CN112036353A - Collaborative filtering bearing current damage fault identification method based on memory - Google Patents

Collaborative filtering bearing current damage fault identification method based on memory Download PDF

Info

Publication number
CN112036353A
CN112036353A CN202010936952.XA CN202010936952A CN112036353A CN 112036353 A CN112036353 A CN 112036353A CN 202010936952 A CN202010936952 A CN 202010936952A CN 112036353 A CN112036353 A CN 112036353A
Authority
CN
China
Prior art keywords
bearing
similarity
data
state
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010936952.XA
Other languages
Chinese (zh)
Inventor
王广斌
贺英航
李学军
弓满锋
宾光富
程欢珂
王腾强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Lingnan Normal University
Original Assignee
Lingnan Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Lingnan Normal University filed Critical Lingnan Normal University
Priority to CN202010936952.XA priority Critical patent/CN112036353A/en
Publication of CN112036353A publication Critical patent/CN112036353A/en
Pending legal-status Critical Current

Links

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
    • 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
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Signal Processing (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Electrochemistry (AREA)
  • Evolutionary Computation (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Acoustics & Sound (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a collaborative filtering bearing current damage fault identification method based on a memory, which comprises the following steps: 1) constructing a joint scoring matrix for bearing state identification; 2) and calculating the bearing state prediction score of the fault data in the bearing state recognition combined scoring matrix according to the training data in the bearing state recognition combined scoring matrix and by combining multiple similarity measurement indexes. The method applies the collaborative filtering recommendation system to the fault identification of the bearing current damage, has simple steps, is easy to realize, and can effectively identify the fault of the bearing current damage.

Description

Collaborative filtering bearing current damage fault identification method based on memory
Technical Field
The invention relates to the field of bearing fault diagnosis, in particular to a collaborative filtering bearing current damage fault identification method based on a memory.
Background
The method separates essential characteristics and other interference parts existing in the fault signal, deeply excavates fault characteristic values, and obtains effective characteristic information from the fault signal as much as possible, so that the method is a core target of fault diagnosis. Because the current damage process of the bearing is a slow and complex process, the signal of the current damage fault is weak and very stable, in order to obtain comprehensive and complete monitoring data, more sensors are often selected for monitoring the current damage fault, so that the density of the sensor arrangement is increased sharply, the monitoring time is also prolonged, and the amount of the faced monitoring data shows explosive increase. In addition, complex additional interference and noise exist in the monitoring process, and the extracted monitoring data are more and more complex. It has become increasingly difficult to extract fault features from fault signals using classical time-frequency domain analysis with limitations, empirical mode decomposition and wavelet decomposition.
With the rapid development of the internet technology, the collaborative filtering recommendation system is also continuously improved and matured as an effective method for solving the problem of information overload, and is widely applied. The collaborative filtering follows the basic assumption that the information demands of users with the same or similar interest preferences are similar when the problem of internet information overload is solved, and the ' crowd ' intelligence ' is fully utilized to filter and screen information. The collaborative filtering recommendation algorithm has the advantages of strong information processing and screening capability, high processing efficiency and the like, provides a good idea for processing massive bearing current damage monitoring data, and opens up a new research direction for fault diagnosis of mechanical equipment.
Collaborative filtering can be generally divided into two categories, Memory-Based (Memory-Based) and Model-Based (Model-Based). And selecting a part of neighbor users with similar interests for the target user based on the collaborative filtering of the memory, and predicting the rating value of the target user to the project according to the rating of the neighbor users. And (3) learning to obtain a complex model according to the training set data based on the model collaborative filtering, and then deducing the scoring value of the target user on the unscored items based on the model and the scored data of the target user. However, the collaborative filtering recommendation system is established on the corresponding scoring matrix, and for the identification of the rolling bearing state, no specific scoring rule exists, and the corresponding scoring matrix cannot be established.
Disclosure of Invention
In order to solve the technical problems, the invention provides a collaborative filtering bearing current damage fault identification method based on a memory, which is simple in algorithm and high in diagnosis precision.
The technical scheme for solving the problems is as follows: a collaborative filtering bearing current damage fault identification method based on a memory is characterized by comprising the following steps:
1) constructing a joint scoring matrix for bearing state identification;
2) and calculating the bearing state prediction score of the fault data in the bearing state recognition combined scoring matrix according to the training data in the bearing state recognition combined scoring matrix and by combining multiple similarity measurement indexes.
In the above method for identifying the current damage fault of the collaborative filtering bearing based on the memory, in the step 1), it is assumed that there are vibration signal data S of u groups of rolling bearings(1)，…，S(h)，S(h+1)，…，S(u)And these rolling bearings exist in v different types of states z(1)，z(2)，…，z(v)The previous h sets of training data S are known(1)，…，S(h)If the existing state exists, the concrete steps in the step 1) are as follows:
(1-1) the ith group of signal data S(i)I ═ 1, …, h, h +1, …, u; decomposing into a layer by the existing wavelet packet technology to obtain b-2a1 sub-band, then the total signal S of the ith set of signal data(i)Expressed as follows:
Figure BDA0002672293220000031
wherein
Figure BDA0002672293220000032
Means that the i-th group of signal data is decomposed into the signal of the b-th sub-band after the a-layer
Figure BDA0002672293220000033
Corresponding energy is
Figure BDA0002672293220000034
w is 0, 1, …, b; the ith group of signal data is decomposed into energy of w sub-band after a layer
Figure BDA0002672293220000035
Comprises the following steps:
Figure BDA0002672293220000036
total energy E of ith group of signal data(i)Comprises the following steps:
Figure BDA0002672293220000037
construction of S by energy(i)Normalized feature vector T of(i)The following were used:
Figure BDA0002672293220000038
wherein
Figure BDA0002672293220000039
Representing the result of dividing the w energy of the sub-band after the i-th group of signal data is decomposed into a layer a by the total energy of the i-th group of signal data,
Figure BDA00026722932200000310
according to the feature vector T(i)Obtaining a bearing feature scoring matrix
Figure BDA00026722932200000311
As shown in the following formula:
Figure BDA00026722932200000312
wherein R represents a real number set;
(1-2) obtaining a bearing state scoring matrix according to the corresponding state of the bearing
Figure BDA00026722932200000313
As shown in the following formula:
Figure BDA00026722932200000314
for training data S(1)，S(2)，…，S(h)Their corresponding existing known status scores are given a maximum value of 1, while the non-existing status scores are given a small value of ≦ 1/10000, for test data S(i′)I' ═ h +1, …, u; its corresponding state Z(t)Is not known to give a zero value, and is recorded as
Figure BDA0002672293220000041
t＝1，…，v；
(1-3) combining the bearing characteristic scoring matrix A and the bearing state scoring matrix B to obtain a joint scoring matrix C for bearing state identification:
Figure BDA0002672293220000042
wherein
Figure BDA0002672293220000043
And d ═ b +1+ v.
In the method for identifying the current damage fault of the collaborative filtering bearing based on the memory, in the step 2), the training data are from the h +1 th row to the u th row of the joint scoring matrix C, and the testing data are from the 1 st row to the h th row of the joint scoring matrix C;
the following assumptions are made: if the bearing data of the fault i has a higher feature score in a certain energy section relative to the feature scores of other energy sections, and if unknown new bearing data also has a higher feature score in the certain energy section compared with other energy sections of the unknown new bearing data, the unknown new bearing data is considered to have the fault i, and according to the assumption, the bearing states of the test data with unknown states are estimated by using the known training data, and the specific steps are as follows:
(2-1) scoring by bearing characteristics, and calculating test data S by adopting cosine similarity(i′)With each training data S(h +1)，…，S(u)Similarity of (2):
simcos(S(i′)，S(h+1))，…，simcos(S(i′)，S(u))
wherein, simcos(S(i′)，S(h+1)) Represents the test data S(i′)And training data S(h+1)Cosine similarity of (d);
(2-2) taking the first N groups of training data with highest similarity
Figure BDA00026722932200000512
And recording it with S(i′)The similarity of (A) is as follows:
Figure BDA0002672293220000051
(2-3) training data with the training data
Figure BDA0002672293220000052
Bearing condition scoring of
Figure BDA0002672293220000053
On the basis of which test data S are calculated(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000054
Figure BDA0002672293220000055
(2-4) repeating the steps (2-1) - (2-3) by adopting the similarity based on the Euclidean distance to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000056
(2-5) repeating the steps (2-1) - (2-3) by adopting the Pearson correlation coefficient to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000057
(2-6) comparing each test data separately
Figure BDA0002672293220000058
And
Figure BDA0002672293220000059
prediction scores, the highest of which is test data S(i′)To state Z(t)Predictive scoring of
Figure BDA00026722932200000510
(2-7) prediction score
Figure BDA00026722932200000511
Corresponding state Z(t)I.e. the predicted state of the identified test data.
In the memory-based collaborative filtering bearing current damage fault identification method, in the step (2-1), the cosine similarity takes the score of each user as a point in an n-dimensional space, and the similarity between the cosine similarity and the point in the n-dimensional space is measured on the basis of the point; taking a vector in a three-dimensional space as an example, if two points in the space respectively represent scores of two users, respectively connecting the two points with an origin point can form two vectors; the similarity between the users is closely related to the relationship between the two vectors, the included angle between the vectors substantially determines the similarity between the two vectors, namely the two users, the smaller the included angle is, the more similar the two vectors are, the higher the similarity is, and vice versa; the cosine of the included angle is used as a numerical value for similarity comparison, the magnitude of the numerical value reflects the degree of similarity, and the value of the numerical value is between-1 and 1; the cosine similarity is calculated by the following formula:
Figure BDA0002672293220000061
wherein m represents the number of the vector, X1Represents a first vector, X2Represents a second vector; xmRepresenting a point in an n-dimensional space, i.e. an n-dimensional vector.
In the memory-based collaborative filtering bearing current damage fault identification method, in the step (2-4), the similarity based on the Euclidean distance takes the score of each user as a point in an n-dimensional space, and the similarity between the users is measured by using the Euclidean distance between vectors; the Euclidean distance between two user vector coordinates is calculated according to the following formula:
Figure BDA0002672293220000062
in order to reflect the positive correlation of the similarity, when the method is actually used, the calculated Euclidean distance is subjected to certain deformation to be used as the value of the final similarity, and the similarity calculation formula based on the Euclidean distance is as follows:
Figure BDA0002672293220000063
the more similar between two users the larger the value, the limit is 1, representing that the preferences of both users are identical.
In the above method for identifying the current damage fault of the collaborative filtering bearing based on the memory, in the step (2-5), the pearson correlation coefficient is a classical similarity estimation method in mathematical statistics, and the calculation method thereof is as follows:
Figure BDA0002672293220000064
from the above formula, the value of the similarity value is between-1 and 1, the higher the absolute value of the value is, the higher the linear correlation is, the limits 1 and-1 represent complete positive correlation or negative correlation, and the value of 0 represents no linear correlation.
The invention has the beneficial effects that:
1. firstly, constructing a joint scoring matrix for bearing state identification, namely obtaining a bearing characteristic scoring matrix according to wavelet sub-band energy, designing a scoring matrix for accurately describing a bearing state, and combining scores of the two different characteristics to obtain the joint scoring matrix for bearing state identification; and then, calculating bearing state prediction scores of fault data in the bearing state recognition combined scoring matrix according to training data in the bearing state recognition combined scoring matrix and in combination with multiple similarity measurement indexes. The method applies the collaborative filtering recommendation system to the fault identification of the bearing current damage, has simple steps, is easy to realize, and can effectively identify the fault of the bearing current damage.
2. The invention utilizes the characteristic of 'information filtering' of a collaborative filtering algorithm to give full play to 'intelligence' of data. Compared with the traditional fault diagnosis method, the method has the advantages that firstly, the expansibility of a diagnosis model is strong, new fault data can be added into a scoring matrix in time to enhance the recognition accuracy of the algorithm, and therefore, the method is suitable for diagnosing the bearing fault with relatively large data volume; secondly, the control parameters of the algorithm are few, and the algorithm is easy to adjust in time in the diagnosis process; and finally, the calculation formula of the similarity is simple, and the calculation amount of the algorithm is small.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings and examples.
As shown in fig. 1, a method for identifying a current damage fault of a collaborative filtering bearing based on a memory is characterized by comprising the following steps:
1) and constructing a joint scoring matrix for bearing state identification.
Vibration signal data S for supposing u groups of rolling bearings(1)，…，S(h)，S(h+1)，…，S(u)And these rolling bearings exist in v different types of states z(1)，z(2)，…，z(v)The previous h sets of training data S are known(1)，…，S(h)If the existing state exists, the concrete steps in the step 1) are as follows:
(1-1) the ith group of signal data S(i)I ═ 1, …, h, h +1, …, u; decomposing into a layer by the existing wavelet packet technology to obtain b-2a1 sub-band, then the total signal S of the ith set of signal data(i)Expressed as follows:
Figure BDA0002672293220000081
wherein
Figure BDA0002672293220000082
Means that the i-th group of signal data is decomposed into the signal of the b-th sub-band after the a-layer
Figure BDA0002672293220000083
Corresponding energy is
Figure BDA0002672293220000084
w is 0, 1, …, b; the ith group of signal data is decomposed into energy of w sub-band after a layer
Figure BDA0002672293220000085
Comprises the following steps:
Figure BDA0002672293220000086
total energy E of ith group of signal data(i)Comprises the following steps:
Figure BDA0002672293220000087
construction of S by energy(i)Normalized feature vector T of(i)The following were used:
Figure BDA0002672293220000088
wherein
Figure BDA0002672293220000089
Representing the result of dividing the w energy of the sub-band after the i-th group of signal data is decomposed into a layer a by the total energy of the i-th group of signal data,
Figure BDA00026722932200000810
according to the feature vector T(i)Designing a bearing characteristic scoring table as shown in table 1, and obtaining a bearing characteristic scoring matrix
Figure BDA00026722932200000811
R represents a real number set as shown in the following formula:
Figure BDA00026722932200000812
TABLE 1
Figure BDA0002672293220000091
(1-2) designing a bearing state rating table according to the corresponding state of the bearing, as shown in Table 2And obtaining a bearing state scoring matrix
Figure BDA0002672293220000092
As shown in the following formula:
Figure BDA0002672293220000093
TABLE 2
Figure BDA0002672293220000094
For training data S(1)，S(2)，…，S(h)Their corresponding existing known status scores are given a maximum value of 1, while the non-existing status scores are given a small value of ≦ 1/10000, for test data S(i′)I' ═ h +1, …, u; its corresponding state Z(t)Is not known to give a zero value, and is recorded as
Figure BDA0002672293220000095
t＝1，…，v：
(1-3) combining the bearing characteristic scoring matrix A and the bearing state scoring matrix B to obtain a joint scoring matrix C for bearing state identification:
Figure BDA0002672293220000101
wherein
Figure BDA0002672293220000102
And d ═ b +1+ v.
2) And calculating the bearing state prediction score of the fault data in the bearing state recognition combined scoring matrix according to the training data in the bearing state recognition combined scoring matrix and by combining multiple similarity measurement indexes. Each column is data, training data are h +1 th column to u th column of the joint scoring matrix C, and testing data are 1 st column to h th column of the joint scoring matrix C.
The important item in the TOP-N recommendation algorithm based on the memory is the similarity measurement, and firstly, three methods for measuring the similarity adopted by the invention are introduced.
a) Cosine similarity:
cosine similarity measures the similarity between each user based on their score as a point in n-dimensional space. Taking the vector in three-dimensional space as an example, if two points in the space respectively represent the scores of two users, then connecting the two points with the origin respectively forms two vectors. The similarity between users is closely related to the relationship between the two vectors, the included angle between the vectors substantially determines the similarity between the two vectors, i.e. the two users, and the smaller the included angle is, the more similar the two vectors are, the higher the similarity is, and vice versa. In practice, the cosine of the included angle is used as a numerical value for similarity comparison, the magnitude of the numerical value reflects the degree of similarity, and the value of the numerical value is between-1 and 1. The cosine similarity is calculated by the following formula:
Figure BDA0002672293220000111
wherein m represents the number of the vector, X1Represents a first vector, X2Represents a second vector; xmRepresenting a point in an n-dimensional space, i.e. an n-dimensional vector.
b) Similarity based on Euclidean distance:
similar to cosine similarity, similarity based on Euclidean distance takes the score of each user as a point in n-dimensional space, except that the Euclidean distance between vectors is used for measuring the similarity between users. The Euclidean distance calculation method between two user vector coordinates is shown as the following formula:
Figure BDA0002672293220000112
however, in order to reflect the positive correlation of the similarity, in practical use, a certain deformation is performed on the calculated euclidean distance as the final similarity value, and the similarity calculation formula based on the euclidean distance is shown as the following formula:
Figure BDA0002672293220000113
the more similar between two users the larger the value, the limit is 1, representing that the preferences of both users are identical.
c) Pearson correlation coefficient:
the pearson correlation coefficient is a classical similarity estimation method in mathematical statistics, and can fully evaluate the linear relationship between two vectors, and the calculation method is shown as the following formula:
Figure BDA0002672293220000114
in the formula, the value of the similarity value is between-1 and 1, the higher the absolute value of the value is, the higher the linear correlation is, the limits 1 and-1 represent complete positive correlation or negative correlation, and the value of 0 represents no linear correlation.
The following assumptions are made in this application: if the bearing data of the fault i has a higher feature score in a certain energy section relative to the feature scores of other energy sections, and if unknown new bearing data also has a higher feature score in the energy section compared with other energy sections of the unknown new bearing data, the unknown new bearing data is considered to have the fault i, and according to the assumption, the known training data is used for estimating the bearing state of the test data with unknown state. Based on the constructed joint scoring matrix for bearing state identification, the aim is to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000129
The method comprises the following specific steps:
(2-1) scoring by bearing characteristics, and calculating test data S by adopting cosine similarity(i′)With each training data S(h +1)，…，S(u)Similarity of (2):
simcos(S(i′)，S(h+1))，…，simcos(S(i′)，S(u))
wherein, simcos(S(i′)，S(h+1)) Represents the test data S(i′)And training data S(h+1)Cosine similarity of (d);
(2-2) taking the first N groups of training data with highest similarity
Figure BDA0002672293220000121
And recording it with S(i′)The similarity of (A) is as follows:
Figure BDA0002672293220000122
(2-3) training data with the training data
Figure BDA0002672293220000123
Bearing condition scoring of
Figure BDA0002672293220000124
On the basis of the above-mentioned raw materials,
Figure BDA0002672293220000125
is part of C (first to h columns, rows b +2 to d), the subscripts thereon all giving definitions as given above: t is 1, …, v; t is t1，…，tNRepresenting training data; calculating test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000126
Figure BDA0002672293220000127
(2-4) repeating the steps (2-1) - (2-3) by adopting the similarity based on the Euclidean distance to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000128
(2-5) repeating the steps (2-1) - (2-3) by adopting the Pearson correlation coefficient to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000131
(2-6) comparing each test data separately
Figure BDA0002672293220000132
And
Figure BDA0002672293220000133
prediction scores, the highest of which is test data S(i′)To state Z(t)Predictive scoring of
Figure BDA0002672293220000134
(2-7) prediction score
Figure BDA0002672293220000135
Corresponding state Z(t)I.e. the predicted state of the identified test data.

Claims (6)

1. A collaborative filtering bearing current damage fault identification method based on a memory is characterized by comprising the following steps:
1) constructing a joint scoring matrix for bearing state identification;
2) and calculating the bearing state prediction score of the fault data in the bearing state recognition combined scoring matrix according to the training data in the bearing state recognition combined scoring matrix and by combining multiple similarity measurement indexes.
2. The memory-based collaborative filtering bearing current damage fault identification method according to claim 1, wherein in the step 1), it is assumed thatVibration signal data S of u groups of rolling bearings(1)，…，S(h)，S(h+1)，…，S(u)And these rolling bearings exist in v different types of states z(1)，z(2)，…，z(v)The previous h sets of training data S are known(1)，…，S(h)If the existing state exists, the concrete steps in the step 1) are as follows:
(1-1) the ith group of signal data S(i)I ═ 1, …, h, h +1, …, u; decomposing into a layer by the existing wavelet packet technology to obtain b-2a1 sub-band, then the total signal S of the ith set of signal data(i)Expressed as follows:
Figure FDA0002672293210000011
wherein
Figure FDA0002672293210000012
Means that the i-th group of signal data is decomposed into the signal of the b-th sub-band after the a-layer
Figure FDA0002672293210000013
Corresponding energy is
Figure FDA0002672293210000014
w is 0, 1, …, b; the ith group of signal data is decomposed into energy of w sub-band after a layer
Figure FDA0002672293210000015
Comprises the following steps:
Figure FDA0002672293210000016
total energy E of ith group of signal data(i)Comprises the following steps:
Figure FDA0002672293210000017
construction of S by energy(i)Normalized feature vector T of(i)The following were used:
Figure FDA0002672293210000018
wherein
Figure FDA0002672293210000021
Figure FDA0002672293210000022
Representing the result of dividing the w energy of the sub-band after the i-th group of signal data is decomposed into a layer a by the total energy of the i-th group of signal data,
Figure FDA0002672293210000023
according to the feature vector T(i)Obtaining a bearing feature scoring matrix
Figure FDA0002672293210000024
As shown in the following formula:
Figure FDA0002672293210000025
wherein R represents a real number set;
(1-2) obtaining a bearing state scoring matrix according to the corresponding state of the bearing
Figure FDA0002672293210000026
As shown in the following formula:
Figure FDA0002672293210000027
for training data S(1)，S(2)，…，S(h)Their corresponding existing known status scores are given a maximum value of 1, while the non-existing status scores are given a small value of ≦ 1/10000, for test data S(i′)I' ═ h +1, …, u; its corresponding state Z(t)Is not known to give a zero value, and is recorded as
Figure FDA0002672293210000028
t＝1，…，v；
(1-3) combining the bearing characteristic scoring matrix A and the bearing state scoring matrix B to obtain a joint scoring matrix C for bearing state identification:
Figure FDA0002672293210000029
wherein
Figure FDA00026722932100000210
And d ═ b +1+ v.
3. The method for identifying the current damage fault of the collaborative filtering bearing based on the memory according to claim 2, wherein in the step 2), training data are from h +1 th column to u th column of a joint scoring matrix C, and testing data are from 1 st column to h th column of the joint scoring matrix C;
the following assumptions are made: if the bearing data of the fault i has a higher feature score in a certain energy section relative to the feature scores of other energy sections, and if unknown new bearing data also has a higher feature score in the certain energy section compared with other energy sections of the unknown new bearing data, the unknown new bearing data is considered to have the fault i, and according to the assumption, the bearing states of the test data with unknown states are estimated by using the known training data, and the specific steps are as follows:
(2-1) scoring by bearing characteristics, and calculating test data S by adopting cosine similarity(i′)With each training data S(h+1)，…，S(u)Similarity of (2):
simcos(S(i′)，S(h+1))，…，simcos(S(i′)，S(u))
wherein, simcos(S(i′)，S(h+1)) Represents the test data S(i′)And training data S(h+1)Cosine similarity of (d);
(2-2) taking the first N groups of training data with highest similarity
Figure FDA0002672293210000031
And recording it with S(i′)The similarity of (A) is as follows:
Figure FDA0002672293210000032
(2-3) training data with the training data
Figure FDA0002672293210000033
Bearing condition scoring of
Figure FDA0002672293210000034
On the basis of which test data S are calculated(i′)To state Z(t)Predictive scoring of
Figure FDA0002672293210000035
Figure FDA0002672293210000036
(2-4) repeating the steps (2-1) - (2-3) by adopting the similarity based on the Euclidean distance to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure FDA0002672293210000037
(2-5) repeating the steps (2-1) - (2-3) by adopting the Pearson correlation coefficient to obtain test data S(i′)To state Z(t)Predictive scoring of
Figure FDA0002672293210000038
(2-6) comparing each test data separately
Figure FDA0002672293210000041
And
Figure FDA0002672293210000042
prediction scores, the highest of which is test data S(i′)To state Z(t)Predictive scoring of
Figure FDA0002672293210000043
(2-7) prediction score
Figure FDA0002672293210000044
Corresponding state Z(t)I.e. the predicted state of the identified test data.
4. The memory-based collaborative filtering bearing current damage fault identification method according to claim 3, wherein in the step (2-1), cosine similarity is measured by taking a score of each user as a point in an n-dimensional space, and similarity between the cosine similarity and the point in the n-dimensional space is measured on the basis of the point; taking a vector in a three-dimensional space as an example, if two points in the space respectively represent scores of two users, respectively connecting the two points with an origin point can form two vectors; the similarity between the users is closely related to the relationship between the two vectors, the included angle between the vectors substantially determines the similarity between the two vectors, namely the two users, the smaller the included angle is, the more similar the two vectors are, the higher the similarity is, and vice versa; the cosine of the included angle is used as a numerical value for similarity comparison, the magnitude of the numerical value reflects the degree of similarity, and the value of the numerical value is between-1 and 1; the cosine similarity is calculated by the following formula:
Figure FDA0002672293210000045
wherein m represents the number of the vector, X1Represents a first vector, X2Represents a second vector; xmRepresenting a point in an n-dimensional space, i.e. an n-dimensional vector.
5. The memory-based collaborative filtering bearing current damage fault identification method according to claim 4, wherein in the step (2-4), the similarity based on Euclidean distance takes the score of each user as a point in an n-dimensional space, and the similarity between the users is measured by using the Euclidean distance between vectors; the Euclidean distance between two user vector coordinates is calculated according to the following formula:
Figure FDA0002672293210000051
in order to reflect the positive correlation of the similarity, when the method is actually used, the calculated Euclidean distance is subjected to certain deformation to be used as the value of the final similarity, and the similarity calculation formula based on the Euclidean distance is as follows:
Figure FDA0002672293210000052
the more similar between two users the larger the value, the limit is 1, representing that the preferences of both users are identical.
6. The method for identifying current damage fault of cooperative filtering bearing based on memory as claimed in claim 3, wherein in the step (2-5), the Pearson correlation coefficient is a classical similarity estimation method in mathematical statistics, and the calculation method is as follows:
Figure FDA0002672293210000053
wherein cov (X, Y) represents the covariance of X and Y, σ represents the standard deviation, σ represents the meanXDenotes the standard deviation, σ, of XYDenotes the standard deviation of Y, E denotes expectation;
from the above formula, the value of the similarity value is between-1 and 1, the higher the absolute value of the value is, the higher the linear correlation is, the limits 1 and-1 represent complete positive correlation or negative correlation, and the value of 0 represents no linear correlation.
CN202010936952.XA 2020-09-08 2020-09-08 Collaborative filtering bearing current damage fault identification method based on memory Pending CN112036353A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010936952.XA CN112036353A (en) 2020-09-08 2020-09-08 Collaborative filtering bearing current damage fault identification method based on memory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010936952.XA CN112036353A (en) 2020-09-08 2020-09-08 Collaborative filtering bearing current damage fault identification method based on memory

Publications (1)

Publication Number Publication Date
CN112036353A true CN112036353A (en) 2020-12-04

Family

ID=73585103

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010936952.XA Pending CN112036353A (en) 2020-09-08 2020-09-08 Collaborative filtering bearing current damage fault identification method based on memory

Country Status (1)

Country Link
CN (1) CN112036353A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113917267A (en) * 2021-10-15 2022-01-11 万帮数字能源股份有限公司 Charging pile troubleshooting method

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111460316A (en) * 2020-03-20 2020-07-28 南京邮电大学 Knowledge system-oriented personalized recommendation method and computer storage medium

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111460316A (en) * 2020-03-20 2020-07-28 南京邮电大学 Knowledge system-oriented personalized recommendation method and computer storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
GUANGBIN WANG等: "Bearing Fault Identification Method Based on Collaborative Filtering Recommendation Technology", 《SHOCK AND VIBRATION》, pages 1 - 10 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113917267A (en) * 2021-10-15 2022-01-11 万帮数字能源股份有限公司 Charging pile troubleshooting method

Similar Documents

Publication Publication Date Title
CN112364779A (en) Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion
CN111721536A (en) Rolling bearing fault diagnosis method for improving model migration strategy
CN110135459B (en) Zero sample classification method based on double-triple depth measurement learning network
CN109299380B (en) Exercise personalized recommendation method based on multi-dimensional features in online education platform
CN109684673B (en) Feature extraction and cluster analysis method for transient stability result of power system
CN110940523B (en) Unsupervised domain adaptive fault diagnosis method
CN110096810B (en) Industrial process soft measurement method based on layer-by-layer data expansion deep learning
CN111414461A (en) Intelligent question-answering method and system fusing knowledge base and user modeling
JP2006172437A (en) Method for determining position of segment boundary in data stream, method for determining segment boundary by comparing data subset with vicinal data subset, program of instruction executable by computer, and system or device for identifying boundary and non-boundary in data stream
CN111784121B (en) Action quality evaluation method based on uncertainty score distribution learning
CN111353373A (en) Correlation alignment domain adaptive fault diagnosis method
CN108898621B (en) Related filtering tracking method based on instance perception target suggestion window
CN114429152A (en) Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption
Zhao et al. A fault diagnosis method for rotating machinery based on CNN with mixed information
CN116701871A (en) Rolling bearing composite fault feature selection method based on causal feature weighting network
CN116401603A (en) Multi-mode bearing fault intelligent diagnosis method based on transfer learning
CN107688647A (en) A kind of study based on collaborative filtering reviews exam pool and recommends method
CN112036353A (en) Collaborative filtering bearing current damage fault identification method based on memory
CN115935187B (en) Nuclear sensitivity alignment network-based mechanical fault diagnosis method under variable working conditions
CN115905976B (en) Attention mechanism-based fault diagnosis method, system and equipment for Highway Bi-LSTM bearing
CN112052880A (en) Underwater sound target identification method based on weight updating support vector machine
CN111479107A (en) No-reference audio and video joint quality evaluation method based on natural audio and video statistics
CN115935817A (en) Rapid model generation method based on diffusion model
CN115470863A (en) Domain generalized electroencephalogram signal classification method based on double supervision
Kasahara et al. Weakly supervised acoustic defect detection in concrete structures using clustering-based augmentation

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