CN101907681B - Analog circuit dynamic online failure diagnosing method based on GSD-SVDD - Google Patents
Analog circuit dynamic online failure diagnosing method based on GSD-SVDD Download PDFInfo
- Publication number
- CN101907681B CN101907681B CN2010102286876A CN201010228687A CN101907681B CN 101907681 B CN101907681 B CN 101907681B CN 2010102286876 A CN2010102286876 A CN 2010102286876A CN 201010228687 A CN201010228687 A CN 201010228687A CN 101907681 B CN101907681 B CN 101907681B
- Authority
- CN
- China
- Prior art keywords
- sample
- fault
- suprasphere
- test
- diagnostic model
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000012360 testing method Methods 0.000 claims abstract description 78
- 238000003745 diagnosis Methods 0.000 claims abstract description 49
- 230000008569 process Effects 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims description 60
- 239000011159 matrix material Substances 0.000 claims description 16
- 230000003278 mimic effect Effects 0.000 claims description 16
- 238000000926 separation method Methods 0.000 claims description 13
- 230000009467 reduction Effects 0.000 claims description 12
- 238000002955 isolation Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 238000004088 simulation Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000003064 k means clustering Methods 0.000 claims description 3
- 239000011692 calcium ascorbate Substances 0.000 claims description 2
- 239000012141 concentrate Substances 0.000 claims description 2
- 238000005086 pumping Methods 0.000 claims description 2
- PPASLZSBLFJQEF-RKJRWTFHSA-M sodium ascorbate Substances [Na+].OC[C@@H](O)[C@H]1OC(=O)C(O)=C1[O-] PPASLZSBLFJQEF-RKJRWTFHSA-M 0.000 claims description 2
- 230000003595 spectral effect Effects 0.000 claims description 2
- 238000013459 approach Methods 0.000 abstract description 2
- 238000000513 principal component analysis Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 230000009466 transformation Effects 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000002405 diagnostic procedure Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004092 self-diagnosis Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Landscapes
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The invention discloses an analog circuit dynamic online failure diagnosing method based on GSD-SVDD, belonging to the technical field of analog circuit failure diagnosis. In an offline test process, a KFCM algorithm is adopted to calculate a failure resolution value of each testable node and an optimal test node set is selected according to the failure resolution value. In an online diagnosis process, a failure diagnosis model is established by adopting an SVDD single classification approach based on a map spatial distance positive and negative sample weighting, test samples are diagnosed by a layered diagnosis method, and a failure class library and the diagnosis model are renewed dynamically. The method effectively reduces the drill and online diagnosis time of the diagnosis model, guarantees the real-time property of the online diagnosis and improves the precision of the failure diagnosis and can dynamically renew parameters of the diagnosis model so as to enable the a diagnosis system to have the self-adaption capability.
Description
Technical field
The present invention relates to a kind of analog-circuit fault diagnosis method, especially a kind of mimic channel on-line fault diagnosis method based on GSD_SVDD.
Background technology
The development trend of modern electronic equipment self-test, self diagnosis, selfreparing has proposed new requirement to analog circuit fault diagnosing; In case certain partial circuit breaks down; Requirement can realize circuit online in real time testing and diagnosing; Under the operate as normal that does not influence circuit, accomplish the isolation location of fault and repair also input coefficient use again.Although two more than ten years of past people have obtained many achievements in that the off-line circuit fault diagnosis is technical, yet online in real time electronic failure Studies on Diagnosis is still immature.
Compare with the off-line simulation circuit fault diagnosis; On-line fault diagnosis has faced more difficulty, not only need overcome extensively exist in the mimic channel non-linear, and fault type is many; There are problems such as tolerance in components and parts, also need satisfy following 2 requirements: (1) real-time.On-line fault diagnosis requires higher to the real-time of system; In case certain partial circuit breaks down; Require diagnostic system can diagnose out the result immediately; So that the system failure is in time repaired, this all requires the speed of data acquisition in the on-line fault diagnosis, data pre-service, data diagnosis faster, can not influence the normal operation of circuit; (2) adaptivity.After breaking down, wants on-line system in time the failure message of diagnosing to be fed back to diagnostic system; The various parameters of real time modifying diagnostic system, the adaptivity of increase system is in the electronic equipment on-line operation process; Because influences such as facility environment, components and parts tolerance; Equipment state constantly changes, and therefore requires the various parameters of diagnostic system to bring in constant renewal in time, adapts to the new requirement of electronic equipment environment.
At present, the artificial intelligence method based on signal Processing is the focus of mimic channel inline diagnosis method research.In conjunction with existing document and patented technology; In the mimic channel radiodiagnosis x process, the information processing technology commonly used has Fourier transform, wavelet transformation; Fractional order signal Processing etc., and artificial intelligence approach commonly used has neural network method, rough set method, SVMs method, Intelligent Fusion method etc.Though the artificial intelligence diagnosis's method diagnosis efficiency based on signal Processing is high; Solved the problem such as ambiguity and uncertainty of fault diagnosis; But still have following deficiency: the signal of (1) online acquisition receives the influence of electronic component tolerance and outside noise; Cause the signal processing time process long, can't satisfy the requirement of inline diagnosis real-time, and the feature samples separability after the signal Processing is not high; (2) the artificial intelligence method need be set up the intelligent diagnostics model usually in advance, and the diagnostic model training time is long to be required to contradict with real-time.(3) in-circuit diagnostic system requires to have adaptivity, and the adaptive ability of intelligence system itself and learning ability have been limited to the accuracy and the real-time of fault diagnosis.(4) diagnostic accuracy that obtains of existing mimic channel on-line fault diagnosis method is undesirable, and misdiagnosis rate is higher.
Summary of the invention
The object of the present invention is to provide a kind of can solve the diagnostic model training time that exists in the existing mimic channel online diagnosing technique of support shaft longly require with real-time to contradict, adaptive ability is strong and the mimic channel on-line fault diagnosis method of the problem that misdiagnosis rate is high.
Thinking of the present invention is through adopting a kind of improved SVDD sorting technique; Promptly the single type of sorting technique of SVDD based on the positive and negative sample weighting of collection of illustrative plates space length (is called for short GSD-SVDD; Below all with); Be used for analog circuit fault diagnosing, longly require conflicting problem with real-time to solve the existing method diagnostic model training time.
The basic thought of SVDD is through in feature space, finding out a suprasphere that surrounds target sample point and let target sample point all be enclosed in the suprasphere through minimizing the volume that suprasphere surrounds (or many as far as possible), but not target sample point is not comprised in the suprasphere as much as possible.SVDD is a kind of single type of disaggregated model, owing to only need one type of training sample, when having higher diagnosis effect, the training speed of diagnostic model is fast, and more traditional artificial intelligence method more can adapt to the real-time of on-line fault diagnosis.But; Traditional SVDD is very sensitive to noise and singular point in training process; When containing some noises or wild value sample in the training sample; These samples that contain " unusually " information usually are positioned near the suprasphere in feature space, and the suprasphere that causes obtaining is not real optimum suprasphere.GSD-SVDD method of the present invention is that positive negative sample carries out weighting according to the collection of illustrative plates space length; Training sample is divided into two types of positive negative samples; Each sample point distributes a fuzzy membership coefficient; Because each sample point has different penalty coefficients, make each sample point different to describing the edge effect ability, can reduce the influence of noise and singular point through the weights of reduction noise spot and singular point; To obtain the description suprasphere that robustness is better, precision is higher, can effectively improve the generalization ability and the diagnostic accuracy of traditional SVDD diagnostic model.
Particularly, the present invention reaches the object of the invention through following technical scheme:
Mimic channel dynamic online failure diagnosing method based on GSD_SVDD is characterized in that, comprises following each step:
A, the optimum test node set of selection from circuit under test;
B, gather the normal sample and the fault sample of circuit under test, the sample of gathering is carried out the pre-service of feature extraction and dimensionality reduction, obtain training normal sample set and fault sample collection through the test node selected in the steps A;
C, normal sample set of training and fault sample collection that step B is obtained use the GSD_SVDD method to train respectively, obtain normal type diagnostic model and failure classes diagnostic model; Said GSD_SVDD method is a kind of single type of sorting technique of SVDD based on the positive and negative sample weighting of collection of illustrative plates space length; This method is according to the collection of illustrative plates space length of training sample; Training sample is divided into positive and negative two types; The collection of illustrative plates space length of training sample is carried out weighting as weights to this training sample, and obtain an optimal spatial suprasphere through finding the solution minimum quadratic programming target training, comprise positive sample in this suprasphere by the positive and negative samples of weighting; And negative sample is positioned at outside the suprasphere, and each sample is endowed different punishment degree according to collection of illustrative plates space length difference during training;
D, the test sample book of gathering the circuit under test on-line operation are carried out the pre-service of filtering, feature extraction and dimensionality reduction;
E, the test sample book that step D is gathered are carried out the layering fault diagnosis: at first, judge whether it is fault with a normal class diagnostic model, if, then adopt failure classes diagnostic model fault location classification, and more new samples storehouse and diagnostic model.
In the technique scheme; Choose optimum test node set and can use traditional optimization method; For example Sensitirity va1ue method and intelligent algorithm optimizing method, but sensitivity method often need be listed circuit equation, not strong to the applicability of large-scale circuit and non-linear circuit; And the intelligent algorithm optimizing method objective function of optimizing need be set, and the optimizing time is often long.Therefore the present invention adopts fault separation value to carry out the selection of optimum test node; The fault separation value of the fault sample of promptly under this node, gathering according to the KFCM algorithm computation; Select optimum test node set according to the size of degree of separation value, this method need not write out the modal equation of circuit, is fit to the linear circuit and the non-linear circuit of random scale; This method has reflected the fault sample separation degree of test node, helps improving the precision of diagnostic system.
In addition; When test sample book is carried out the layering fault diagnosis; Whether the present invention judges greater than this suprasphere radius whether test sample book belongs to such according to test sample book and normal sample class and fault sample class suprasphere centre of sphere distance; But in the reality diagnosis, can exist a plurality of failure classes supraspheres to satisfy above-mentioned condition, thereby can't judge the affiliated classification of test sample book.For addressing this problem, the present invention adopts Bayes decision rule to judge the failure classes that test sample book is affiliated, and its discriminant function is:
Wherein, N
iBe i class fault sample number in the training sample, i=1 ..., c, c are fault sample class number, N is all training sample sums, r
iBe the radius of i class GSD-SVDD diagnostic model suprasphere, d
i(z) be the distance of the test sample book z distance i class diagnostic model suprasphere centre of sphere.
In sum, the present invention uses KFCM algorithm computation circuit in off-line test each can survey the fault separation value of node, selects optimum test node set according to the size of degree of separation value; In the inline diagnosis process; Employing is carried out the structure of fault diagnosis model based on single type of sorting technique of SVDD (GSD-SVDD) of the positive and negative sample weighting of collection of illustrative plates space length, and adopts the layering diagnostic method test sample book to be diagnosed and dynamically updated fault class libraries and diagnostic model.Compare prior art, the inventive method has the following advantages:
(1) the present invention selects optimum test node set according to the fault separation value that can survey node in the circuit; Need not write out modal equation; The linear circuit and the non-linear circuit that are fit to random scale have reflected the fault sample separation degree of test node, have improved the precision of diagnostic system.
(2) the present invention adopts the single type of sorting technique of SVDD based on the positive and negative sample weighting of collection of illustrative plates space length to make up diagnostic model; Both had that traditional SVDD diagnostic method model training speed is fast, the good advantage of diagnosis real-time; Reduce simultaneously the influence of noise and singular point again, improved the generalization ability and the diagnostic accuracy of diagnostic model.
(3) the present invention adopts the layering diagnostic method to diagnose online new samples, only needs to calculate the distance of the test sample book and the normal sample spherical ball heart, has practiced thrift the time of judging test sample book and other failure classes relation, has improved the diagnosis efficiency of circuit normal condition greatly.
(4) the present invention adopts GSD-SVDD dynamic diagnosis model; This diagnostic model has the ability of the new fault category of identification; The dynamic simultaneously parameter of revising diagnostic model; Guaranteed that diagnostic system has adaptive ability and learning ability, selectively upgraded training sample set and sample diagnostic model, the time of having reduced inline diagnosis.
Description of drawings
Fig. 1 is a FB(flow block) of the present invention;
Fig. 2 is GSD SVDD sorter model figure of the present invention;
Fig. 3 is the process flow diagram that dynamically updates GSD SVDD diagnostic model in the inventive method.
Embodiment
Below in conjunction with accompanying drawing technical scheme of the present invention is elaborated:
Shown in accompanying drawing 1, mimic channel dynamic online failure diagnosing method of the present invention specifically comprises following each step:
A, the optimum test node set of selection from circuit under test;
The present invention selects optimum test node set according to the fault separation value that can survey node in the circuit, specifically comprises following each step:
A1, adopt the running status of emulation tool simulation circuit under test, and add pumping signal identical when working with circuit under test at circuit input end;
Adopt the running status of Pspice emulation tool simulation circuit under test in this embodiment.
A2, gather the fault signature sample, calculate the degree of fault isolation of each fault signature sample set; Specifically according to following each step:
A201, can survey the node place at each respectively and gather the fault signature sample, obtain and can survey the fault signature sample set that node is counted same number, and each can survey the respectively corresponding fault signature sample set of node;
The cluster centre v of each fault sample class in A202, the use KFCM algorithm computation fault signature sample set
i, i=1 ..., c, c are the number of fault sample class; Wherein, the objective function of KFCM algorithm is:
Constraint condition is:
(k=1; 2 ... n)
In the formula, U=[u
Ik] be to be subordinate to matrix, V=[v
i] be the cluster centre matrix, u
IkBetween (0,1), u
IkRepresent k data points Φ (x
k) degree of type of belonging to i, and m ∈ [1, ∞) be a weighted index, v
iBe the cluster centre in the input space, i=1 ..., c, n are the data sample numbers;
A203, calculate the degree of fault isolation of each fault signature sample set according to following formula:
Wherein, J
i(x) be the degree of fault isolation that can survey the fault signature sample set x that gathers under the node at i, i=1 ..., c;
A3, with the fault separation value that calculates according to sorting from big to small, choose preceding ξ
2The pairing node of surveying of individual fault separation value is as optimum test node, ξ
2Integer for 1-10.
The optimum test node number of choosing in this step can be chosen according to actual needs, and the common situation when this embodiment is considered actual diagnosis is defined as 1-10 with this number; Certainly, according to actual needs, also can choose the optimum test node more than 10.
B, gather the normal sample and the fault sample of circuit under test, the sample of gathering is carried out the pre-service of feature extraction and dimensionality reduction, obtain training normal sample set and fault sample collection through the test node selected in the steps A;
In this embodiment, gather sample and sample carried out pre-service and specifically carry out according to following steps:
B1, gather the voltage signal can survey node, and to the data equal interval sampling of voltage signal:
Among the Pspice during image data; Though SF is provided with equal; But the packing density of still gathering sometimes is inconsistent, therefore in the present invention the data of gathering is carried out equal interval sampling, in this embodiment the voltage signal of gathering is uniformly-spaced gathered 500 data points.
B2, the voltage signal behind the equal interval sampling in B1 step is carried out FFT;
The general wavelet transformation that adopts carries out the data pre-service in analog circuit fault diagnosing, yet wavelet transformation need be according to the feature selecting optimal wavelet conversion number of plies of transform data, requires to contradict with the real-time of on-line fault diagnosis.The present invention selects fast fourier transformation algorithm to carry out the data pre-service, voltage signal is carried out Fast Fourier Transform (FFT) belong to prior art, specifically can repeat no more referring to document (Digital Signal Analysis and Processing, Yan Qingming, 2009, Electronic Industry Press) here.
B3, the data behind the Fourier transform in B2 step are carried out principal component analysis (PCA) PCA dimensionality reduction;
The characteristic dimension that is obtained by step B2 is still very high, and the training speed of the diagnostic model that can slow down is very high to the real-time requirement in the mimic channel on-line fault diagnosis, therefore, is necessary the data after the data are carried out dimensionality reduction.The principal ingredient that the present invention adopts the PCA method to extract characteristic is carried out dimension-reduction treatment, and concrete PCA analytical procedure is following:
1) the covariance matrix E of computational data collection sample x;
2) eigenwert [e of the proper vector of calculating covariance matrix E
1, e
2... e
h], h is the dimension of data set, the characteristic value of levying by big to little ordering [e '
1, e '
2... e '
h];
3) the data set sample is projected among the space that eigenvector opens.
For PCA, the dimension of confirming dimensionality reduction is a problem of facing a difficult choice.If dimension r is too small, then the dimension of data is low, is convenient to analyze, and has also reduced noise simultaneously, but possibly lose some Useful Informations.The present invention is according to contribution rate of accumulative total (Accumulated Contribution Rate, ACR) the next principal character dimension of confirming extraction of main composition preceding r in the PCA transformation matrix.ACR is calculated by following formula:
Wherein, e '
iBe the eigenwert of the proper vector after the ordering, r is the dimension of dimensionality reduction variable.
B4, the data that B3 step is obtained are carried out normalization and are handled, and dwindle the difference between the relative size between the data, and the training sample that obtains normal sample and fault sample is gathered, and wherein the normalized formula of data sequence X is following:
Wherein, X (i) is i data in the data sequence, X
MinAnd X
MaxBe respectively data minimum with maximum in the data sequence.
C, normal sample set of training and fault sample collection that step B is obtained use the GSD_SVDD method to train respectively, obtain normal type diagnostic model and failure classes diagnostic model; Said GSD_SVDD method is a kind of single type of sorting technique of SVDD based on the positive and negative sample weighting of collection of illustrative plates space length; Shown in accompanying drawing 2; This method is divided into positive and negative two types according to the collection of illustrative plates space length of training sample with training sample, and the collection of illustrative plates space length of training sample is carried out weighting as weights to this training sample; And obtain an optimal spatial suprasphere through finding the solution minimum quadratic programming target training by the positive and negative samples of weighting; Comprise positive sample in this suprasphere, and negative sample is positioned at outside the suprasphere, each sample is endowed different punishment degree according to collection of illustrative plates space length difference during training; Specifically comprise following each step:
C1, training sample x is mapped to Laplce's spectral space of figure;
C2, the cluster centre that adopts every type of fault sample of k-means clustering algorithm calculating and sample are to the collection of illustrative plates space length of cluster centre, as the weights of each training sample; This step is specifically carried out according to following each step:
C201, use Gaussian function structure similar matrix A ∈ R
Txt, matrix element A
Ij=exp (|| S
i-S
j||
2/ 2 σ
2), and when i=j, A
Ii=0;
C202, structure degree matrix D, (i i) is the i row element sum of similar matrix A to the element D on the degree matrix principal diagonal, and other elements are 0.Structure Laplce matrix,
C203, Laplce's matrix L is carried out characteristic value decomposition, h pairing proper vector x of eigenvalue of maximum before finding out
1, x
2... x
h, structural matrix X=[x
1, x
2..., x
h] ∈ R
N * h, wherein feature vector, X is by the row storage; C204, the capable vector of X is carried out normalization, remembers that normalized matrix is Y,
C205, regard each row among the Y as space R
hIn sample.The luv space data are promptly represented by the collection of illustrative plates characteristic vector data;
C206, Y is carried out k-mean cluster (it is a type that every type of fault sample gathers), obtain cluster centre, calculate among the Y each sample to distances of clustering centers { d
1, d
2... d
h, i.e. the collection of illustrative plates space length of sample.
The k-means clustering algorithm that relates among the present invention is a prior art, but detailed content list of references (data mining and Knowledge Discovery, Li Xiongfei, Li Jun, 2003, Higher Education Publishing House).
C3, threshold xi is set
1, the collection of illustrative plates space length is a negative sample greater than the sample of threshold value, is positive sample, wherein threshold xi less than the sample of threshold value
1Set basis negative sample number can not surpass 10% rule of positive sample number;
C4, training suprasphere obtain the GSD_SVDD diagnostic model; Wherein the GSD_SVDD diagnostic model obtains by finding the solution following optimum quadratic programming:
Wherein, a is the centre of sphere of being trained the suprasphere that obtains by training sample, and R is the radius of sphericity of this suprasphere, C
i, C
iBe respectively the punishment constant of positive negative sample, ξ
i, ξ
lBe respectively the slack variable of positive negative sample, m
i, m
iBe respectively the weights of positive negative sample, be the collection of illustrative plates space length of sample, N
i, N
lBe respectively positive and negative sample number, their sums are total sample number, and wherein positive sample set is given label y
i=1, the negative sample collection is given label y
l=-1.
D, the test sample book of gathering the circuit under test on-line operation are carried out the pre-service of filtering, feature extraction and dimensionality reduction;
Circuit for on-line operation; The data of online acquisition can receive noise; At first need denoising; Consider the real-time requirement of inline diagnosis, still adopt fast fourier transformation algorithm to carry out denoising, fault signature extracts consistent with step described in the step B with the preprocessing process of dimensionality reduction thereafter.
E, the test sample book that step D is gathered are carried out the layering fault diagnosis: at first, judge whether it is fault with a normal class diagnostic model, if, then adopt failure classes diagnostic model fault location classification, and more new samples storehouse and diagnostic model.
This step specifically comprises following each step:
E1, according to the judgment criterion between test sample book and the diagnostic model suprasphere, judge whether it is fault with a normal type of diagnostic model, if failure classes, then execution in step E2; Otherwise, execution in step E3:
E2, according to the judgment criterion between test sample book and the diagnostic model suprasphere, with the fault category of failure classes diagnostic model fault location sample;
As have only a fault diagnosis model suprasphere to satisfy condition, execution in step E3 then; If when having a plurality of fault diagnosis model supraspheres to satisfy, adopt the affiliated failure classes of Bayes decision rule judgement sample z, execution in step E3 then, wherein the Bayes decision rule discriminant function is:
Wherein, N
iBe i class fault sample number in the training sample, i=1 ..., c, c are fault sample class number, N is all training sample sums, r
iBe the radius of i class GSDSVDD diagnostic model suprasphere, d
i(z) be the distance of the test sample book z distance i class diagnostic model suprasphere centre of sphere.
E3, renewal training sample database and diagnostic model, shown in accompanying drawing 3, its step is carried out according to following process:
E301, calculate test sample book z to the centre of sphere of t class suprasphere apart from d
i(z), wherein t judges the classification that test sample book is affiliated among step e 1 or the E2;
E302, judging distance d
t(z) whether more than or equal to the radius of t class suprasphere, if less than, then test sample book z is the non-support vector of such suprasphere, does not influence the decision boundary of spheroid, need not upgrade j class training sample database and train diagnostic model again; If more than or equal to, then test sample book z is the support vector of t class suprasphere, is positioned at outside suprasphere border or the suprasphere, influences the decision boundary of suprasphere, execution in step E303;
E303, add test sample book z to t class training sample and concentrate, train suprasphere again, upgrade the diagnostic model of t class.
Judgment criterion between above-mentioned test sample book and the diagnostic model suprasphere, specifically carry out according to following method:
Judge whether test sample book z belongs to such diagnostic model and only need judge between test sample book z and the diagnostic model suprasphere whether meet the following conditions, as satisfy this condition, judge that then it belongs to such; Otherwise, do not belong to such:
Wherein, K is a kernel function; X is the training sample of suprasphere, i, j=1....n; N is a number of training,
Work as α
iBe called support vector at＞0 o'clock, R is the radius of this suprasphere.
Claims (5)
1. based on the mimic channel dynamic online failure diagnosing method of GSD_SVDD, it is characterized in that, comprise following each step:
A, the optimum test node set of selection from circuit under test;
B, gather the normal sample and the fault sample of circuit under test, the sample of gathering is carried out the pre-service of feature extraction and dimensionality reduction, obtain training normal sample set and fault sample collection through the test node selected in the steps A;
C, normal sample set of training and fault sample collection that step B is obtained use the GSD_SVDD method to train respectively, obtain normal type diagnostic model and failure classes diagnostic model; Said GSD_SVDD method is a kind of single type of sorting technique of SVDD based on the positive and negative sample weighting of collection of illustrative plates space length; This method is according to the collection of illustrative plates space length of training sample; Training sample is divided into positive and negative two types; The collection of illustrative plates space length of training sample is carried out weighting as weights to this training sample, and obtain an optimal spatial suprasphere through finding the solution minimum quadratic programming target training, comprise positive sample in this suprasphere by the positive and negative samples of weighting; And negative sample is positioned at outside the suprasphere, and each sample is endowed different punishment degree according to collection of illustrative plates space length difference during training;
D, the test sample book of gathering the circuit under test on-line operation are carried out the pre-service of filtering, feature extraction and dimensionality reduction;
E, the test sample book that step D is gathered are carried out the layering fault diagnosis: at first, judge whether it is fault with a normal class diagnostic model, if, then adopt failure classes diagnostic model fault location classification, and more new samples storehouse and diagnostic model.
2. according to claim 1 based on the mimic channel dynamic online failure diagnosing method of GSD_SVDD, it is characterized in that step C specifically comprises following each step:
C1, training sample x is mapped to Laplce's spectral space of figure;
C2, the cluster centre that adopts every type of fault sample of k-means clustering algorithm calculating and sample are to the collection of illustrative plates space length of cluster centre, as the weights of each training sample;
C3, threshold xi is set
1, the collection of illustrative plates space length is a negative sample greater than the sample of threshold value, is positive sample, wherein threshold xi less than the sample of threshold value
1Set basis negative sample number can not surpass 10% rule of positive sample number;
C4, training suprasphere obtain the GSD_SVDD diagnostic model; Wherein the GSD_SVDD diagnostic model obtains by finding the solution following optimum quadratic programming:
Wherein, a is the centre of sphere of being trained the suprasphere that obtains by training sample, and R is the radius of sphericity of this suprasphere, C
i, C
lBe respectively the punishment constant of positive negative sample, ξ
i, ξ
lBe respectively the slack variable of positive negative sample, m
i, m
lBe respectively the weights of positive negative sample, be the collection of illustrative plates space length of sample, N
i, N
lBe respectively positive and negative sample number, their sums are total sample number, and wherein positive sample set is given label y
i=1, the negative sample collection is given label y
l=-1.
3. according to claim 1 based on the mimic channel dynamic online failure diagnosing method of GSD_SVDD, it is characterized in that said step e specifically comprises following each step:
E1, according to the judgment criterion between test sample book and the diagnostic model suprasphere, judge whether it is fault with a normal type of diagnostic model, if failure classes, then execution in step E2; Otherwise, execution in step E3:
E2, according to the judgment criterion between test sample book and the diagnostic model suprasphere, with the fault category of failure classes diagnostic model fault location sample;
As have only a fault diagnosis model suprasphere to satisfy condition, execution in step E3 then; If when having a plurality of fault diagnosis model supraspheres to satisfy, adopt the affiliated failure classes of Bayes decision rule judgement sample z, execution in step E3 then, wherein the Bayes decision rule discriminant function is:
Wherein, N
iBe i class fault sample number in the training sample, i=1 ..., c, c are fault sample class number, N is all training sample sums, r
iBe the radius of i class GSD_SVDD diagnostic model suprasphere, d
i(z) be the distance of the test sample book z distance i class diagnostic model suprasphere centre of sphere;
E3, renewal training sample database and diagnostic model, its step is carried out according to following process:
E301, calculate test sample book z to the centre of sphere of t class suprasphere apart from d
t(z), wherein t judges the classification that test sample book is affiliated among step e 1 or the E2;
E302, judging distance d
t(z) whether more than or equal to the radius of t class suprasphere, if less than, then test sample book z is the non-support vector of such suprasphere, does not influence the decision boundary of spheroid, need not upgrade j class training sample database and train diagnostic model again; If more than or equal to, then test sample book z is the support vector of t class suprasphere, is positioned at outside suprasphere border or the suprasphere, influences the decision boundary of suprasphere, execution in step E303;
E303, add test sample book z to t class training sample and concentrate, train suprasphere again, upgrade the diagnostic model of t class.
4. the mimic channel dynamic online failure diagnosing method based on GSD_SVDD as claimed in claim 3 is characterized in that, the judgment criterion between test sample book described in step e 1 and the E2 and the diagnostic model suprasphere is specifically carried out according to following method:
Judge whether test sample book z belongs to such diagnostic model and only need judge between test sample book z and the diagnostic model suprasphere whether meet the following conditions, as satisfy this condition, judge that then it belongs to such; Otherwise, do not belong to such:
Wherein, K is a kernel function, and x is the training sample of suprasphere, i, and j=1....n, n are number of training,
Work as a
iBe called support vector at＞0 o'clock, R is the radius of this suprasphere.
5. like claim 1,2,3 or 4 said mimic channel dynamic online failure diagnosing methods based on GSD_SVDD; It is characterized in that; From circuit under test, selecting optimum test node set described in the steps A, is to select to obtain through comparing the size that respectively can survey the degree of fault isolation of node in the circuit under test; Specifically carry out according to following each step:
A1, adopt the running status of emulation tool simulation circuit under test, and add pumping signal identical when working with circuit under test at circuit input end;
A2, gather the fault signature sample, calculate the degree of fault isolation of each fault signature sample set; Specifically according to following each step:
A201, can survey the node place at each respectively and gather the fault signature sample, obtain and can survey the fault signature sample set that node is counted same number, and each can survey the respectively corresponding fault signature sample set of node;
The cluster centre v of each fault sample class in A202, the use KFCM algorithm computation fault signature sample set
i, i=1 ..., c, c are the number of fault sample class; Wherein, the objective function of KFCM algorithm is:
Constraint condition is:
(k=1; 2 ... n)
In the formula, U=[u
Ik] be to be subordinate to matrix, V=[v
i] be the cluster centre matrix, u
IkBetween (0,1), u
IkRepresent k data points Φ (x
k) degree of type of belonging to i, and m ∈ [1, ∞) be a weighted index, v
iBe the cluster centre in the input space, i=1 ..., c, n are the data sample numbers;
A203, calculate the degree of fault isolation of each fault signature sample set according to following formula:
Wherein, J
i(x) be the degree of fault isolation that can survey the fault signature sample set x that gathers under the node at i, i=1 ..., c;
A3, with the fault separation value that calculates according to sorting from big to small, choose preceding ξ
2The pairing node of surveying of individual fault separation value is as optimum test node, ξ
2Integer for 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102286876A CN101907681B (en) | 2010-07-15 | 2010-07-15 | Analog circuit dynamic online failure diagnosing method based on GSD-SVDD |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102286876A CN101907681B (en) | 2010-07-15 | 2010-07-15 | Analog circuit dynamic online failure diagnosing method based on GSD-SVDD |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101907681A CN101907681A (en) | 2010-12-08 |
| CN101907681B true CN101907681B (en) | 2012-07-04 |
Family
ID=43263196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010102286876A Expired - Fee Related CN101907681B (en) | 2010-07-15 | 2010-07-15 | Analog circuit dynamic online failure diagnosing method based on GSD-SVDD |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101907681B (en) |
Families Citing this family (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8868359B2 (en) * | 2011-04-29 | 2014-10-21 | General Electric Company | Device and method for detecting and locating defects in underground cables |
| CN102339389B (en) * | 2011-09-14 | 2013-05-29 | 清华大学 | Fault detection method for one-class support vector machine based on density parameter optimization |
| CN102421126A (en) * | 2011-10-31 | 2012-04-18 | 北京邮电大学 | Interruption prewarning method and device in SON (Self Organization Network) system |
| CN102520341B (en) * | 2011-12-05 | 2013-12-18 | 南京航空航天大学 | Analog circuit fault diagnosis method based on Bayes-KFCM (Kernelized Fuzzy C-Means) algorithm |
| CN102749572B (en) * | 2012-06-26 | 2015-03-25 | 南京航空航天大学 | IECMAC parameter identification-based power electronic circuit failure predicting method |
| CN102914737A (en) * | 2012-09-21 | 2013-02-06 | 安徽科技学院 | Network fault diagnosis method for complex circuit |
| US9262726B2 (en) * | 2013-01-17 | 2016-02-16 | Applied Materials, Inc. | Using radial basis function networks and hyper-cubes for excursion classification in semi-conductor processing equipment |
| CN103076556B (en) * | 2013-01-21 | 2013-10-16 | 中国民航大学 | Method for selecting function-maintenance testing points of avionic assembly |
| CN103675525B (en) * | 2013-11-14 | 2017-01-18 | 南京航空航天大学 | DC-DC converter health monitoring and fault prediction method based on multiple SVDD models |
| CN103675544B (en) * | 2013-12-09 | 2016-01-20 | 华南理工大学 | Fault Signal Analyses in HV Transmission based on optimized algorithm detects and method for waveform identification |
| CN103995237A (en) * | 2014-05-09 | 2014-08-20 | 南京航空航天大学 | Satellite power supply system online fault diagnosis method |
| CN103995528B (en) * | 2014-05-09 | 2017-04-19 | 南京航空航天大学 | Intelligent self-repairing technology for main circuit of power converter |
| CN104134073B (en) * | 2014-07-31 | 2017-09-05 | 郑州航空工业管理学院 | One kind is based on the normalized remote sensing image list class sorting technique of a class |
| CN106294038B (en) * | 2015-05-25 | 2019-10-18 | 阿里巴巴集团控股有限公司 | A kind of generation of fault spectrum, detection method and device based on fault spectrum |
| CN105425150B (en) * | 2015-11-09 | 2018-08-31 | 江苏科技大学 | A kind of Method of Motor Fault Diagnosis based on RBF and PCA-SVDD |
| CN105467305B (en) * | 2016-01-13 | 2018-06-26 | 安徽理工大学 | Buck-Boost faults feature extracting method under a kind of variable working condition |
| CN107102630B (en) * | 2016-02-19 | 2019-12-31 | 同济大学 | Controller card fault detection system for magnetic-levitation train |
| CN106249101A (en) * | 2016-06-30 | 2016-12-21 | 湖南大学 | A kind of intelligent distribution network fault identification method |
| CN106295712A (en) * | 2016-08-19 | 2017-01-04 | 苏州大学 | A kind of fault detection method and system |
| CN106682814B (en) * | 2016-11-28 | 2021-01-01 | 华北电力大学 | Wind turbine generator fault intelligent diagnosis method based on fault knowledge base |
| CN106886213B (en) * | 2017-03-13 | 2019-10-18 | 北京化工大学 | A kind of batch process fault detection method based on core similarity Support Vector data description |
| CN107505512A (en) * | 2017-04-27 | 2017-12-22 | 中国人民解放军海军航空工程学院 | Electronics intelligent fault monitoring method and device based on JTAG technologies |
| CN107300856B (en) * | 2017-06-30 | 2020-04-17 | 哈尔滨理工大学 | Rotary machine residual life prediction method based on FDA and SVDD |
| CN107843779B (en) * | 2017-09-26 | 2021-03-19 | 国网河南省电力公司民权县供电公司 | Power system fault recording classification analysis method and system based on fuzzy clustering |
| CN108241298A (en) * | 2018-01-09 | 2018-07-03 | 南京航空航天大学 | A kind of aerogenerator method for diagnosing faults based on FWA-RNN models |
| CN108564376A (en) * | 2018-04-20 | 2018-09-21 | 阿里巴巴集团控股有限公司 | Risk control method, device, server and readable storage medium storing program for executing |
| CN108945514B (en) * | 2018-06-19 | 2021-02-05 | 中国航空工业集团公司北京长城航空测控技术研究所 | State detection method based on airplane parameter data |
| CN109376413B (en) * | 2018-10-08 | 2022-08-02 | 四川大学 | Online neural network fault diagnosis method based on data driving |
| CN109726770A (en) * | 2019-02-19 | 2019-05-07 | 金陵科技学院 | A kind of analog circuit fault testing and diagnosing method |
| CN110046419B (en) * | 2019-04-11 | 2023-04-18 | 天津工业大学 | CTMA-DL algorithm-based dynamic equipment fault type online diagnosis method |
| CN110781922A (en) * | 2019-09-27 | 2020-02-11 | 北京淇瑀信息科技有限公司 | Sample data generation method and device for machine learning model and electronic equipment |
| CN110796172A (en) * | 2019-09-27 | 2020-02-14 | 北京淇瑀信息科技有限公司 | Sample label determination method and device for financial data and electronic equipment |
| CN111667379A (en) * | 2020-05-26 | 2020-09-15 | 湖南科技大学 | Fault diagnosis method based on wind power data rising dimension and spherical data fitting |
| CN112051836B (en) * | 2020-09-11 | 2021-09-24 | 江苏科技大学 | Underwater robot propeller state monitoring method based on multi-core model |
| CN112461543B (en) * | 2020-10-28 | 2022-09-13 | 山东科技大学 | Rotary machine fault diagnosis method based on multi-classification support vector data description |
| CN113884292B (en) * | 2021-10-09 | 2023-07-21 | 西安西电开关电气有限公司 | SVDD-based mechanical fault diagnosis method and system for switchgear |
| CN114330599B (en) * | 2022-03-14 | 2022-06-17 | 中汽信息科技(天津)有限公司 | Engine fault identification method, equipment and system based on deep learning |
| CN114880922A (en) * | 2022-04-15 | 2022-08-09 | 大连理工大学 | Civil aviation-oriented flight data analysis and performance trend prediction system |
| CN117686225B (en) * | 2024-02-02 | 2024-04-12 | 浙江大学 | Permanent magnet synchronous motor bearing fault degree diagnosis method and system |
| CN118607390A (en) * | 2024-08-07 | 2024-09-06 | 国网山西省电力公司电力科学研究院 | Intelligent diagnosis and isolation device and method for power distribution network line faults |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101614787A (en) * | 2009-07-07 | 2009-12-30 | 南京航空航天大学 | Analogical Electronics method for diagnosing faults based on M-ary textural classification device |
-
2010
- 2010-07-15 CN CN2010102286876A patent/CN101907681B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101614787A (en) * | 2009-07-07 | 2009-12-30 | 南京航空航天大学 | Analogical Electronics method for diagnosing faults based on M-ary textural classification device |
Non-Patent Citations (4)
| Title |
|---|
| David M j T , Robert P W D.Support vector data description.《Machnie Learning》.2004,45－66. |
| David M j T, Robert P W D.Support vector data description.《Machnie Learning》.2004,45-66. * |
| 唐静远等.基于SVDD和D-S理论的模拟电路故障诊断.《测控技术》.2008,(第09期), * |
| 唐静远等.采用模糊支持向量机的模拟电路故障诊断新方法.《电子测量与仪器学报》.2009,(第06期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101907681A (en) | 2010-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101907681B (en) | Analog circuit dynamic online failure diagnosing method based on GSD-SVDD | |
| CN106156401B (en) | Multi-combination classifier based data driving system state model online identification method | |
| CN109765333A (en) | A kind of Diagnosis Method of Transformer Faults based on GoogleNet model | |
| CN107392901A (en) | A kind of method for transmission line part intelligence automatic identification | |
| CN108178037A (en) | A kind of elevator faults recognition methods based on convolutional neural networks | |
| CN113435546B (en) | Migratable image recognition method and system based on differentiation confidence level | |
| CN112633339A (en) | Bearing fault intelligent diagnosis method, bearing fault intelligent diagnosis system, computer equipment and medium | |
| CN112765890B (en) | Dynamic domain adaptive network-based multi-working-condition rotating machine residual life prediction method | |
| US10417083B2 (en) | Label rectification and classification/prediction for multivariate time series data | |
| CN104050361A (en) | Intelligent analysis early warning method for dangerousness tendency of prison persons serving sentences | |
| CN112700324A (en) | User loan default prediction method based on combination of Catboost and restricted Boltzmann machine | |
| CN110175640A (en) | A kind of Fault Diagnosis Method of Electro-hydraulic based on machine learning | |
| CN111833175A (en) | Internet financial platform application fraud behavior detection method based on KNN algorithm | |
| CN114266289A (en) | Complex equipment health state assessment method | |
| CN115165366A (en) | Variable working condition fault diagnosis method and system for rotary machine | |
| CN117475191A (en) | Bearing fault diagnosis method for feature alignment domain antagonistic neural network | |
| US20040096107A1 (en) | Method and computer program product for determining an efficient feature set and an optimal threshold confidence value for a pattern recogniton classifier | |
| JP7355299B2 (en) | Learning dataset generation system, learning server, and learning dataset generation program | |
| CN115393666A (en) | Small sample expansion method and system based on prototype completion in image classification | |
| CN115271442A (en) | Modeling method and system for evaluating enterprise growth based on natural language | |
| Ebenezer et al. | A comprehensive analysis of handling imbalanced dataset | |
| CN109902731A (en) | A kind of detection method and device of the performance fault based on support vector machines | |
| CN109145968A (en) | A kind of power quality event classification method, system, device and readable storage medium storing program for executing | |
| CN114139598A (en) | Fault diagnosis method and diagnosis framework based on deep cost sensitive convolution network | |
| CN111061626A (en) | Test case priority ranking method based on neuron activation frequency analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120704 Termination date: 20160715 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |