vibration signals as training samples, clustering the training samples, and forming a group of code words by the clustering mass center and the clustering variance of each type so as to construct a codebook;
step (2) dividing the cycle: carrying out periodic division on the vibration signal monitored in real time;
and (3) calculating a correlation diagram, which specifically comprises the following steps:
step (3.1): forming a group of vibration data by two vibration data with a preset time interval in a period;
step (3.2): randomly selecting two groups of code words in a codebook;
step (3.3): calculating a mapping value from one vibration data in each group of vibration data to a selected group of coding words and a mapping value from the other vibration data to another selected group of coding words by using a Gaussian kernel function, multiplying the two mapping values, and finally accumulating the product results of the mapping values of all the groups of vibration data to obtain a correlation diagram of the current period of the vibration signal;
step (3.4): repeating the steps (3.2) to (3.3) on the vibration signal of each period obtained by continuous monitoring to obtain a correlation diagram sequence with the same number as the period number;
calculating an entropy value: and performing dimensionality reduction treatment on the obtained correlation diagram sequence by using the entropy value to obtain the characteristic reflecting the mechanical state operation trend.
2. The method for extracting characteristics of vibration signals of rotating machinery as claimed in claim 1, wherein in the step (1) of constructing the codebook, training samples are clustered by using a K-means algorithm.
3. The method for extracting the vibration signal characteristic of the rotary machine according to claim 1, wherein in the step (2) of dividing the period, the vibration signal monitored in real time is periodically divided by using time series analysis, and the specific process includes:
step (2.1): determining a set of candidate vibration signal segments a based on the newly monitored vibration signal;
step (2.2): randomly selecting a candidate segment B with the end point being the same as the starting point of the selected vibration signal segment, and calculating the matching loss between the candidate segment B and the newly monitored vibration signal by using a dynamic time warping method;
step (2.3): the time period of the candidate segment B corresponding to the minimum matching loss is the optimal period;
step (2.4): the vibration signal is periodically divided according to the optimum period.
4. A vibration signal feature extraction device for a rotary machine, comprising a vibration signal feature extraction processor, the vibration signal feature extraction processor comprising:
a build codebook module configured to: using the collected vibration signals as training samples, clustering the training samples, and forming a group of code words by the clustering mass center and the clustering variance of each type so as to construct a codebook;
a split period module configured to: carrying out periodic division on the vibration signal monitored in real time;
a compute dependency graph module configured to:
forming a group of vibration data by two vibration data with a preset time interval in a period;
randomly selecting two groups of code words in a codebook;
calculating a mapping value from one vibration data in each group of vibration data to a selected group of coding words and a mapping value from the other vibration data to another selected group of coding words by using a Gaussian kernel function, multiplying the two mapping values, and finally accumulating the product results of the mapping values of all the groups of vibration data to obtain a correlation diagram of the current period of the vibration signal;
obtaining a correlation diagram sequence with the number identical to the number of the vibration signal periods;
a compute entropy module configured to: and performing dimensionality reduction treatment on the obtained correlation diagram sequence by using the entropy value to obtain the characteristic reflecting the mechanical state operation trend.
5. A rotary machine oriented vibration signal feature extraction device as claimed in claim 4, characterized in that in said codebook building module training samples are clustered using K-means algorithm.
6. A rotary machine oriented vibration signal feature extraction apparatus as claimed in claim 4, wherein in said split period module, the real-time monitored vibration signal is periodically divided using time series analysis, said split period module is further configured to:
determining a set of candidate vibration signal segments a based on the newly monitored vibration signal;
randomly selecting a candidate segment B with the end point being the same as the starting point of the selected vibration signal segment, and calculating the matching loss between the candidate segment B and the newly monitored vibration signal by using a dynamic time warping method;
the time period of the candidate segment B corresponding to the minimum matching loss is the optimal period;
the vibration signal is periodically divided according to the optimum period.
7. A method of monitoring an operating condition of a rotary machine, comprising:
and (3) extracting characteristics: the method for extracting the vibration signal characteristic facing the rotary machine is adopted to obtain the characteristic reflecting the running trend of the machine state according to any one of claims 1 to 3;
calculating the degree of abnormality: based on the acquired characteristics of the reaction mechanical state operation trend, carrying out anomaly measurement on the reaction mechanical state operation trend by using the Euclidean distance;
a decision analysis step: and (4) adopting hypothesis testing to make a decision on the obtained abnormality degree.
8. An operating condition monitoring method of a rotary machine according to claim 7, wherein in said decision analysis step, it is checked whether or not a change in the operating condition of the rotary machine has occurred using a 3 σ rule in hypothesis testing.
9. An operation state monitoring device of a rotary machine, comprising:
a rotary machine-oriented vibration signal feature extraction apparatus according to any one of claims 4 to 6;
vibration signal feature extraction element still links to each other with the state monitoring treater, the state monitoring treater includes:
a calculate outliers module configured to: based on the acquired characteristics of the reaction mechanical state operation trend, carrying out anomaly measurement on the reaction mechanical state operation trend by using the Euclidean distance;
a decision analysis module configured to: and (4) adopting hypothesis testing to make a decision on the obtained abnormality degree.
10. An operation state monitoring device of a rotary machine according to claim 9, wherein in said decision analysis module, it is checked whether or not a change in the operation state of the rotary machine has occurred using a 3 σ rule in hypothesis test.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810586819.9A CN108955855B (en) | 2018-06-08 | 2018-06-08 | Vibration signal feature extraction method, monitoring method and device for rotary machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810586819.9A CN108955855B (en) | 2018-06-08 | 2018-06-08 | Vibration signal feature extraction method, monitoring method and device for rotary machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108955855A CN108955855A (en) | 2018-12-07 |
| CN108955855B true CN108955855B (en) | 2020-05-08 |
Family
ID=64493498
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810586819.9A Active CN108955855B (en) | 2018-06-08 | 2018-06-08 | Vibration signal feature extraction method, monitoring method and device for rotary machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108955855B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109655267B (en) * | 2019-01-07 | 2020-12-18 | 江苏大学 | Method for extracting fault features of automobile hub bearing |
| CN111929044B (en) * | 2020-07-15 | 2023-08-08 | 西门子工厂自动化工程有限公司 | Method, apparatus, computing device and storage medium for monitoring device status |
| CN111985380B (en) * | 2020-08-13 | 2022-07-05 | 山东大学 | Bearing degradation process state monitoring method, system, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2102776C1 (en) * | 1995-06-19 | 1998-01-20 | Юрий Петрович Кострыгин | Vibroseismic survey method |
| CN105675320A (en) * | 2016-01-06 | 2016-06-15 | 山东大学 | Method for real time monitoring mechanical system operation status on the basis of acoustic signal analysis |
| CN106599271A (en) * | 2016-12-22 | 2017-04-26 | 江苏方天电力技术有限公司 | Emission monitoring time series data abnormal value detection method for coal-fired unit |
| CN107610464A (en) * | 2017-08-11 | 2018-01-19 | 河海大学 | A kind of trajectory predictions method based on Gaussian Mixture time series models |
-
2018
- 2018-06-08 CN CN201810586819.9A patent/CN108955855B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2102776C1 (en) * | 1995-06-19 | 1998-01-20 | Юрий Петрович Кострыгин | Vibroseismic survey method |
| CN105675320A (en) * | 2016-01-06 | 2016-06-15 | 山东大学 | Method for real time monitoring mechanical system operation status on the basis of acoustic signal analysis |
| CN106599271A (en) * | 2016-12-22 | 2017-04-26 | 江苏方天电力技术有限公司 | Emission monitoring time series data abnormal value detection method for coal-fired unit |
| CN107610464A (en) * | 2017-08-11 | 2018-01-19 | 河海大学 | A kind of trajectory predictions method based on Gaussian Mixture time series models |
Non-Patent Citations (1)
| Title |
|---|
| Application of wavelet energy and Shannon entropy for feature;Hojat Heidari Bafroui;《Neurocomputing》;20141231;第437-445页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108955855A (en) | 2018-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ahmad et al. | A hybrid prognostics technique for rolling element bearings using adaptive predictive models | |
| CN111353482B (en) | LSTM-based fatigue factor recessive anomaly detection and fault diagnosis method | |
| CN108955855B (en) | Vibration signal feature extraction method, monitoring method and device for rotary machine | |
| Rohani Bastami et al. | Estimation of remaining useful life of rolling element bearings using wavelet packet decomposition and artificial neural network | |
| CN113834657B (en) | Bearing fault early warning and diagnosis method based on improved MSET and frequency spectrum characteristics | |
| Zhang et al. | Degradation modeling and maintenance decisions based on Bayesian belief networks | |
| US20130148817A1 (en) | Abnormality detection apparatus for periodic driving system, processing apparatus including periodic driving system, abnormality detection method for periodic driving system, and computer program | |
| EP3822595B1 (en) | Predictive maintenance for robotic arms using vibration measurements | |
| US20070239629A1 (en) | Cluster Trending Method for Abnormal Events Detection | |
| US20140039806A1 (en) | Estimating remaining useful life from prognostic features discovered using genetic programming | |
| CN111122162B (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
| Carino et al. | Remaining useful life estimation of ball bearings by means of monotonic score calibration | |
| Li et al. | A particle filtering-based approach for remaining useful life predication of rolling element bearings | |
| JP2012128840A (en) | Abnormality monitoring system and abnormality monitoring method | |
| Zhang et al. | Deep transfer learning-based hierarchical adaptive remaining useful life prediction of bearings considering the correlation of multistage degradation | |
| Wong et al. | Effects of compressed sensing on classification of bearing faults with entropic features | |
| US11449781B2 (en) | Plant abnormality prediction system and method | |
| Yu et al. | Fault diagnosis of rolling element bearing using multi-scale Lempel-Ziv complexity and mahalanobis distance criterion | |
| Singleton et al. | Discovering the hidden health states in bearing vibration signals for fault prognosis | |
| Techane et al. | Rotating machinery prognostics and application of machine learning algorithms: Use of deep learning with similarity index measure for health status prediction | |
| CN111950757A (en) | Method for predicting degradation state trend of key parts of rotary machine | |
| Chen et al. | Performance degradation assessment of rotary machinery based on a multiscale Tsallis permutation entropy method | |
| JP7127477B2 (en) | LEARNING METHOD, APPARATUS AND PROGRAM, AND EQUIPMENT FAILURE DIAGNOSIS METHOD | |
| Zhang et al. | A hybrid feature selection and health indicator construction scheme for delay-time-based degradation modelling of rolling element bearings | |
| Choi et al. | Prognostics by classifying degradation stage on Lambda architecture |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |