CN109447187B - Motor fault diagnosis method and system - Google Patents

Motor fault diagnosis method and system Download PDF

Info

Publication number
CN109447187B
CN109447187B CN201811590223.2A CN201811590223A CN109447187B CN 109447187 B CN109447187 B CN 109447187B CN 201811590223 A CN201811590223 A CN 201811590223A CN 109447187 B CN109447187 B CN 109447187B
Authority
CN
China
Prior art keywords
fault
data set
probability density
calculating
density function
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
Application number
CN201811590223.2A
Other languages
Chinese (zh)
Other versions
CN109447187A (en
Inventor
阳春华
魏焱烽
陈志文
彭涛
杨超
陶宏伟
桂卫华
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.)
Central South University
Original Assignee
Central South 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 Central South University filed Critical Central South University
Priority to CN201811590223.2A priority Critical patent/CN109447187B/en
Publication of CN109447187A publication Critical patent/CN109447187A/en
Application granted granted Critical
Publication of CN109447187B publication Critical patent/CN109447187B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Tests Of Circuit Breakers, Generators, And Electric Motors (AREA)

Abstract

The invention relates to the field of motor fault diagnosis, and discloses a motor fault diagnosis method and a system, which can find the abnormal condition of motor operation and carry out fault diagnosis, and are convenient to use and easy to implement; the method of the invention comprises the following steps: selecting a normal training data set, calculating a first detection statistic and calculating a detection threshold; selecting different types of fault data as a fault training data set, calculating second detection statistics, calculating a fault probability density function of the second detection statistics by adopting a kernel density estimation method, and constructing probability density function sets of all types of fault samples; selecting a test data set, calculating a third detection statistic according to the test data set, comparing the third detection statistic with a detection threshold value, and judging whether the motor fails; and if the fault occurs, calculating a fault density function of the third detection statistic by adopting a kernel density estimation method, and constructing a probability density function set of the test data set so as to diagnose the fault type.

Description

Motor fault diagnosis method and system
Technical Field
The invention relates to the field of motor fault diagnosis, in particular to a motor fault diagnosis method and system.
Background
The motor is an electric device widely used in various industries, such as a high-speed train traction transmission system, a wind power generator, a new energy automobile driving motor, a track traction motor, a ship motor and the like. In practical application, the motor is affected by frequent starting, load fluctuation, harsh working environment and other factors, so that the motor running state is abnormal and further a fault is inevitably generated. If the faults cannot be diagnosed and discovered in time, the faults can be continuously worsened, finally the whole system is out of work, and huge losses are brought to industrial production.
In actual operation of the motor, such as switching between states of starting acceleration, constant speed, braking deceleration and the like, dynamic characteristics shown under different working conditions are different, and changes of system parameters and measurement data are complex. For example, when the motor operates under the working condition of starting acceleration, the output torque is large, the corresponding starting current is large, and the voltage of the motor is increased; under the working condition of constant speed, if the motor runs at low speed, the current and the magnetomotive force of a motor winding are large; if the motor runs at a speed higher than the rated speed, the motor also needs to be subjected to weak magnetic control, and the magnetomotive force of the motor obviously changes; under the braking condition, taking electric braking as an example, when the motor works in a generator mode, the output torque is reversed, mechanical energy is converted into electric energy to be fed back to a power grid again or consumed through a braking resistor, and the current, magnetomotive force, rotating speed and the like of a stator of the motor are greatly changed. Therefore, the dynamic change of the motor operation is complex and has a plurality of parameters, and the factors causing the motor to break down are numerous.
Therefore, how to find the abnormal condition of the motor operation in time and carry out fault diagnosis becomes a problem which needs to be solved urgently.
Disclosure of Invention
The invention aims to provide a motor fault diagnosis method and system, which can find abnormal conditions of motor operation in time and carry out fault diagnosis, and are convenient to use and easy to implement.
In order to achieve the above object, the present invention provides a motor fault diagnosis method, comprising the steps of:
s1: selecting normal operation data of a motor to be detected as a normal training data set, calculating a first detection statistic according to the normal training data set, and calculating a detection threshold according to the first detection statistic;
s2: selecting different types of fault data from historical fault operation data as a fault training data set, calculating second detection statistics according to the fault training data set, calculating a fault probability density function of the second detection statistics by adopting a kernel density estimation method, and constructing probability density function sets of all types of fault samples;
s3: selecting real-time operation data of a motor to be tested as a test data set, calculating third detection statistic according to the test data set, comparing the third detection statistic with the detection threshold, if the value of the third detection statistic is larger than the detection threshold, judging that a fault occurs, and entering S4; otherwise, the test data set is selected again until the motor is detected to be in fault;
s4: and calculating a probability density function of the third detection statistic by adopting a kernel density estimation method, constructing a probability density function set of a test data set, and diagnosing the fault type according to the distance between the probability density function set of the test data set and the probability density function sets of all types of fault samples in the S2.
Preferably, the S1 specifically includes the following steps:
s11: the normal training data set X is represented as:
Figure BDA0001920072880000021
in the formula, m is the number of the motor sensors, N is the number of sampling points, x is a motor operation data sample collected according to a time sequence,
Figure BDA0001920072880000022
is a real number set;
s12: carrying out normalization pretreatment on the data set X to obtain a data set
Figure BDA0001920072880000023
Comprises the following steps:
Figure BDA0001920072880000024
computing a data set
Figure BDA0001920072880000025
The calculation formula of the mean matrix μ is:
Figure BDA0001920072880000026
in the formula (I), the compound is shown in the specification,
Figure BDA0001920072880000027
computing a data set
Figure BDA0001920072880000028
The calculation formula of the covariance matrix S is as follows:
Figure BDA0001920072880000029
calculating a first detection statistic by adopting a sliding window method, wherein the calculation formula is as follows:
Figure BDA0001920072880000031
wherein l (k) is the kth normal training sample set
Figure BDA0001920072880000032
Is detected in the first detection statistic of (a),
Figure BDA0001920072880000033
for the kth normal training sample set
Figure BDA0001920072880000034
The average value matrix of (a) is,
Figure BDA0001920072880000035
for the kth normal training sample set
Figure BDA0001920072880000036
Of the covariance matrix, S-1Is the inverse matrix of the covariance matrix of the normal training data set, | | | is a two-norm, tr () is the trace of the matrix;
after h sliding windows, h first detection statistics of the normal training data set are obtained as follows: l (1), …, l (h);
s13: randomly extracting B statistics from h first detection statistics of a normal training data set to form a set { l }1,…,lBB < h, and reordering the B sample samples in descending order: l(1)＜l(2)＜…＜l(w)And (3) making the w-th maximum statistic value w be lambda multiplied by alpha, wherein alpha is an allowable false alarm rate, and lambda is a random sampling frequency, and obtaining the following result after repeating lambda sampling:
Figure BDA0001920072880000037
calculating a detection threshold JthThe calculation formula is as follows:
Figure BDA0001920072880000038
preferably, the S2 specifically includes the following steps:
s21: selecting n types of fault data samples from historical fault operation data, wherein the f type of fault samples form a fault training data set XfComprises the following steps:
Figure BDA0001920072880000039
wherein f is 1, …, n;
for data set XfCarrying out normalization processing to obtain a fault training data set
Figure BDA00019200728800000310
Comprises the following steps:
Figure BDA00019200728800000311
wherein f is 1, …, n;
s22: computing a data set
Figure BDA00019200728800000312
Is a mean matrix of
Figure BDA00019200728800000313
The calculation formula is as follows:
Figure BDA0001920072880000041
in the formula (I), the compound is shown in the specification,
Figure BDA0001920072880000042
wherein d is a slave data set
Figure BDA0001920072880000043
The number of samples extracted in (1), q is the qth fault training data set, and i is 1, …, m, q is 1, …, t;
computing a data set
Figure BDA0001920072880000044
Covariance matrix of
Figure BDA0001920072880000045
The calculation formula is as follows:
Figure BDA0001920072880000046
q fault training sample set for calculating f fault by adopting sliding window method
Figure BDA0001920072880000047
Second detection statistic lf(q) the calculation formula is:
Figure BDA0001920072880000048
after t sliding windows, t second detection statistics of the f-th type fault training sample set are obtained as follows: lf(1),…,lf(t)；
S23: calculating a qth fault training sample set of the f-type fault by adopting a kernel density estimation method
Figure BDA0001920072880000049
Second detection statistic lf(q) fault probability density function Pf(q) the calculation formula is:
Figure BDA00019200728800000410
the probability density function of the t second detection statistics for the class f fault is Pf(1)，…，Pf(t)；
From t probability density functions P by sliding window methodf(1),…,PfG data are sequentially extracted in (t), and an r probability density function set corresponding to the f type fault is formed as follows:
Figure BDA00019200728800000411
wherein r is 1, …, rl。
Preferably, the step of calculating the second detection statistic according to the fault training data set in S3 specifically includes the following steps:
s31: selecting real-time operation data of the motor to be tested as a test data set Y:
Figure BDA0001920072880000051
carrying out normalization pretreatment on the test data set Y to obtain a data set
Figure BDA0001920072880000052
Comprises the following steps:
Figure BDA0001920072880000053
calculating a mean matrix of the test data set Y
Figure BDA0001920072880000054
The calculation formula is as follows:
Figure BDA0001920072880000055
in the formula (I), the compound is shown in the specification,
Figure BDA0001920072880000056
and i is 1, …, m, q is 1, …, t, where d is the number of samples extracted from the data set Y and q is the qth test training data set;
calculating a covariance matrix for test dataset Y
Figure BDA0001920072880000057
The calculation formula is as follows:
Figure BDA0001920072880000058
calculating the qth test sample set Y by adopting a sliding window methodqdThird detection statistic ly(q) the calculation formula is:
Figure BDA0001920072880000059
after t sliding windows, t third detection statistics of the test sample set are obtained as follows: ly(1),…,ly(t)。
Preferably, the normal data includes: the operating voltage, current, power, and speed of the motor.
Preferably, the S4 specifically includes the following steps:
S41：calculating the qth test sample set Y by using a nuclear density estimation methodqdThird detection statistic lyProbability density function P of (q)y(q) the calculation formula is:
Figure BDA0001920072880000061
wherein q is 1, …, t;
t third detection statistics l corresponding to the set of test samplesy(q) a probability density function of Py(1)，…，Py(t)；
From t probability density functions P by sliding window methody(1)，…，PyG data are sequentially extracted in (t), and the probability density function set of the r test sample is formed as follows:
Figure BDA0001920072880000062
wherein r is 1, …, rl；
Calculating the distance between the probability density function set of the r test sample and the probability density function set of the f type fault sample, wherein the calculation formula is as follows:
Figure BDA0001920072880000063
the distance between the probability density function set of the r test sample and the probability density function set of the n types of fault samples is as follows:
Figure BDA0001920072880000064
r under f-type faultlThe distance values are:
Figure BDA0001920072880000065
form it into a distance set
Figure BDA0001920072880000066
Then r islProbability density function set and n classes of individual test samplesThe set of distances between the fault sample probability density function sets are respectively as follows:
Figure BDA0001920072880000067
set D of the above distancesfThe distance data in (1) is rearranged from small to large
Figure BDA0001920072880000068
Calculating the lower quartile of each distance data by adopting a box type graph area calculation method
Figure BDA0001920072880000069
Median number
Figure BDA00019200728800000610
Upper quartile
Figure BDA00019200728800000611
Upper limit of UmaxAnd lower limit UminComprises the following steps:
Figure BDA0001920072880000071
in the formula (I), the compound is shown in the specification,
Figure BDA0001920072880000072
it is indicated that the direction is positive in the upward direction,
Figure BDA0001920072880000073
to get
Figure BDA0001920072880000074
Centralizing data at corresponding positions;
all D are plotted against the calculated values of equation (20)f1, …, n; respectively calculate all corresponding DfArea S of box plotf：
Sf＝IQRf×W；(23)
In the formula, W is the width of the bottom edge;
s42: judging all DfAnd judging the fault type corresponding to the box type graph as the fault type of the motor to be tested.
Preferably, the first detection statistic, the second detection statistic, and the third detection statistic are all of KL detection statistics.
Preferably, in S41, the distance between the r-th test sample probability density function set and the n-type fault sample probability density function set is KL divergence between the r-th test sample probability density function set and the n-type fault sample probability density function set.
As a general inventive concept, the present invention also provides a motor fault diagnosis system, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the above method when executing the computer program.
The invention has the following beneficial effects:
the invention provides a motor fault diagnosis method and system, which adopt a kernel density estimation method to calculate a fault probability density function, diagnose the fault type by comparing the probability density function of real-time operation data with the fault probability density functions of all types of faults, can find the abnormal condition of motor operation in time and carry out fault diagnosis, and have wide applicability, high accuracy, convenient use and easy implementation; the maintenance work can be conveniently and timely arranged, and the motor safety maintenance device has important significance in improving the safe operation of the motor.
The present invention will be described in further detail below with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a flow chart of a method for diagnosing electrode faults in accordance with a preferred embodiment of the present invention;
FIG. 2 is a waveform diagram of a fault detection result of a fault in a broken bar according to a preferred embodiment of the present invention;
FIG. 3 is a fault diagnosis result box diagram of the preferred embodiment of the present invention;
FIG. 4 is a waveform illustrating the detection of an air gap bearing fault in accordance with a preferred embodiment of the present invention;
FIG. 5 is a boxed view of the air gap bearing fault diagnosis result of the preferred embodiment of the present invention;
FIG. 6 is a waveform diagram of inter-turn short fault detection in accordance with a preferred embodiment of the present invention;
FIG. 7 is a boxed view of the inter-turn short fault diagnosis result of the preferred embodiment of the present invention;
FIG. 8 is a waveform illustrating the detection of an air gap eccentricity fault in accordance with a preferred embodiment of the present invention;
fig. 9 is a box-type view of the air gap eccentricity fault diagnosis result of the preferred embodiment of the present invention.
Detailed Description
The embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways as defined and covered by the claims.
Unless otherwise defined, all terms of art used hereinafter have the same meaning as commonly understood by one of ordinary skill in the art. The use of "first," "second," and similar terms in the description and in the claims of the present application do not denote any order, quantity, or importance, but rather the intention is to distinguish one element from another. Also, the use of the terms "a" or "an" and the like do not denote a limitation of quantity, but rather denote the presence of at least one.
Example 1
Referring to fig. 1, the present embodiment provides a motor fault diagnosis method, including the following steps:
s1: selecting normal operation data of a motor to be detected as a normal training data set, calculating a first detection statistic according to the normal training data set, and calculating a detection threshold according to the first detection statistic;
s2: selecting different types of fault data from historical fault operation data as a fault training data set, calculating second detection statistics according to the fault training data set, calculating a fault probability density function of the second detection statistics by adopting a kernel density estimation method, and constructing probability density function sets of all types of fault samples;
s3: selecting real-time operation data of the motor to be tested as a test data set, calculating third detection statistic according to the test data set, comparing the third detection statistic with a detection threshold, if the value of the third detection statistic is larger than the detection threshold, judging that a fault occurs, and entering S4; otherwise, the test data set is selected again until the motor is detected to be in fault;
s4: and calculating a fault density function of the third detection statistic by adopting a kernel density estimation method, constructing a probability density function set of the test data set, and diagnosing the fault type according to the distance between the probability density function set of the test data set and the probability density function sets of all types of fault samples in S2.
The motor fault diagnosis method can find the abnormal condition of the motor operation in time and carry out fault diagnosis, and has the advantages of wide applicability, high accuracy, convenient use and easy implementation; the maintenance work can be conveniently and timely arranged, and the motor safety maintenance device has important significance in improving the safe operation of the motor.
Specifically, the motor failure in the rail transit traction drive control system is taken as an example for explanation.
In the embodiment, the three-phase current of the motor is derived from the phase a, the phase b and the phase c of the rotor; the sampling time is 60000ms, which corresponds to 60000 samples; selecting a sample from the normal operation data to form a normal training data set, and calculating KL detection statistic (Kullback-Leibler divergence) of the normal operation data, wherein the normal operation data comprises parameters such as operation voltage, current, power, rotating speed and the like of the motor. In the embodiment, the KL detection statistic of the normal training data set is the first detection statistic, and a detection threshold value is obtained by a self-service method; selecting different types of fault samples from fault operation data to form a fault training data set, calculating KL detection statistics of the fault training data set, and calculating probability density functions of different types of faults by adopting a kernel density estimation method; selecting a sample from the field operation data to form a test data set, calculating KL detection statistics of the test data set, and comparing the KL detection statistics with a detection threshold to judge whether the motor fails; if the fault occurs, calculating KL divergence (distance) between the probability density function set of the test sample and the probability density function sets of all types of fault samples and box type graph areas of all types of faults; and judging the type of the current fault with the smallest area.
Preferably, the implementation steps of the invention are elaborated in combination with a motor air gap eccentricity fault in a traction drive control system fault injection benchmark software (TDCS-FIB V2.0) platform:
firstly, selecting a sample from normal operation data of a motor without air gap eccentric fault to form a normal training data set:
Figure BDA0001920072880000091
carrying out normalization pretreatment on the data set X to obtain a data set
Figure BDA0001920072880000092
Comprises the following steps:
Figure BDA0001920072880000093
calculating a data set by the above formula (1) and formula (2)
Figure BDA0001920072880000094
And the data set
Figure BDA0001920072880000095
The covariance matrix S.
Further, calculating KL detection statistics of the normal training sample set, specifically: from normal training data sets using a sliding window method
Figure BDA0001920072880000096
Extracting z-200 samples, and forming a kth normal training sample set as follows:
Figure BDA0001920072880000097
in the formula, k is 1, …, h, in this embodiment, the step length of each sliding window movement is set to 1, and h is N-z +1 normal training sample sets are extracted; then calculating a normal training sample set
Figure BDA0001920072880000101
Is a mean matrix of
Figure BDA0001920072880000102
Sum covariance matrix
Figure BDA0001920072880000103
Computing the kth Normal training sample set
Figure BDA0001920072880000104
KL detection statistic l (k) of (a), formula:
Figure BDA0001920072880000105
the h KL detection statistics of the normal training data set are obtained as follows: l (1), …, l (59801).
Note that the normal training sample set
Figure BDA0001920072880000106
The calculation method is specifically as follows: using sliding window method to extract data from data set
Figure BDA0001920072880000107
Extracting z samples to form the kth sample set
Figure BDA0001920072880000108
k is 1, …, h; in this embodiment, the step length of each sliding window movement is set to 1, and N-z +1 normal training sample sets are extracted altogether.
Further, a detection threshold J is calculatedth。
Collecting known air gap eccentric fault data to form a fault training data set; injecting air gap eccentric fault information based on the fault of the traction transmission control system, wherein the fault information comprises that the fault type is an air gap eccentric fault, the fault degree is 0.3, and the fault time is 40000ms, and the fault information is specifically shown in the following table 1:
TABLE 1 TDCS-FIB V2.0 platform Motor Fault
Figure BDA0001920072880000109
In this embodiment, the air gap eccentricity fault data sample is:
Figure BDA00019200728800001010
in the present embodiment, f is 1;
the fault data samples obtained by preprocessing are as follows:
Figure BDA00019200728800001011
preferably, the step length of each sliding window movement is set to 1, further, t-N-d +1 sliding window method acquisitions are performed on the data set, and each time d-200 fault samples are taken, then:
Figure BDA0001920072880000111
wherein q is 1, …, t;
calculating the mean matrix of the fault samples each time a sliding window is performed
Figure BDA0001920072880000112
Covariance matrix
Figure BDA0001920072880000113
Q-th fault training sample set for calculating air gap eccentricity fault
Figure BDA0001920072880000114
KL detection statistic l1(q) the formula:
Figure BDA0001920072880000115
after t sliding windows, t KL detection statistics of the air gap eccentric fault training sample set are obtained as follows: l1(1),…,l1(59801)；
Q fault training sample set for calculating air gap eccentric fault
Figure BDA0001920072880000116
KL detection statistic l1(q) fault probability density function P1(q) is:
Figure BDA0001920072880000117
probability density of failure function of air gap eccentricity is P1(1)，…，P1(59801). In the same way, the other 3 kinds of fault data of the motor are collected from the platform: broken bar fault, bearing fault, turn-to-turn short circuit fault. And respectively calculating the fault probability density function P of the 3 fault dataf(q)，f＝2,…,4；
From 59801 probability density functions P by adopting sliding window methodf(1),…,Pf(t), f is 1, …,4, sequentially extracting g is 100 data, and forming the r-th probability density function set corresponding to the f-th fault
Figure BDA0001920072880000118
The step length of each sliding window movement is 1, and r is extractedlT-g +1 f-th fault sample probability density function sets;
further, collecting motor operation data injected with air gap eccentric faults from real-time operation data of a traction drive control system fault injection reference software (TDCS-FIB V2.0) platform to form a test data set:
Figure BDA0001920072880000119
the method comprises the following steps of (1) collecting a test data set by a sliding window method for t-N-d +1 times, wherein d-200 samples are taken each time:
Figure BDA0001920072880000121
calculating the mean matrix of the corresponding data set each time a sliding window is performed
Figure BDA0001920072880000122
Sum covariance matrix
Figure BDA0001920072880000123
Computing the qth test sample set YqdKL detection statistic ly(q) is:
Figure BDA0001920072880000124
then the t KL detection statistics of the test sample set are: ly(1),…,ly(59801)。
And then judging the fault condition of the test data. Specifically, the average value of t KL detection statistics of the test sample set and a detection threshold J are calculatedthComparing; if the average value of the detection statistics of the test sample set is larger than the threshold value JthJudging the running state of the motor to be a fault;
after the occurrence of the fault is detected, calculating the q test sample set Y by adopting a nuclear density estimation methodqdKL detection statistic lyProbability density function P of (q)y(q) the formula:
Figure BDA0001920072880000125
then t KL detection statistics l corresponding to the test sample sety(q) a probability density function of Py(1)，…，Py(59801)。
From 59801 probability density functions P by adopting sliding window methody(1)，…，Py(59801) Sequentially extracting g-100 data to form the probability density function set of the r-th test sample
Figure BDA0001920072880000126
In this embodiment, the step length of each sliding window movement is 1, and r is extracted altogetherlT-g +1 sets of probability density functions.
Calculating the probability density function set of the r test sample
Figure BDA0001920072880000127
Probability density function set of class f fault sample
Figure BDA0001920072880000128
KL divergence (distance) between
Figure BDA0001920072880000129
Figure BDA00019200728800001210
In this embodiment, the KL divergence (distance) between the r-th test sample probability density function set and the 4-class fault sample probability density function set is:
Figure BDA00019200728800001211
r under f-type faultlThe individual KL divergence (distance) values are:
Figure BDA00019200728800001212
form it into a distance set
Figure BDA0001920072880000131
Then r islThe KL divergence (distance) set between the probability density function set of each test sample and the probability density function set of the n-type fault samples is respectively as follows:
Figure BDA0001920072880000132
and further, judging the fault type based on the area of the box type graph. Put KL divergence (distance) set DfIn the order from small to large
Figure BDA0001920072880000133
Calculating the lower quartile of each distance data
Figure BDA0001920072880000134
Median number
Figure BDA0001920072880000135
Upper quartile
Figure BDA0001920072880000136
Upper limit UmaxAnd lower limit UminAll D are plotted using the calculated values of equation 16f1, …, n; respectively calculate all corresponding DfArea S of box plotf：
Sf＝IQRf×W；(13)
In the formula, W is the base width, and in this embodiment, the value is 1. Specifically, when a fault occurs, the area calculation value of the box type diagram is shown in table 2 below, the waveform diagram of the fault detection result is shown in fig. 2, and the box type diagram of the fault diagnosis result is shown in fig. 3; when a bearing fault occurs, the area calculation value of the box type diagram is shown in the following table 3, the fault detection result waveform diagram is shown in fig. 4, and the fault diagnosis result box type diagram is shown in fig. 5; when a turn-to-turn short circuit fault occurs, the area calculation value of the box diagram is shown in the following table 4, the fault detection result waveform diagram is shown in fig. 6, and the fault diagnosis result box diagram is shown in fig. 7;
TABLE 2 area calculation values of box plots in the event of a broken bar fault
Figure BDA0001920072880000137
TABLE 3 area calculation values of box plot in case of bearing failure
Figure BDA0001920072880000138
TABLE 4 area calculation values of box plots in the event of turn-to-turn short circuit fault
Figure BDA0001920072880000141
Judging the type of the fault with the smallest area:
min(Sf)→f；
wherein min (S)f) → f denotes taking SfThe minimum value of the medium area corresponds to the fault, in the embodiment, the fault is the motor air gap eccentric fault, the waveform diagram of the fault detection result is shown in fig. 8, and the diagnosis result is shown in the chart box type diagram in fig. 9. In fig. 9, the box-shaped area corresponding to the motor air gap eccentricity fault is the smallest.
Example 2
In correspondence with the above method embodiments, the present embodiment provides a motor fault diagnosis system, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (9)

1. A motor fault diagnosis method is characterized by comprising the following steps:
s1: selecting normal operation data of a motor to be detected as a normal training data set, calculating a first detection statistic according to the normal training data set, and calculating a detection threshold according to the first detection statistic;
s2: selecting different types of fault data from historical fault operation data as a fault training data set, calculating second detection statistics according to the fault training data set, calculating a fault probability density function of the second detection statistics by adopting a kernel density estimation method, and constructing probability density function sets of all types of fault samples;
s3: selecting real-time operation data of a motor to be tested as a test data set, calculating third detection statistic according to the test data set, comparing the third detection statistic with the detection threshold, if the value of the third detection statistic is larger than the detection threshold, judging that a fault occurs, and entering S4; otherwise, the test data set is selected again until the motor is detected to be in fault;
s4: and calculating a probability density function of the third detection statistic by adopting a kernel density estimation method, constructing a probability density function set of a test data set, and diagnosing the fault type according to the distance between the probability density function set of the test data set and the probability density function sets of all types of fault samples in the S2.
2. The motor fault diagnosis method according to claim 1, wherein the S1 specifically includes the steps of:
s11: the normal training data set X is represented as:
Figure FDA0003016582950000011
in the formula, m is the number of the motor sensors, N is the number of sampling points, x is a motor operation data sample collected according to a time sequence,
Figure FDA0003016582950000012
is a real number set;
s12: carrying out normalization pretreatment on the data set X to obtain a data set
Figure FDA0003016582950000013
Comprises the following steps:
Figure FDA0003016582950000014
computing a data set
Figure FDA0003016582950000015
The calculation formula of the mean matrix μ is:
Figure FDA0003016582950000016
in the formula (I), the compound is shown in the specification,
Figure FDA0003016582950000021
computing a data set
Figure FDA0003016582950000022
The calculation formula of the covariance matrix S is as follows:
Figure FDA0003016582950000023
calculating a first detection statistic by adopting a sliding window method, wherein the calculation formula is as follows:
Figure FDA0003016582950000024
wherein l (k) is the kth normal training sample set
Figure FDA0003016582950000025
Is detected in the first detection statistic of (a),
Figure FDA0003016582950000026
for the kth normal training sample set
Figure FDA0003016582950000027
Z is the number of samples of the kth normal training sample set,
Figure FDA0003016582950000028
for the kth normal training sample set
Figure FDA0003016582950000029
Of the covariance matrix, S-1Is the inverse matrix of the covariance matrix of the normal training data set, | | | is a two-norm, tr () is the trace of the matrix;
after h sliding windows, h first detection statistics of the normal training data set are obtained as follows: l (1), …, l (h);
s13: randomly extracting B statistics from h first detection statistics of a normal training data set to form a set { l }1,…,lBB < h, and reordering the B sample samples in descending order: l(1)＜l(2)＜…＜l(w)Let w be the maximum statistical value l(w)λ × α, where α is an allowable false alarm rate, λ is a random sampling number, and after λ -times of sampling is repeated, the following results are obtained:
Figure FDA00030165829500000210
calculating a detection threshold JthThe calculation formula is as follows:
Figure FDA00030165829500000211
3. the motor fault diagnosis method according to claim 2, wherein the S2 specifically includes the steps of:
s21: selecting n types of fault data samples from historical fault operation data, wherein the f type of fault samples form a fault training data set XfComprises the following steps:
Figure FDA0003016582950000031
wherein f is 1, …, n;
for data set XfCarrying out normalization processing to obtain a fault training data set
Figure FDA0003016582950000032
Comprises the following steps:
Figure FDA0003016582950000033
wherein f is 1, …, n;
s22: computing a data set
Figure FDA0003016582950000034
Is a mean matrix of
Figure FDA0003016582950000035
The calculation formula is as follows:
Figure FDA0003016582950000036
in the formula (I), the compound is shown in the specification,
Figure FDA0003016582950000037
wherein d is a slave data set
Figure FDA0003016582950000038
The number of the samples extracted in the step (2),q is the qth fault training data set, and i is 1, …, m, q is 1, …, t, t is the number of fault training data sets collected by the sliding window method;
computing a data set
Figure FDA0003016582950000039
Covariance matrix of
Figure FDA00030165829500000310
The calculation formula is as follows:
Figure FDA00030165829500000311
q fault training sample set for calculating f fault by adopting sliding window method
Figure FDA00030165829500000312
Second detection statistic lf(q) the calculation formula is:
Figure FDA00030165829500000313
after t sliding windows, t second detection statistics of the f-th type fault training sample set are obtained as follows: lf(1),…,lf(t)；
S23: calculating a qth fault training sample set of the f-type fault by adopting a kernel density estimation method
Figure FDA00030165829500000314
Second detection statistic lf(q) fault probability density function Pf(q) the calculation formula is:
Figure FDA0003016582950000041
probability density of t second detection statistics corresponding to class f faultDegree function of Pf(1)，…，Pf(t)；
From t probability density functions P by sliding window methodf(1),…,PfG data are sequentially extracted in (t), and an r probability density function set corresponding to the f type fault is formed as follows:
Figure FDA0003016582950000042
wherein r is 1, …, rl，rlThe number of the f-th type fault sample probability density function sets.
4. The method according to claim 3, wherein the step of calculating the third detection statistic from the fault training data set in the step S3 specifically includes the steps of:
s31: selecting real-time operation data of the motor to be tested as a test data set Y:
Figure FDA0003016582950000043
carrying out normalization pretreatment on the test data set Y to obtain a data set
Figure FDA0003016582950000044
Comprises the following steps:
Figure FDA0003016582950000045
computing a test data set
Figure FDA0003016582950000046
Is a mean matrix of
Figure FDA0003016582950000047
The calculation formula is as follows:
Figure FDA0003016582950000048
in the formula (I), the compound is shown in the specification,
Figure FDA0003016582950000049
and i is 1, …, m, q is 1, …, t, where d is the number of samples extracted from the data set Y and q is the qth test training data set;
calculating a covariance matrix for test dataset Y
Figure FDA00030165829500000410
The calculation formula is as follows:
Figure FDA0003016582950000051
calculating the qth test sample set Y by adopting a sliding window methodqdThird detection statistic ly(q) the calculation formula is:
Figure FDA0003016582950000052
after t sliding windows, t third detection statistics of the test sample set are obtained as follows: ly(1),…,ly(t)。
5. The motor fault diagnosis method according to claim 1, wherein in the S1, the normal operation data includes: the operating voltage, current, power and speed of the motor.
6. The motor fault diagnosis method according to claim 4, wherein the S4 specifically includes the steps of:
s41: calculating the qth test sample set Y by using a nuclear density estimation methodqdThird detection statistic lyProbability density function of (q)Py(q) the calculation formula is:
Figure FDA0003016582950000053
wherein q is 1, …, t;
t third detection statistics l corresponding to the set of test samplesy(q) a probability density function of Py(1)，…，Py(t)；
From t probability density functions P by sliding window methody(1)，…，PyG data are sequentially extracted in (t), and the probability density function set of the r test sample is formed as follows:
Figure FDA0003016582950000054
wherein r is 1, …, rl；
Calculating the distance between the probability density function set of the r test sample and the probability density function set of the f type fault sample, wherein the calculation formula is as follows:
Figure FDA0003016582950000055
the distance between the probability density function set of the r test sample and the probability density function set of the n types of fault samples is as follows:
Figure FDA0003016582950000056
r under f-type faultlThe distance values are:
Figure FDA0003016582950000057
form it into a distance set
Figure FDA0003016582950000061
Then r islOf distances between the set of probability density functions of individual test samples and the set of probability density functions of n-type fault samplesThe sets are respectively:
Figure FDA0003016582950000062
set D of the above distancesfThe distance data in (1) is rearranged from small to large
Figure FDA0003016582950000063
Calculating the lower quartile of each distance data by adopting a box type graph area calculation method
Figure FDA0003016582950000064
Median number
Figure FDA0003016582950000065
Upper quartile
Figure FDA0003016582950000066
Upper limit UmaxAnd lower limit UminComprises the following steps:
Figure FDA0003016582950000067
in the formula
Figure FDA0003016582950000068
It is indicated that the direction is positive in the upward direction,
Figure FDA0003016582950000069
to get
Figure FDA00030165829500000610
Centralizing data at corresponding positions;
all D are plotted against the calculated values of equation (20)f1, …, n; respectively calculate all corresponding DfArea S of box plotf：
Sf＝IQRf×W； (23)
In the formula, W is the width of the bottom edge;
s42: judging all DfAnd judging the fault type corresponding to the box type graph as the fault type of the motor to be tested.
7. The motor fault diagnosis method according to claim 4, characterized in that the first detection statistic, the second detection statistic, and the third detection statistic are all of the type of KL detection statistic.
8. The motor fault diagnosis method according to claim 6, wherein in S41, the distance between the r-th test sample probability density function set and the n-type fault sample probability density function set is KL divergence between the r-th test sample probability density function set and the n-type fault sample probability density function set.
9. A motor fault diagnosis system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method according to any of the preceding claims 1 to 8 are implemented when the computer program is executed by the processor.
CN201811590223.2A 2018-12-25 2018-12-25 Motor fault diagnosis method and system Active CN109447187B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811590223.2A CN109447187B (en) 2018-12-25 2018-12-25 Motor fault diagnosis method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811590223.2A CN109447187B (en) 2018-12-25 2018-12-25 Motor fault diagnosis method and system

Publications (2)

Publication Number Publication Date
CN109447187A CN109447187A (en) 2019-03-08
CN109447187B true CN109447187B (en) 2021-06-15

Family

ID=65535562

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811590223.2A Active CN109447187B (en) 2018-12-25 2018-12-25 Motor fault diagnosis method and system

Country Status (1)

Country Link
CN (1) CN109447187B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110222765B (en) * 2019-06-06 2022-12-27 中车株洲电力机车研究所有限公司 Method and system for monitoring health state of permanent magnet synchronous motor
CN110428004B (en) * 2019-07-31 2021-02-05 中南大学 Mechanical part fault diagnosis method based on deep learning under data imbalance
CN110399854B (en) * 2019-07-31 2020-10-23 中南大学 Rolling bearing fault classification method based on mixed feature extraction
CN110729054B (en) * 2019-10-14 2023-04-11 深圳平安医疗健康科技服务有限公司 Abnormal diagnosis behavior detection method and device, computer equipment and storage medium
CN110806733B (en) * 2019-10-30 2021-09-21 中国神华能源股份有限公司国华电力分公司 Thermal power plant equipment monitoring method and device and electronic equipment
CN111160776A (en) * 2019-12-30 2020-05-15 华东理工大学 Method for detecting abnormal working condition in sewage treatment process by utilizing block principal component analysis
CN112098850B (en) * 2020-09-21 2024-03-08 山东工商学院 Lithium ion battery voltage fault diagnosis method and system based on SDO algorithm
CN112668867B (en) * 2020-12-24 2024-08-06 中国电力科学研究院有限公司 Equipment failure rate assessment method and system based on field data volume
CN113051092B (en) * 2021-02-04 2022-05-17 中国人民解放军国防科技大学 Fault diagnosis method based on optimized kernel density estimation and JS divergence
CN113123932B (en) * 2021-02-23 2023-01-13 北京华能新锐控制技术有限公司 Real-time dynamic fault diagnosis method for wind turbine generator

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4261206A (en) * 1978-08-18 1981-04-14 Mitsubishi Denki Kabushiki Kaisha Fault diagnostic device for winding of electric machinery and apparatus
CN103868692A (en) * 2014-03-18 2014-06-18 电子科技大学 Rotary machine fault diagnosis method based on kernel density estimation and K-L divergence
CN104392082A (en) * 2014-07-10 2015-03-04 中山火炬职业技术学院 Diagnosis method for initial failure of gearbox of wind generating set based on vibration monitoring
CN106710653A (en) * 2016-12-05 2017-05-24 浙江大学 Real-time data abnormal diagnosis method for monitoring operation of nuclear power unit
CN108107360A (en) * 2017-12-05 2018-06-01 中国电子产品可靠性与环境试验研究所 Electrical fault discrimination method and system
CN108830006A (en) * 2018-06-27 2018-11-16 中国石油大学(华东) Linear-nonlinear industrial processes fault detection method based on the linear evaluation factor

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8040139B2 (en) * 2009-02-16 2011-10-18 Maxim Integrated Products, Inc. Fault detection method for detecting leakage paths between power sources and chassis

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4261206A (en) * 1978-08-18 1981-04-14 Mitsubishi Denki Kabushiki Kaisha Fault diagnostic device for winding of electric machinery and apparatus
CN103868692A (en) * 2014-03-18 2014-06-18 电子科技大学 Rotary machine fault diagnosis method based on kernel density estimation and K-L divergence
CN104392082A (en) * 2014-07-10 2015-03-04 中山火炬职业技术学院 Diagnosis method for initial failure of gearbox of wind generating set based on vibration monitoring
CN106710653A (en) * 2016-12-05 2017-05-24 浙江大学 Real-time data abnormal diagnosis method for monitoring operation of nuclear power unit
CN108107360A (en) * 2017-12-05 2018-06-01 中国电子产品可靠性与环境试验研究所 Electrical fault discrimination method and system
CN108830006A (en) * 2018-06-27 2018-11-16 中国石油大学(华东) Linear-nonlinear industrial processes fault detection method based on the linear evaluation factor

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Electric motor defects diagnosis based on kernel density estimation and Kullback–Leibler divergence in quality control scenario;Francesco Ferracuti 等;《Engineering ApplicationsofArtificial Intelligence》;20150505;第44卷;第25-32页 *
Minor fault diagnosis based on fractional-order model of permanent magnet synchronous motor;Wei Yu 等;《IECON 2017-43rd Annual Conference of the IEEE Industrial Electronics Society》;20171218;第8082-8086页 *

Also Published As

Publication number Publication date
CN109447187A (en) 2019-03-08

Similar Documents

Publication Publication Date Title
CN109447187B (en) Motor fault diagnosis method and system
CN103698699B (en) A kind of asynchronous motor malfunction monitoring diagnostic method based on model
CN103926533B (en) Permagnetic synchronous motor loss of excitation on-line fault diagnosis method and system
CN103926506B (en) Based on the rotor of steam turbo generator Winding Short Fault Diagnosis method of constructor
Sin et al. Induction machine on-line condition monitoring and fault diagnosis-a survey
Almounajjed et al. Diagnosis of stator fault severity in induction motor based on discrete wavelet analysis
Xu et al. Data-driven inter-turn short circuit fault detection in induction machines
CN103983452B (en) Utilize the method that hybrid domain characteristic vector and grey correlation analysis carry out Fault Pattern Recognition to epicyclic gearbox
CN113748326A (en) Method for estimating severity of bearing failure for induction motor
CN103792463B (en) Steam-turbine generator rotor winding short trouble diagnosis method based on fictitious power principle
EP3702238A1 (en) Diagnostic system and a method of diagnosing faults
CN110222765B (en) Method and system for monitoring health state of permanent magnet synchronous motor
RU2431152C2 (en) Method of diagnostics of electrically driven mechanisms and systems
CN113325314A (en) Motor fault diagnosis method
CN106093626B (en) A method of the small fault for diagnosing CRH5 type high ferro inverters
Saidi et al. Stator current bi-spectrum patterns for induction machines multiple-faults detection
Zhou et al. A summary of health prognostics methods for industrial robots
Ciszewski Induction motor bearings diagnostic indicators based on MCSA and normalized triple covariance
Sharp Prognostic approaches using transient monitoring methods
Dias et al. An experimental approach for diagnosis of adjacent and nonadjacent broken bars in induction motors at very low slip
Mbo'o et al. Bearing damage diagnosis by means of the linear discriminant analysis of stator current feature
Navarro et al. Condition monitoring system for characterization of electric motor ball bearings with distributed fault using fuzzy inference tools
CN114112390B (en) Nonlinear complex system early fault diagnosis method
Khelfi et al. Temporal Envelope Estimation of Stator Current by Peaks Detection for IM Fault Diagnosis
Vinayak et al. Inter phase fault detection in inverter fed induction motor using wavelet transform

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