CN109242008B - Compound fault identification method under incomplete sample class condition - Google Patents

Compound fault identification method under incomplete sample class condition Download PDF

Info

Publication number
CN109242008B
CN109242008B CN201810974303.1A CN201810974303A CN109242008B CN 109242008 B CN109242008 B CN 109242008B CN 201810974303 A CN201810974303 A CN 201810974303A CN 109242008 B CN109242008 B CN 109242008B
Authority
CN
China
Prior art keywords
sample
fault
normal
formula
samples
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
CN201810974303.1A
Other languages
Chinese (zh)
Other versions
CN109242008A (en
Inventor
易辉
刘宇芳
张霞
陈溪
朱浩男
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.)
Nanjing Tech University
Original Assignee
Nanjing Tech 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 Nanjing Tech University filed Critical Nanjing Tech University
Priority to CN201810974303.1A priority Critical patent/CN109242008B/en
Publication of CN109242008A publication Critical patent/CN109242008A/en
Application granted granted Critical
Publication of CN109242008B publication Critical patent/CN109242008B/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
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Test And Diagnosis Of Digital Computers (AREA)

Abstract

The invention provides a composite fault identification method under the condition of incomplete sample category, which comprises the following steps: generating a fault detector; generating an evaluation function; sending the sample to be detected to a fault detector, and judging whether the output result has a fault or not; respectively sending the samples to be tested to corresponding evaluation functions to obtain corresponding index values; comprehensively evaluating the indexes; and setting a threshold value theta to diagnose whether a composite fault exists or not. The composite fault identification method respectively calculates the high-dimensional spatial positions of the samples to be detected in the fault A classifier and the fault B classifier, then estimates the possibility of AB faults occurring simultaneously, and further realizes the identification of the composite faults on the premise of not needing AB composite fault samples.

Description

Compound fault identification method under incomplete sample class condition
Technical Field
The invention relates to a compound fault identification method, in particular to a compound fault identification method under the condition of incomplete sample categories.
Background
Under the background of artificial intelligence and big data, the diagnosis system of the current industrial system gradually changes from relying on an accurate mathematical model to relying on industrial process data and a data driving algorithm. A fault diagnosis method based on a Support Vector Machine (SVM) is a typical data-driven fault diagnosis method, which avoids the complex and even unrealizable mechanism modeling of a diagnostic object, and simultaneously avoids the requirement of an intelligent diagnosis algorithm with minimized experience risk, such as an artificial neural network, on the size of a training sample set, and can realize effective fault diagnosis on the premise of a small sample amount, thereby obtaining wide research and application in recent years.
The SVM bi-classifier is a typical bi-classifier, and given training sample sets A and B, the built SVM bi-classifier can judge whether unknown samples belong to class A or class B with higher probability. The essence of fault diagnosis based on the SVM method is that a nonlinear fault sample set is projected to a high-dimensional fault space, and a classification hyperplane is made through the SVM so that the fault space is divided into corresponding fault intervals. And the fault detection and isolation are realized by positioning the fault section of the sample to be detected in the high-dimensional space.
The performance of the traditional fault diagnosis based on SVM fundamentally depends on the quality of a data set, and accurate diagnosis can be realized by training all class samples at one time, namely the sample set is required to be complete. By taking fig. 1 as an example, if only a normal sample, a fault a sample and a fault B sample are collected, the conventional SVM diagnosis method can only diagnose three working states of normal, fault a and fault B, and cannot judge whether a composite fault occurs (fault A, B occurs simultaneously); if the diagnosis of the state needs to be realized, a fault sample under the condition of AB fault concurrence needs to be collected, and then a new classifier is constructed.
In actual engineering, the fault itself usually belongs to a small probability event, the composite fault is a small probability of a small probability, and a sample is extremely difficult to obtain, so that the sample set is usually incomplete.
Disclosure of Invention
The invention aims to: the method for identifying the compound faults under the condition of incomplete sample classes can calculate the high-dimensional spatial positions of samples to be detected in a fault A classifier and a fault B classifier respectively, then estimate the possibility of AB faults occurring simultaneously, and further realize the identification of the compound faults on the premise of not needing AB compound fault samples.
In order to achieve the above object, the present invention provides a method for identifying a compound fault under the condition of incomplete sample class, which comprises the following steps:
step 1, generating a fault detector, which comprises the following specific steps: collecting Normal sample XNormalConstructing hypersphere omega by using support vector data description method1If the sample to be measured falls into the hypersphere omega1If the range is within the range, the sample to be detected belongs to a normal sample, otherwise, the sample to be detected belongs to a fault sample;
step 2, generating an evaluation function, which comprises the following specific steps:
step 2.1, collecting a fault sample set X of the fault AA；
Step 2.2, respectively, normal sample XNormalAnd fault sample set XASetting corresponding labels, marking the normal sample as 1 and the fault sample as-1, and further obtaining a label set
Figure BDA0001777025240000021
N is the total number of samples;
step 2.3, solving the normal sample X by using a support vector machineNormalAnd fault sample set XAThe classification hyperplane w x + b is 0, w is a weight vector, b is a bias, and then the solution of the hyperplane is converted into the following quadratic programming problem:
min:＜w·w＞ (3)
S.t.yi(＜w·xi＞+b)≥1,i＝1,...,N
the support vector machine method projects the existing sample to a linearly separable high-dimensional space, if w x + b is larger than 0, the sample is considered to be a normal sample, and if w x + b is less than or equal to 0, the sample is considered to be a fault sample;
step 2.3, evaluating the sample by using the distance between the sample to be measured and the hyperplane, and giving the sample x to be measuredtThen the evaluation function for the a fault is:
Figure BDA0001777025240000022
in the formula, wA、bAAnd classifying the weight vector and the offset of the hyperplane by using the normal sample and the A fault sample, and generating an evaluation function of the B fault in the same way:
Figure BDA0001777025240000023
in the formula, wB、bBWeight vector and bias for classifying hyperplane with normal samples and B fault samples；
Step 3, sending the sample to be detected to a fault detector, if the high-dimensional projection of the sample is in the hypersphere, the sample is a normal sample, the output result is no fault, otherwise, the fault occurs, and entering step 4;
step 4, respectively sending the samples to be tested to the evaluation function f of the A faultsA(xt) And B fault evaluation function fB(xt) In the method, an index value eta is obtainedA＝fA(xt) And ηB＝fB(xt)；
Step 5, comprehensively evaluating the indexes: giving a comprehensive evaluation function D, and calculating according to the formula:
Figure BDA0001777025240000031
and 6, setting a threshold value theta, wherein theta is more than or equal to 0 and less than 1, when D is less than or equal to theta, the diagnosis result is no composite fault, and if D is more than theta, the diagnosis result is composite fault.
Further, in step 1, a hyper-sphere omega is built1The method comprises the following specific steps:
is provided with
Figure BDA0001777025240000032
For a known set of target samples, N is the number of samples, hypersphere
Figure BDA0001777025240000036
) Can be combined with
Figure BDA0001777025240000033
Is completely contained, wherein a is the centre of sphere, R is the radius of the hypersphere, has:
Figure BDA0001777025240000034
in the formula, C is a given penalty factor, xi is misjudgment loss, the solution is realized by combining Lagrangian formula with an activity set method, and the solving formula is as follows:
Figure BDA0001777025240000035
in the formula, Z is a sample element in the test sample Z, and for the test sample Z, if the above formula is satisfied, the test sample Z belongs to the target class.
Further, in step 6, the value of θ is set to 0.2.
The invention has the beneficial effects that: the composite fault identification method respectively calculates the high-dimensional spatial positions of the samples to be detected in the fault A classifier and the fault B classifier, then estimates the possibility of AB faults occurring simultaneously, and further realizes the identification of the composite faults on the premise of not needing AB composite fault samples.
Drawings
FIG. 1 is the essence of SVM based fault diagnosis in the prior art;
FIG. 2 is a schematic flow diagram of the process of the present invention;
FIG. 3 is a schematic diagram of a hypersphere structure of the present invention;
FIG. 4 is a schematic diagram of a method for projecting a conventional sample into a linearly separable high-dimensional space.
Detailed Description
As shown in fig. 2, the method for identifying a compound fault under the incomplete sample category condition disclosed by the invention comprises the following steps:
step 1, generating a fault detector, which comprises the following specific steps: collecting Normal sample XNormalAdopting Support Vector Data Description (SVDD) method to construct hyper-sphere omega1If the sample to be measured falls into the hypersphere omega1If the range is within the range, the sample to be detected belongs to a normal sample, otherwise, the sample to be detected belongs to a fault sample;
step 2, generating an evaluation function, which comprises the following specific steps:
step 2.1, collecting a fault sample set X of the fault AA；
Step 2.2, respectively, normal sample XNormalAnd fault sample set XAThe corresponding label is set, the normal sample is marked 1,the fault sample is marked as-1, and a label set is further obtained
Figure BDA0001777025240000041
N is the total number of samples;
step 2.3, solving a normal sample X by utilizing a Support Vector Machine (SVM)NormalAnd fault sample set XAThe classification hyperplane w x + b is 0, w is a weight vector, b is a bias, and then the solution of the hyperplane is converted into the following quadratic programming problem:
min:＜w·w＞ (3)
S.t.yi(＜w·xi＞+b)≥1,i＝1,...,N
the support vector machine method projects the existing sample to a linearly separable high-dimensional space, as shown in fig. 4, if w x + b is greater than 0, the sample is considered as a normal sample, and if w x + b is less than or equal to 0, the sample is considered as a fault sample;
step 2.3, evaluating the sample by using the distance between the sample to be measured and the hyperplane, and giving the sample x to be measuredtThen the evaluation function for the a fault is:
Figure BDA0001777025240000042
in the formula, wA、bAAnd classifying the weight vector and the offset of the hyperplane by using the normal sample and the A fault sample, and generating an evaluation function of the B fault in the same way:
Figure BDA0001777025240000043
in the formula, wB、bBClassifying the weight vector and the offset of the hyperplane by using a normal sample and a B fault sample;
step 3, sending the sample to be detected to a fault detector, if the high-dimensional projection of the sample is in the hypersphere, the sample is a normal sample, the output result is no fault, otherwise, the fault occurs, and entering step 4;
step 4, respectively sending the samples to be detectedEvaluation function f to A faultA(xt) And B fault evaluation function fB(xt) In the method, an index value eta is obtainedA＝fA(xt) And ηB＝fB(xt)；
Step 5, comprehensively evaluating the indexes: giving a comprehensive evaluation function D, and calculating according to the formula:
Figure BDA0001777025240000051
and 6, setting a threshold value theta, wherein theta is more than or equal to 0 and less than 1, when D is less than or equal to theta, the diagnosis result is no composite fault, if D is more than theta, the diagnosis result is composite fault, the value of theta is not too high, and the value of theta is preferably set to be 0.2.
Further, in step 1, the SVDD method tries to adopt a minimum hypersphere structure, and includes all training samples to establish hypersphere Ω1The method comprises the following specific steps:
is provided with
Figure BDA0001777025240000052
For a known set of target samples, N is the number of samples, hypersphere
Figure BDA0001777025240000056
Can be combined with
Figure BDA0001777025240000053
Is completely contained, wherein a is the centre of sphere, R is the radius of the hypersphere, has:
Figure BDA0001777025240000054
in the formula, C is a given penalty factor, xi is misjudgment loss, the solution is realized by combining Lagrangian formula with an activity set method, and the solving formula is as follows:
Figure BDA0001777025240000055
in the formula, Z is a sample element in the test sample Z, and for the test sample Z, if the above formula is satisfied, the test sample Z belongs to the target class, as shown in fig. 3.
The composite fault identification method under the incomplete sample class condition disclosed by the invention respectively calculates the high-dimensional spatial positions of the sample to be detected in the fault A classifier and the fault B classifier, then estimates the possibility of AB faults occurring simultaneously, and further realizes the identification of the composite fault on the premise of not needing AB composite fault samples.

Claims (3)

1. A composite fault identification method under the condition of incomplete sample category is characterized by comprising the following steps:
step 1, generating a fault detector, which comprises the following specific steps: collecting Normal sample XNormalConstructing hypersphere omega by using support vector data description method1If the sample to be measured falls into the hypersphere omega1If the range is within the range, the sample to be detected belongs to a normal sample, otherwise, the sample to be detected belongs to a fault sample;
step 2, generating an evaluation function, which comprises the following specific steps:
step 2.1, collecting a fault sample set X of the fault AA；
Step 2.2, respectively, normal sample XNormalAnd fault sample set XASetting corresponding labels, marking the normal sample as 1 and the fault sample as-1, and further obtaining a label set
Figure FDA0003169081630000011
N is the total number of samples;
step 2.3, solving the normal sample X by using a support vector machineNormalAnd fault sample set XAThe classification hyperplane w x + b is 0, w is a weight vector, b is a bias, and then the solution of the hyperplane is converted into the following quadratic programming problem:
min:<w·w> (3)
S.t.yi(<w·xi>+b)≥1,i＝1,...,N
the support vector machine method projects the existing sample to a linearly separable high-dimensional space, if w x + b is larger than 0, the sample is considered to be a normal sample, and if w x + b is less than or equal to 0, the sample is considered to be a fault sample;
step 2.4, evaluating the sample by using the distance between the sample to be measured and the hyperplane, and giving the sample x to be measuredtThen the evaluation function for the a fault is:
Figure FDA0003169081630000012
in the formula, wA、bAAnd classifying the weight vector and the offset of the hyperplane by using the normal sample and the A fault sample, and generating an evaluation function of the B fault in the same way:
Figure FDA0003169081630000013
in the formula, wB、bBClassifying the weight vector and the offset of the hyperplane by using a normal sample and a B fault sample;
step 3, sending the sample to be detected to a fault detector, if the high-dimensional projection of the sample to be detected is in the hypersphere, the sample to be detected is a normal sample, the output result is no fault, otherwise, the fault occurs, and the step 4 is entered;
step 4, respectively sending the samples to be tested to the evaluation function f of the A faultsA(xt) And B fault evaluation function fB(xt) In the method, an index value eta is obtainedA＝fA(xt) And ηB＝fB(xt)；
Step 5, comprehensively evaluating the indexes: giving a comprehensive evaluation function D, and calculating according to the formula:
Figure FDA0003169081630000021
and 6, setting a threshold value theta, wherein theta is more than or equal to 0 and less than 1, when D is less than or equal to theta, the diagnosis result is no composite fault, and if D is more than theta, the diagnosis result is composite fault.
2. The method for identifying the compound fault under the incomplete sample class condition according to claim 1, wherein in the step 1, a hyper-sphere omega is built1The method comprises the following specific steps:
is provided with
Figure FDA0003169081630000022
For a given target sample set, N is the total number of samples, and the hyper-sphere Ω ═ a, R can be computed
Figure FDA0003169081630000023
Is completely contained, wherein a is the centre of sphere, R is the radius of the hypersphere, has:
Figure FDA0003169081630000024
in the formula, C is a given penalty factor, xi is misjudgment loss, the solution is realized by combining Lagrangian formula with an activity set method, and the solving formula is as follows:
Figure FDA0003169081630000025
in the formula, Z is a sample element in the test sample Z, and for the test sample Z, if the above formula is satisfied, the test sample Z belongs to the target class.
3. The method for identifying the compound fault under the incomplete sample class condition according to claim 1, wherein in step 6, the value of θ is set to 0.2.
CN201810974303.1A 2018-08-24 2018-08-24 Compound fault identification method under incomplete sample class condition Active CN109242008B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810974303.1A CN109242008B (en) 2018-08-24 2018-08-24 Compound fault identification method under incomplete sample class condition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810974303.1A CN109242008B (en) 2018-08-24 2018-08-24 Compound fault identification method under incomplete sample class condition

Publications (2)

Publication Number Publication Date
CN109242008A CN109242008A (en) 2019-01-18
CN109242008B true CN109242008B (en) 2021-10-12

Family

ID=65067985

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810974303.1A Active CN109242008B (en) 2018-08-24 2018-08-24 Compound fault identification method under incomplete sample class condition

Country Status (1)

Country Link
CN (1) CN109242008B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109885032B (en) * 2019-03-13 2021-01-12 东北大学 Heterogeneous data collaborative modeling industrial fault detection method based on neighborhood projection preservation
CN113050602B (en) * 2021-03-26 2022-08-09 杭州电子科技大学 Industrial process fault classification method based on robust semi-supervised discriminant analysis

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103048133A (en) * 2012-12-03 2013-04-17 陕西科技大学 Bayesian network-based rolling bearing fault diagnosis method
CN103136587A (en) * 2013-03-07 2013-06-05 武汉大学 Power distribution network operating state classification recognition method based on support vector machine
CN106021771A (en) * 2016-05-30 2016-10-12 天河国云（北京）科技有限公司 Method and device for diagnosing faults

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103048133A (en) * 2012-12-03 2013-04-17 陕西科技大学 Bayesian network-based rolling bearing fault diagnosis method
CN103136587A (en) * 2013-03-07 2013-06-05 武汉大学 Power distribution network operating state classification recognition method based on support vector machine
CN106021771A (en) * 2016-05-30 2016-10-12 天河国云（北京）科技有限公司 Method and device for diagnosing faults

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
《Fault Diagnosis in Condition of Sample Type Incompleteness Using Support Vector Data Description》;Hui Yi等;《Mathematical Problems in Engineering》;20151231;第2015卷;第1-10页 *
《不完备故障类别下基于Multi-SVDD的高压隔离开关故障诊断方法》;陈士刚 等;《电工技术学报》;20180630;第33卷(第11期);第2439-2447页 *
《基于支持向量机的故障诊断及应用研究》;易辉;《中国博士学位论文全文数据库信息科技辑》;20131015(第10期);正文第1-96页 *

Also Published As

Publication number Publication date
CN109242008A (en) 2019-01-18

Similar Documents

Publication Publication Date Title
CN110263846B (en) Fault diagnosis method based on fault data deep mining and learning
CN103914064B (en) Based on the commercial run method for diagnosing faults that multi-categorizer and D-S evidence merge
CN108536971B (en) Bayesian model-based structural damage identification method
CN105022021B (en) A kind of state identification method of the Electric Energy Tariff Point Metering Device based on multiple agent
CN105117602B (en) A kind of metering device running status method for early warning
CN106355030A (en) Fault detection method based on analytic hierarchy process and weighted vote decision fusion
CN106371427A (en) Industrial process fault classification method based on analytic hierarchy process and fuzzy fusion
CN108921230A (en) Method for diagnosing faults based on class mean value core pivot element analysis and BP neural network
CN102340811A (en) Method for carrying out fault diagnosis on wireless sensor networks
WO2011034805A1 (en) Supervised fault learning using rule-generated samples for machine condition monitoring
CN101738998B (en) System and method for monitoring industrial process based on local discriminatory analysis
CN103901880A (en) Industrial process fault detection method based on multiple classifiers and D-S evidence fusion
CN109491338A (en) A kind of relevant method for diagnosing faults of multimode procedure quality based on sparse GMM
CN109033513A (en) Method for diagnosing fault of power transformer and diagnosing fault of power transformer device
CN109242008B (en) Compound fault identification method under incomplete sample class condition
CN111680875A (en) Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model
CN111767657B (en) Nuclear power system fault diagnosis method and system
CN108268632A (en) Abnormal information data identifies machine learning method
CN107942994A (en) A kind of satellite temperature control system method for diagnosing faults based on temperature curve feature
CN113225346A (en) Network operation and maintenance situation assessment method based on machine learning
CN113283491A (en) Fault diagnosis method of electric vehicle alternating current charging pile based on optimized deep confidence network
CN116824517B (en) Substation operation and maintenance safety control system based on visualization
CN110598902A (en) Water quality prediction method based on combination of support vector machine and KNN
CN111122811A (en) Sewage treatment process fault monitoring method of OICA and RNN fusion model
CN115220133A (en) Multi-meteorological-element rainfall prediction method, device, equipment and storage medium

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