CN107357275B - Non-gaussian industrial process fault detection method and system - Google Patents
Non-gaussian industrial process fault detection method and system Download PDFInfo
- Publication number
- CN107357275B CN107357275B CN201710623082.9A CN201710623082A CN107357275B CN 107357275 B CN107357275 B CN 107357275B CN 201710623082 A CN201710623082 A CN 201710623082A CN 107357275 B CN107357275 B CN 107357275B
- Authority
- CN
- China
- Prior art keywords
- vector
- value
- residual generation
- residual
- iteration
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 51
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 43
- 238000000034 method Methods 0.000 claims abstract description 82
- 230000008569 process Effects 0.000 claims abstract description 45
- 238000000342 Monte Carlo simulation Methods 0.000 claims abstract description 26
- 238000010219 correlation analysis Methods 0.000 claims abstract description 20
- 238000005070 sampling Methods 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims description 86
- 238000012545 processing Methods 0.000 claims description 23
- 238000005516 engineering process Methods 0.000 claims description 13
- 241001269238 Data Species 0.000 claims description 9
- 238000000354 decomposition reaction Methods 0.000 claims description 7
- 238000012544 monitoring process Methods 0.000 abstract description 5
- 238000003745 diagnosis Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 6
- 238000013480 data collection Methods 0.000 description 5
- 238000012880 independent component analysis Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 4
- 238000007619 statistical method Methods 0.000 description 3
- 238000010438 heat treatment Methods 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000010998 test method Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 235000015220 hamburgers Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000013011 mating Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41885—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Complex Calculations (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
Abstract
The present invention relates to industrial process monitoring and fault diagnosis field, a kind of non-gaussian industrial process fault detection method and system are disclosed, with convenient and practical ground on-line checking non-gaussian industrial process.The method of the present invention includes: the first step, chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, establishes residual generation using Canonical Correlation Analysis；Second step, sampling Monte Carlo method determine the respective threshold of corresponding residual generation；Third step carries out real-time online detection according to identified corresponding residual generation threshold value and to industrial process data.
Description
Technical field
The present invention relates to industrial process monitoring and fault diagnosis fields more particularly to a kind of non-gaussian industrial process failure to examine
Survey method and system.
Background technique
With the rapid development of information technology and data acquisition technology, factory and enterprise have production number quite abundant
According to resource, industrial big data era slowly arrives.In industrial processes, due to severe production environment and longtime running,
Equipment can inevitably be lost or even failure, these abnormal conditions not only influence the quality of product, but also will affect
The safe and stable operation of system, it some times happens that great production accident, the traditional detection method for fully relying on operative employee is insufficient
To solve complicated process test problems.
Data Over Cable in actual production process, nonlinear, Gauss, non-gaussian etc., for different lifes
Feature possessed by production process should select different fault detection methods, to guarantee the maximization of efficiency of fault diagnosis.
It is found through patent consulting, due to the feature more than production process variable, in recent years based on the process of multi-variate statistical analysis
Detection method, such as Independent component analysis technology (Independent component analysis, ICA), pivot analysis
(Principal component analysis's, PCA) and offset minimum binary (Partial least squares, PLS) changes
It has been widely used in the on-line checking of nongausian process into technology etc..But ICA and PCA method mainly considers individual data
Collection, does not make full use of the correlation between variable, and PLS method is primarily used to solve to return to ask although it is contemplated that correlation
Topic.Canonical correlation analysis technology (Canonical correlation analysis, CCA) considers the correlation between variable,
It can be considered a kind of extension of above-mentioned basic fundamental.Therefore, it is more to handle to have sampled the method based on canonical correlation analysis by the present invention
The problem on line detection of variable production process.
On the other hand, for nongausian process, the existing nongausian process detection method master based on multi-variate statistical analysis
It is divided into two classes, one kind is based on existing Multielement statistical analysis method combination Threshold, main threshold value determination side
Method has based on gauss hybrid models (Gaussian mixture model, GMM), based on kernel function estimation (Kernel-based)
With based on sequence quantile method (Sequential quantile estimation, SQE) etc., although these methods are answered
With, but it is still limited by the problems such as kernel functional parameter is chosen；Another kind of is the method based on ICA, and such method is not necessarily to become process
Amount carries out Gaussian Profile it is assumed that by finding out independent pivot, constructs corresponding statistic and carries out fault detection, but such methods are only examined
Consider forms data collection, and the determination of threshold value is also based on GMM, kernel-based, SQE etc..Therefore, how data to be made full use of
Correlation and more simple and practical Threshold are measured in real time nongausian process, at present could not also be enough preferable
Solution.
Summary of the invention
Present invention aims at a kind of non-gaussian industrial process fault detection method and system is disclosed, to exist convenient and practically
Line detects non-gaussian industrial process.
To achieve the above object, the invention discloses a kind of non-gaussian industrial process fault detection methods, comprising:
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes typical case
Correlation analysis method establishes residual generation；It specifically includes:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate in vector y m
Measured value of the variable at the ith sample moment；
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two
The mean value of each variable in data set is 0, obtains new data setWith
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the cross covariance of y
Matrix；
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And
Singular value decomposition, Υ=Γ Σ R are carried out to correlation matrix ΥT, wherein Γ is the left singular vector of the correlation matrix, and R is institute
The right singular vector of correlation matrix is stated, Σ is diagonal matrix；Then the first residual error is obtained using canonical correlation analysis technology to occur
Device raWith the second residual generation rb:
ra=LTy-ΣTJTu,rb=JTu-ΣLTy
Wherein,
Second step, sampling Monte Carlo method determine the first threshold and the second residual error of first residual generation respectively
The second threshold of generator；It specifically includes:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and set
Set initial thresholdCurrent iteration number s=0；
Step B2, according to Monte Carlo number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value
Are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate that statistic is corresponding when iteration s
Present threshold value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, raAnd rbThe residual signals of respectively the first residual generation and the output of the second residual generation；WithRespectively the first residual generation and the corresponding statistic of the second residual generation；WithRespectively corresponding residual error letter
Number covariance matrix inverse matrix；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf
It is invalid, continue iteration, enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε >
0 is a constant, and Δ is iteration tolerance (also known as: iteration step length)；
Third step carries out real-time online detection to industrial process data, specifically includes:
The variable data of the vector u and the vector y that acquire in real time are obtained, and according to first residual generation
Obtain corresponding residual signals with second residual generation, and calculate separately corresponding residual signals statistic and with it is corresponding
The first threshold and the second threshold be compared, if the statistic of corresponding first residual generation of gained has been more than institute
The statistic for stating first threshold or corresponding second residual generation has been more than the second threshold, then exports fault warning.
Above-mentioned first residual generation of the invention and the second residual generation are able to achieve good complementation, it is ensured that the present invention
The precision of detection and the time delay of energy optimizing detection, so that detection delay reduces, with better detection performance.As this hair
A kind of bright realization deteriorated, the first residual generation of dependence that can be simple in above-mentioned detection process or the second residual generation
It is detected, under the premise of reducing Correlation method for data processing amount, precision compares the inspection that the first and second residual generations combine
Survey method slightly reduces.
It corresponds to the above method, invention additionally discloses a kind of mating non-gaussian industrial process events for executing the above method
Hinder detection system.
The invention has the following advantages:
1, present invention firstly provides a kind of non-height combined based on canonical correlation analysis and Monte Carlo threshold learning
This process online test method, this method can be with on-line checking non-gaussian industrial process；
2, detection method proposed by the present invention is in real time, when industrial process breaks down, can timely to detect
Come；
3, the present invention is based on process datas, convenient and practical independent of accurate process model.
Below with reference to accompanying drawings, the present invention is described in further detail.
Detailed description of the invention
The attached drawing constituted part of this application is used to provide further understanding of the present invention, schematic reality of the invention
It applies example and its explanation is used to explain the present invention, do not constitute improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is non-gaussian industrial process fault detection method flow chart disclosed by the embodiments of the present invention.
Specific embodiment
The embodiment of the present invention is described in detail below in conjunction with attached drawing, but the present invention can be defined by the claims
Implement with the multitude of different ways of covering.
Embodiment 1
A kind of non-gaussian industrial process fault detection method, referring to Fig.1, comprising:
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes typical case
Correlation analysis method establishes residual generation.The step can carry out under off-line state, specifically include:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate in vector y m
Measured value of the variable at the ith sample moment.
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two
The mean value of each variable in data set is 0, obtains new data setWith
In step A2, the realization step of average value processing is gone to include:
1, Estimation of Mean,
2, average value processing is carried out to each sample:
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the cross covariance of y
Matrix.
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And it is right
Correlation matrix Υ carries out singular value decomposition, Υ=Γ Σ RT, wherein Γ is the left singular vector of the correlation matrix, and R is described
The right singular vector of correlation matrix, Σ are diagonal matrix；Then the first residual generation is obtained using canonical correlation analysis technology
raWith the second residual generation rb:
ra=LTy-ΣTJTu,rb=JTu-ΣLTy
Wherein,
Second step, sampling Monte Carlo method determine the first threshold and the second residual error of first residual generation respectively
The second threshold of generator；It specifically includes:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and set
Set initial thresholdCurrent iteration number s=0；
Step B2, according to Monte Carlo number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value
Are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate that statistic is corresponding when iteration s
Present threshold value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, raAnd rbThe residual signals of respectively the first residual generation and the output of the second residual generation；WithRespectively the first residual generation and the corresponding statistic of the second residual generation；WithRespectively corresponding residual error letter
Number covariance matrix inverse matrix；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf
It is invalid, continue iteration, enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε >
0 is a constant, and Δ is iteration tolerance；
Third step carries out real-time online detection to industrial process data, specifically includes:
The variable data of the vector u and the vector y that acquire in real time are obtained, and according to first residual generation
Obtain corresponding residual signals with second residual generation, and calculate separately corresponding residual signals statistic and with it is corresponding
The first threshold and the second threshold be compared, if the statistic of corresponding first residual generation of gained has been more than institute
The statistic for stating first threshold or corresponding second residual generation has been more than the second threshold, then exports fault warning.
For above-mentioned third step, by taking the 1st sampling instant as an example, specific step is as follows for on-line monitoring:
Equalization is gone to handle the sample data collected in real time；When the 1st sampling collected by data collection system
The measurement data at quarter is denoted asAnd equalization is carried out using the mean value that off-line training step obtains
Processing, obtains new dataWith
Obtain residual signals；Corresponding residual signals r is respectively obtained by the first and second residual generationsaAnd rbIt is as follows:
Counting statistics amount；According to obtained residual signals, corresponding statistic is established respectivelyWithIt is as follows:
Wherein,WithThe inverse matrix of the covariance matrix of respectively corresponding residual signals；
CompareWithWith the size of respective threshold, if statistic is all not more than respective threshold, u (1), y (1)
It is normal sample；Otherwise (1) u, y (1) is fault sample, shows the faulty generation of process.
[concrete application example]
Illustrate effectiveness of the invention in conjunction with the example of continuous stirred tank heating process.The data of embodiment are from moral
The TR682CSTH semi-physical emulation platform of state hamburger G.U.N.T Geraetebau GmbH exploitation.
This example has non-Gaussian system by injection zero-mean coloured noise come simulation process.6 variables of the process are chosen,
According to procedural knowledge, by liquid level L1, temperature in the kettle T1With temperature T in heating compartment2Constitute u vector；Fluid temperature T will be inputted3, store
Fountain solution temperature T4With enter flow quantity F1Constitute y vector.Data set is divided into three parts: training dataset, threshold learning data set and
Test data set.Sampling time is set as 1s, and Monte-Carlo Simulation number K=800, training dataset is by 1000 sample groups
At each is made of threshold learning data set 5000 samples, and test data set is made of 850 samples.
For the fault detection effect of testing algorithm, step and random two kinds of fault types are introduced respectively, as shown in table 1
Table 1, the description of CSTH procedure fault:
| Serial number | Fault type | Failure introduces the time |
| 1 | The stuck failure of flow valve | 401s |
| 2 | Liquid level sensor accuracy decline | 401s |
Implementation steps of the invention are explained in detail below with reference to the example of CSTH process:
First part: off-line training establishes residual generation.
1) training data and threshold learning data set are acquired.The mistake under accidental conditions is chosen using historical data base
Number of passes evidence, is used respectivelyIt indicates
Two variable data collection of process.
2) average value processing is removed.Average value processing is carried out to data set respectively, obtains new data set:
Covariance matrix is constructed, to data set into after normalized, the covariance matrix needed for formula (1) estimation,
Σu, ΣyAnd Σuy。
Construct residual generation；According to estimated obtained covariance matrix, obtained needed for residual generation using formula
Parameter J, L and Σ it is as follows then to obtain two residual generations: ra=LTy-ΣTJTu,rb=JTu-ΣLTy。
Threshold value；Respective threshold J is determined according to above-mentioned threshold learning methodth,aAnd Jth,b。
Second part: on-line monitoring, by taking the 1st sampling instant as an example, the specific implementation steps are as follows for on-line monitoring:
Equalization is gone to handle the sample data collected in real time；When the 1st sampling collected by data collection system
The measurement data at quarter is denoted asAnd equalization is carried out using the mean value that off-line training step obtains
Processing, obtains new dataWithThen corresponding residual signals r is obtainedaAnd rb。
Later, according to obtained residual signals, corresponding statistic is established respectivelyWithCompareWith Jth,a,And Jth,bIf statistic is not more than respective threshold, u (1), y (1) is normal sample；Otherwise (1) u, y (1) are failure samples
This, shows the faulty generation of process, successively detects remaining sample.The result of fault detection such as table 2.
Table 2:
The testing result of 2 kinds of fault conditions is given in Table 2, and the calculating of verification and measurement ratio and detection delay is using Multi simulation running
The method being averaged calculates.From the fault detection rate of table 2 and detection delay it can be seen that the method for the present invention monitors work on-line
The feasibility and validity of industry process, and compared with the CCA method of traditional processing Gaussian process, the failure of the method for the present invention
Verification and measurement ratio is higher and detection delay is lower, has better detection performance.
Non-gaussian industrial process fault detection method disclosed in the present embodiment, above-mentioned first residual generation and the second residual error
Generator is able to achieve good complementation, it is ensured that the time delay of precision and energy optimizing detection that the present invention detects, so that detection delay
Property reduce, have better detection performance.
Embodiment 2
As a kind of realization deteriorated of above-described embodiment 1, the first residual error of dependence that can be simple in above-mentioned detection process
Generator or the second residual generation are detected, and under the premise of reducing Correlation method for data processing amount, precision is compared to first and the
The detection method that two residual generations combine slightly reduces.
Non-gaussian industrial process fault detection method (can refer to above-described embodiment 1) disclosed in the present embodiment, comprising:
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes typical case
Correlation analysis method establishes residual generation；It specifically includes:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate in vector y m
Measured value of the variable at the ith sample moment；
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two
The mean value of each variable in data set is 0, obtains new data setWith
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the cross covariance of y
Matrix；
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And
Singular value decomposition, Υ=Γ Σ R are carried out to correlation matrix ΥT, wherein Γ is the left singular vector of the correlation matrix, and R is institute
The right singular vector of correlation matrix is stated, Σ is diagonal matrix；Then the first residual error is obtained using canonical correlation analysis technology to occur
Device ra:
ra=LTy-ΣTJTu
Wherein,
Second step, sampling Monte Carlo method determine the first threshold of first residual generation, specifically include:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and set
Set initial thresholdCurrent iteration number s=0；
Step B2, according to Monte Carlo number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value
Are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate that statistic is corresponding when iteration s
Present threshold value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, raFor the residual signals of first residual generation output；For first residual generation
Statistic；For the inverse matrix of the covariance matrix of corresponding residual signals；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf
It is invalid, continue iteration, enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε >
0 is a constant, and Δ is iteration tolerance；
Third step carries out real-time online detection to industrial process data, specifically includes:
The vector u and the real-time variable data of vector y are acquired, and phase is obtained according to first residual generation
Residual signals answered, and calculate the statistic of corresponding residual signals and be compared with the first threshold, if gained corresponding the
The statistic of one residual generation has been more than the first threshold, then exports fault warning.
Embodiment 3
The present embodiment is similar with above-described embodiment 2, and disclosing a kind of non-gaussian industrial process fault detection method (can refer to
State embodiment 1), comprising:
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes typical case
Correlation analysis method establishes residual generation；It specifically includes:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate in vector y m
Measured value of the variable at the ith sample moment；
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two
The mean value of each variable in data set is 0, obtains new data setWith
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the cross covariance of y
Matrix；
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And it is right
Correlation matrix Υ carries out singular value decomposition, Υ=Γ Σ RT, wherein Γ is the left singular vector of the correlation matrix, and R is described
The right singular vector of correlation matrix, Σ are diagonal matrix；Then the second residual generation is obtained using canonical correlation analysis technology
rb:
rb=JTu-ΣLTy
Wherein,
Second step, sampling Monte Carlo method determine the second threshold of second residual generation；It specifically includes:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and set
Set initial thresholdCurrent iteration number s=0；
Step B2, according to Monte Carlo number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value
Are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate that statistic is corresponding when iteration s
Present threshold value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, rbFor the residual signals of second residual generation output；For second residual generation
Statistic；For the inverse matrix of the covariance matrix of corresponding residual signals；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf
It is invalid, continue iteration, enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε >
0 is a constant, and Δ is iteration tolerance；
Third step carries out real-time online detection to industrial process data, specifically includes:
The variable data of the vector u and the vector y that acquire in real time are obtained, and according to second residual generation
Corresponding residual signals are obtained, and calculates the statistic of corresponding residual signals and is compared with the second threshold, if gained
The statistic of corresponding second residual generation has been more than the second threshold, then exports fault warning.
To sum up, non-gaussian industrial process fault detection method disclosed in the present embodiment, has the advantages that
1, present invention firstly provides a kind of non-height combined based on canonical correlation analysis and Monte Carlo threshold learning
This process online test method, this method can be with on-line checking non-gaussian industrial process；
2, detection method proposed by the present invention is in real time, when industrial process breaks down, can timely to detect
Come；
3, the present invention is based on process datas, convenient and practical independent of accurate process model, and matched system
Letter is realized in (developing the known technology that corollary system is those skilled in the art according to the above method, this will not be repeated here) exploitation
List, system run all right are reliable.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (6)
1. a kind of non-gaussian industrial process fault detection method characterized by comprising
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes canonical correlation
Analysis method establishes residual generation；It specifically includes:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate m in vector y
Measured value of a variable at the ith sample moment；
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two data
The mean value for each variable concentrated is 0, obtains new data setWith
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the Cross-covariance of y；
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And to correlation
Matrix Υ carries out singular value decomposition, Υ=Γ Σ RT, wherein Γ is the left singular vector of the correlation matrix, and R is the correlation
The right singular vector of matrix, Σ are diagonal matrix；Then the first residual generation is obtained using canonical correlation analysis technology to export
Residual signals raWith the residual signals r of the second residual generation outputb:
ra=LTy-ΣTJTu,rb=JTu-ΣLTy
Wherein,
Second step, sampling Monte Carlo method determine that the first threshold of first residual generation and the second residual error occur respectively
The second threshold of device；It specifically includes:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and be arranged just
Beginning threshold valueCurrent iteration number s=0；
Step B2, according to Monte-Carlo Simulation number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate that statistic is corresponding current when iteration s
Threshold value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, raAnd rbThe residual signals of respectively the first residual generation and the output of the second residual generation；WithPoint
It Wei not the first residual generation and the corresponding statistic of the second residual generation；WithThe association of respectively corresponding residual signals
The inverse matrix of variance matrix；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf not at
It is vertical then continue iteration, it enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε > 0 is
One constant, Δ are iteration tolerance；
Third step carries out real-time online detection to industrial process data, specifically includes:
The variable data of the vector u and the vector y that acquire in real time are obtained, and according to first residual generation and institute
State the second residual generation and obtain corresponding residual signals, and calculate separately corresponding residual signals statistic and with corresponding institute
It states first threshold and the second threshold is compared, if the statistic of corresponding first residual generation of gained has been more than described the
One threshold value or the statistic of corresponding second residual generation have been more than the second threshold, then export fault warning.
2. a kind of for executing the non-gaussian industrial process fault detection system of method as described in claim 1.
3. a kind of non-gaussian industrial process fault detection method characterized by comprising
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes canonical correlation
Analysis method establishes residual generation；It specifically includes:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate m in vector y
Measured value of a variable at the ith sample moment；
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two data
The mean value for each variable concentrated is 0, obtains new data setWith
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the Cross-covariance of y；
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And to correlation
Matrix Υ carries out singular value decomposition, Υ=Γ Σ RT, wherein Γ is the left singular vector of the correlation matrix, and R is the correlation
The right singular vector of matrix, Σ are diagonal matrix；Then the first residual generation is obtained using canonical correlation analysis technology to export
Residual signals ra:
ra=LTy-ΣTJTu
Wherein,
Second step, sampling Monte Carlo method determine the first threshold of first residual generation, specifically include:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and be arranged just
Beginning threshold valueCurrent iteration number s=0；
Step B2, according to Monte-Carlo Simulation number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate statistic corresponding current threshold when iteration s
Value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, raFor the residual signals of first residual generation output；For the statistics of first residual generation
Amount；For the inverse matrix of the covariance matrix of corresponding residual signals；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf not at
It is vertical then continue iteration, it enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε > 0 is
One constant, Δ are iteration tolerance；
Third step carries out real-time online detection to industrial process data, specifically includes:
The vector u and the real-time variable data of vector y are acquired, and is obtained accordingly according to first residual generation
Residual signals, and calculate the statistic of corresponding residual signals and be compared with the first threshold, if gained corresponding first is residual
The statistic of difference generator has been more than the first threshold, then exports fault warning.
4. a kind of for executing the non-gaussian industrial process fault detection system of method as claimed in claim 3.
5. a kind of non-gaussian industrial process fault detection method characterized by comprising
The first step chooses a certain number of normal history data sets, carries out mean value to data set and pre-processes, utilizes canonical correlation
Analysis method establishes residual generation；It specifically includes:
Step A1, the process data under accidental conditions is chosen using historical data base, used respectivelyTwo variable datas of expression process
Collection, wherein l is measurand number in vector u, and m is measurand number in vector y, and N is independent sample points；Then u (i), i=
1 ..., N indicate that measured value of the l variable at the ith sample moment in vector u, y (i), i=1 ..., N indicate m in vector y
Measured value of a variable at the ith sample moment；
Step A2, average value processing is carried out to data set；Average value processing is carried out respectively to two datasets, so that two data
The mean value for each variable concentrated is 0, obtains new data setWith
Step A3, after data set being normalized, estimate required covariance matrix:
Wherein, Σu, ΣyAnd ΣuyRespectively indicate the covariance matrix of u, the covariance matrix and u of y and the Cross-covariance of y；
Step A4, the covariance matrix obtained according to estimated by establishes correlation matrix Υ are as follows:And to correlation
Matrix Υ carries out singular value decomposition, Υ=Γ Σ RT, wherein Γ is the left singular vector of the correlation matrix, and R is the correlation
The right singular vector of matrix, Σ are diagonal matrix；Then the second residual generation is obtained using canonical correlation analysis technology to export
Residual signals rb:
rb=JTu-ΣLTy
Wherein,
Second step, sampling Monte Carlo method determine the second threshold of second residual generation；It specifically includes:
Step B1, the sufficiently big normal history data set of number of samples is chosen, determines Monte-Carlo Simulation number K, and be arranged just
Beginning threshold valueCurrent iteration number s=0；
Step B2, according to Monte-Carlo Simulation number, K data set is chosen in the s times iteration
Step B3, the statistic of corresponding residual generation is expressed as Js, counting function is expressed asCounting function value are as follows:
Wherein,Indicate the statistics magnitude of k-th of sample moment when iteration s,Indicate statistic corresponding current threshold when iteration s
Value,Indicate the count value of k-th of sample moment when iteration s；The calculating of the statistic are as follows:
Wherein, rbFor the residual signals of second residual generation output；For the statistics of second residual generation
Amount；For the inverse matrix of the covariance matrix of corresponding residual signals；
Step B4, after for K Monte-Carlo Simulation, estimate rate of false alarm value:
If step B5,Establishment then returns to present threshold valueFor required threshold value, and enableIf invalid
Then continue iteration, enablesAnd return step B3；Wherein, pFARFor specified maximum allowable rate of false alarm, ε > 0 is one
Constant, Δ are iteration tolerance；
Third step carries out real-time online detection to industrial process data, specifically includes:
The variable data of the vector u and the vector y that acquire in real time are obtained, and is obtained according to second residual generation
Corresponding residual signals, and calculate the statistic of corresponding residual signals and be compared with the second threshold, if gained is corresponding
The statistic of second residual generation has been more than the second threshold, then exports fault warning.
6. a kind of for executing the non-gaussian industrial process fault detection system of method as claimed in claim 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710623082.9A CN107357275B (en) | 2017-07-27 | 2017-07-27 | Non-gaussian industrial process fault detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710623082.9A CN107357275B (en) | 2017-07-27 | 2017-07-27 | Non-gaussian industrial process fault detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107357275A CN107357275A (en) | 2017-11-17 |
| CN107357275B true CN107357275B (en) | 2019-08-27 |
Family
ID=60286354
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710623082.9A Active CN107357275B (en) | 2017-07-27 | 2017-07-27 | Non-gaussian industrial process fault detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107357275B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108170648B (en) * | 2017-12-15 | 2021-05-18 | 宁波大学 | non-Gaussian process monitoring method based on known data regression |
| CN110046054A (en) * | 2018-01-17 | 2019-07-23 | 中兴通讯股份有限公司 | Virtual machine method for detecting abnormality, device, equipment and computer readable storage medium |
| CN110580488B (en) * | 2018-06-08 | 2022-04-01 | 中南大学 | Multi-working-condition industrial monitoring method, device, equipment and medium based on dictionary learning |
| CN108803465B (en) * | 2018-06-19 | 2019-08-23 | 哈尔滨工业大学 | A kind of distributed plug and play fault monitoring method based on closed-loop data driving |
| CN108646719B (en) * | 2018-07-05 | 2021-04-06 | 中南大学 | Weak fault detection method and system |
| CN109203995A (en) * | 2018-07-10 | 2019-01-15 | 中南大学 | A kind of trailer system major loop ground fault diagnosis method and system |
| CN109507972B (en) * | 2018-12-19 | 2020-06-16 | 中国计量大学 | Industrial production process fault monitoring method based on layered non-Gaussian monitoring algorithm |
| CN109523195B (en) * | 2018-12-19 | 2023-01-06 | 中国计量大学 | Sample multi-modal classification-based fault detection method for hierarchical non-Gaussian algorithm |
| CN110209145B (en) * | 2019-05-16 | 2020-09-11 | 浙江大学 | Carbon dioxide absorption tower fault diagnosis method based on nuclear matrix approximation |
| CN110187206B (en) * | 2019-05-22 | 2021-11-02 | 中国人民解放军国防科技大学 | Fault detection method for suspension system in non-Gaussian process under complex working condition |
| CN110879372B (en) * | 2019-12-03 | 2021-04-23 | 中南大学 | Traction system main loop earth fault diagnosis method and system based on feature correlation |
| CN111209973B (en) * | 2020-01-09 | 2023-04-07 | 浙江理工大学 | Process monitoring method based on mixed kernel PCA-CCA and kernel density estimation |
| CN113189968B (en) * | 2021-05-08 | 2022-08-26 | 哈尔滨工业大学 | Distributed fault diagnosis method for interconnected industrial process |
| CN114545907B (en) * | 2022-03-15 | 2023-12-19 | 中南大学 | Fault detection method of flight control system based on filter |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105739489A (en) * | 2016-05-12 | 2016-07-06 | 电子科技大学 | Batch process fault detecting method based on ICA-KNN |
| CN106054859A (en) * | 2016-05-30 | 2016-10-26 | 宁波大学 | Double-layer integrated type industrial process fault detection method based on modified independent component analysis (MICA) |
| CN106092625A (en) * | 2016-05-30 | 2016-11-09 | 宁波大学 | The industrial process fault detection method merged based on correction type independent component analysis and Bayesian probability |
| CN106569487A (en) * | 2016-11-16 | 2017-04-19 | 北京理工大学 | Fault detection method for servo system based on observer |
-
2017
- 2017-07-27 CN CN201710623082.9A patent/CN107357275B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105739489A (en) * | 2016-05-12 | 2016-07-06 | 电子科技大学 | Batch process fault detecting method based on ICA-KNN |
| CN106054859A (en) * | 2016-05-30 | 2016-10-26 | 宁波大学 | Double-layer integrated type industrial process fault detection method based on modified independent component analysis (MICA) |
| CN106092625A (en) * | 2016-05-30 | 2016-11-09 | 宁波大学 | The industrial process fault detection method merged based on correction type independent component analysis and Bayesian probability |
| CN106569487A (en) * | 2016-11-16 | 2017-04-19 | 北京理工大学 | Fault detection method for servo system based on observer |
Non-Patent Citations (3)
| Title |
|---|
| 一种基于改进KICA的非高斯过程故障检测方法;蔡连芳等;《化工学报》;20120930;第63卷(第9期);第2864-2868页 * |
| 基于ALSICA算法的多工况非高斯过程故障检测方法;徐莹等;《计算机与应用化学》;20141228;第31卷(第12期);第1522-1526页 * |
| 基于FRICA算法的非高斯过程故障检测方法;徐莹等;《上海应用技术学院学报（自然科学版）》;20150630;第15卷(第2期);第153-158页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107357275A (en) | 2017-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107357275B (en) | Non-gaussian industrial process fault detection method and system | |
| CN105700518B (en) | A kind of industrial process method for diagnosing faults | |
| CN107016236B (en) | Power grid false data injection attack detection method based on nonlinear measurement equation | |
| CN110738274A (en) | nuclear power device fault diagnosis method based on data driving | |
| CN109613428A (en) | It is a kind of can be as system and its application in motor device fault detection method | |
| CN109085805B (en) | Industrial process fault detection method based on multi-sampling-rate factor analysis model | |
| CN106092625B (en) | The industrial process fault detection method merged based on amendment type independent component analysis and Bayesian probability | |
| CN108549908B (en) | Chemical process fault detection method based on multi-sampling probability kernel principal component model | |
| CN109522948A (en) | A kind of fault detection method based on orthogonal locality preserving projections | |
| CN101738998B (en) | System and method for monitoring industrial process based on local discriminatory analysis | |
| CN107451392A (en) | A kind of method for predicting residual useful life containing multiple dependent degeneration processes | |
| CN110348150A (en) | A kind of fault detection method based on dependent probability model | |
| CN110231447A (en) | The method, apparatus and terminal device of water quality abnormality detection | |
| CN109240274A (en) | A kind of industrial process method for diagnosing faults based on higher order dependencies | |
| CN109409425A (en) | A kind of fault type recognition method based on neighbour's constituent analysis | |
| CN106599367A (en) | Method for detecting abnormal state of spacecraft | |
| CN111122811A (en) | Sewage treatment process fault monitoring method of OICA and RNN fusion model | |
| CN109840671A (en) | Operational development effect calculates equipment, operational development effect calculation method and recording medium | |
| CN112766301B (en) | Oil extraction machine indicator diagram similarity judging method | |
| CN108153267A (en) | A kind of Industrial Process Monitoring method based on error Principal Component Analysis Model | |
| EP3506232A1 (en) | Method for learning latest data considering external influences in early warning system and system for same | |
| CN110209145B (en) | Carbon dioxide absorption tower fault diagnosis method based on nuclear matrix approximation | |
| CN110119579B (en) | OICA-based complex industrial process fault monitoring method | |
| CN113721182B (en) | Method and system for evaluating reliability of online performance monitoring result of power transformer | |
| CN117630797A (en) | Ammeter health state detection method, system and storage medium based on working current |
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 |