CN107590506B - A kind of complex device method for diagnosing faults of feature based processing - Google Patents
A kind of complex device method for diagnosing faults of feature based processing Download PDFInfo
- Publication number
- CN107590506B CN107590506B CN201710706030.8A CN201710706030A CN107590506B CN 107590506 B CN107590506 B CN 107590506B CN 201710706030 A CN201710706030 A CN 201710706030A CN 107590506 B CN107590506 B CN 107590506B
- Authority
- CN
- China
- Prior art keywords
- feature
- value
- data
- characteristic
- fault
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The present invention provides a kind of complex device method for diagnosing faults of feature based processing, and step is as follows：1st, action current signal is acquired in real time；2nd, section partition is carried out to action current curve；3rd, section partition is carried out to operation curve；4th, the high dimensional feature of structure device action current curve represents data set；5th, data set, which carries out feature selecting, to be represented to high dimensional feature；6th, data set, which carries out feature extraction, to be represented to high dimensional feature；7th, division processing is carried out to character representation data set；8th, optimizing solution is carried out to the parameter of SVM；9th, SVM supervised learnings are carried out, obtain fault diagnosis model；10th, the accuracy rate of diagnosis of fault diagnosis model is verified using test set data；By above-mentioned steps, the complex device method for diagnosing faults of feature based processing can be realized, and the SVM methods for passing through parameter optimization realize fault diagnosis grader, complete fault identification and the analysis of characteristic to equipment operating current curve.
Description
Technical field
The present invention provides a kind of complex device method for diagnosing faults of feature based processing, it is related at a kind of feature based
The realization of the complex device method for diagnosing faults of reason belongs to complex device reliability, complex device fault diagnosis field.
Background technology
Fault diagnosis be according to the equipment running status information searching source of trouble, and determine one of corresponding decision it is comprehensive new
Emerging science.Fault diagnosis technology is started from the 1960s, by the development of more than 50 years, has been achieved for significant progress.Its
Intension develops to the fault diagnosis of complication system from the fault diagnosis of simple device, and nowadays fault diagnosis technology is in each Ministry of Industry
Door achieves extensive use.At present, two problems are mainly concerned with for the fault diagnosis of complex device：First, different operating mould
The corresponding feature description of complex device under formula, difficult point are how to carry out concise and efficient feature to the working condition of equipment
Combination；Second is that the structure of fault diagnosis model, difficult point is how the adaptability and fault-tolerance of diagnostic model are designed,
Model is allow to make intelligent decision according to the different working condition of complex device, failure is carried out accurately to judge and identify.
For two above problem, the present invention proposes a kind of complex device method for diagnosing faults of feature based processing, and this method passes through
The fusion feature expression of Intelligent partition, fischer (Fisher) feature selecting, principal component analysis (Principal Component
Analysis, abbreviation PCA) etc. technologies character representation, selection and extraction are carried out to the current signal of complex device with constitutive characteristic
Data, and exercise supervision to characteristic study using support vector machines (Support Vector Machine, abbreviation SVM),
Fault diagnosis model is established, for the fault diagnosis of complex device.
The processing of this method feature based and the correlation theories such as support vector machines technological incorporation accident analysis and troubleshooting, side
Method and technology are realized, while complex device maintainability is improved, reach improve complex device reliability, safety, can
With the purpose of property.
Invention content
(1) the object of the invention
In the research of complex device fault diagnosis, there are the problem of have：
● the input information of fault diagnosis is indefinite.Although the research for fault diagnosis sorting technique is more, for
The discussion that failure validity feature represents is less.Although there are some character representation methods at present, according to indefinite；
● the machine learning method training time is long, trained cumbersome.Although the optimization algorithm based on machine learning is varied,
It is but poor for the machine learning method effect of optimization of fault diagnosis；
● the research that breakdown judge is carried out to complex device is in the majority.Fault identification is in fault diagnosis to be ignored either
The status of auxiliary, judgement and identification cannot be in combination with carrying out complete complex device fault diagnosis.
In view of the above-mentioned problems, the present invention by overcome the deficiencies in the prior art, proposes that a kind of complexity of feature based processing is set
Standby method for diagnosing faults.The fusion feature that this method carries out Intelligent partition according to the action current data of equipment first represents.So
Right fusion feature is screened using Fisher feature selection approach afterwards, retain conducive to fault diagnosis characteristic, and
The feature of screening is further extracted using PCA methods on the basis of this, the fortune of equipment is represented with minimum, most effective feature combination
Row data.Finally using genetic algorithm (Genetic Algorithm, abbreviation GA) and particle group optimizing method (Particle
Swarm Optimization, abbreviation PSO) optimization algorithm (GA-PSO algorithms) that is combined and SVM methods structure fault diagnosis
Sorter model realizes the fault diagnosis to equipment characteristic.It can be seen that the complex device failure based on machine learning is examined
Disconnected method can not only be indicated the validity feature of failure, the diagnostic model of optimization can also be used to judge failure
With identification, a kind of new solution is provided, and existing method for diagnosing faults is innovated for fault diagnosis field.
(2) technical solution
A kind of complex device method for diagnosing faults of feature based processing of the present invention, its step are as follows：
Step 1 acquires the action current signal of complex device according to the suitable sampling interval in real time, obtains equipment
Operation curve data, and analyze action current tracing pattern relationship corresponding with equipment fault type；
Step 2 carries out section partition, and extract each section operation curve from time domain angle to the action current curve of equipment
The temporal characteristics information of data；
Step 3 carries out section partition, and extract each section operation curve data from codomain angle to the operation curve of equipment
Current value characteristic information；
Step 4, by the temporal characteristics information of device action and current value feature fusion, structure device action electric current is bent
The high dimensional feature of line represents data set；
Step 5 represents that data set carries out feature selecting using Fisher methods to high dimensional feature, filters out and most represents
Property, separability can best feature, and reject it is equivocal, be not easy to sentence another characteristic, so as to reduce the dimension of character representation collection；
Step 6 represents that data set carries out feature extraction using PCA methods to high dimensional feature, eliminates redundancy, further
Reduce the dimension of character representation collection；
Step 7 carries out division processing to character representation data set, obtains SVM training sample sets data and test sample collection
Data；
Step 8, on the basis of training sample set, optimizing solution is carried out to the parameter of SVM using GA-PSO optimization algorithms；
Step 9 carries out SVM supervised learnings using the parameter and training set sample data of optimization, obtains fault diagnosis model；
Step 10 verifies the accuracy rate of diagnosis of fault diagnosis model using test set data.
It is wherein, described " analysis action current tracing pattern relationship corresponding with equipment fault type " in step 1,
The practice is as follows：Normal mode and typical fault pattern to equipment are analyzed, and are summarized the fault type of each fault mode, are occurred
Phenomenon and producing cause search the correspondence of device action current curve form and switch breakdown type, for subsequent
Troubleshooting step.
Wherein, described " carrying out section partition to the operation curve of equipment from time domain angle " in step 2, the practice is such as
Under：Analytical equipment carries out the current signal data of different actions, the current value domain information acted, according to current field value information
Time cut-point is correspondingly arranged to the time domain (x-axis) of operation curve, can be so not same district by the temporal partitioning of operation curve
Section, each time-domain interval represent equipment and different action durations occur.
Wherein, " the temporal characteristics information for extracting each section operation curve data " in step 2, the practice is such as
Under：Usage time describes the physical quantity of operation curve data as independent variable, is set with most basic, most intuitive expression way reaction
The working condition of standby each action section.The present invention is by maximum value, minimum value, average value, variance yields, virtual value, mean square deviation, Feng Feng
Value, kurtosis value, peak factor, shape factor, the pulse factor and nargin factor etc. are used as time domain charactreristic parameter, and time-domain is drawn
The each section divided all carries out the acquisition of as above 12 kinds of characteristic parameters.
Wherein, described " carrying out section partition to the operation curve of equipment from codomain angle " in step 3, the practice is such as
Under：First, action current curve is projected to electric current codomain (y-axis), according to the acquisition precision of device action current signal
Projection interval precision is set, here to avoid influence of the projection section to segmentation current value region, setting projection section
Accuracy value is consistent with acquisition precision value.Then, to scattered current value point zero setting of the projection section points less than 5 points, it is not counted in system
Meter.Finally, electric current codomain is segmented according to the current information of the action of equipment；
Wherein, " the current value characteristic information for extracting each section operation curve data " in step 3, the practice is such as
Under：According to electric current codomain piecewise analysis, the non-zero current region of each section is extracted, and these regions are carried out with statistical nature
Parameter represents.The present invention is using sum, mean value, very poor, mode, variance, standard deviation, intermediate value, mean absolute error etc. as statistics
Characteristic parameter carries out each section the acquisition of above-mentioned 8 kinds of characteristic parameters, to represent the action current feature of equipment.
Wherein, " by the temporal characteristics information of device action and current value feature fusion " in step 4,
The practice is as follows：First, 12 characteristic informations of each time-domain interval of division are summarized, forms multidimensional temporal signatures vector.
Then, 8 characteristic informations of each electric current codomain section of division are summarized, forms multidimensional current value characteristic of field vector；Most
Afterwards, the high dimensional feature that this two multidimensional characteristic vectors are carried out with fusion as device action current curve represents collection.
Wherein, " the Fisher methods " in steps of 5, refers to a kind of two category feature selection methods of classics, the party
Method is using the ratio of the between class distance of each mark sheet indicating value and inter- object distance as Fisher criterion functions, and functional value is bigger, then
Represent that this feature is stronger for the discriminating performance of Classification and Identification.Using the Fisher criterion functions of single feature, feature is arranged
Sequence simultaneously selects the discriminating stronger feature of performance, can form new low-dimensional character representation collection, realize the purpose of Feature Dimension Reduction；
It is wherein, described " representing high dimensional feature that collection data carry out feature selecting using Fisher methods " in steps of 5,
Its practice is as follows：First, using two category feature back-and-forth method of Fisher criterion, each typical fault pattern and normal mode are calculated
Between criterion function value, the d dimensional features F of the device action current curve of the i-th class fault modedFisher criterion functions be：
Here, inter-class variance SB,dWith variance within clusters SW,dIt is respectively defined as：
SB,d=(m1,d-mi,d)2I=1,2 ... n
Here, mi,dAnd σi,dIt is feature F respectivelydMean value and standard deviation in the i-th class.Then, it is determined that feature selecting mark
Standard, the present invention take the mode of " cross semi-selection ", i.e., using the half of the Fisher criterion function value maximum values of the fault mode as
Standard, the Fisher criterion functions value of each dimensional feature data is more than this standard, then is chosen, less than this standard, is then dropped；
Finally, the characteristic selected is combined together, forms the character representation set for being conducive to distinguish typical fault pattern.
Wherein, " the PCA methods " in step 6, refers to a kind of feature extracting method, available for Data Dimensionality Reduction, and
Find element effective and important in data and structure.The main purpose of PCA methods is to go to explain original using less variable
Most of variable in data, and ensure that the information loss of former data is minimum；
It is wherein, described " data set, which carries out feature extraction, to be represented to high dimensional feature using PCA methods " in step 6,
The practice is as follows：
Step 6.1 eigencenter.High dimensional feature is represented that every one-dimensional characteristic attribute data in data set A all subtracts this
The mean value of dimension, make the matrix B obtained after transformation all becomes 0 per one-dimensional mean value；
The covariance matrix C of step 6.2 calculating matrix B；
Step 6.3 calculates the characteristic value and feature vector of covariance matrix C；
Characteristic value is ranked up by step 6.4 according to sequence from big to small, maximum k is selected, then by it
Corresponding k feature vector is respectively as Column vector groups into eigenvectors matrix；
Step 6.5 projects to sample point in the feature vector of selection, obtains new k dimension data collection.Wherein, the selection of k
Foundation is ratio that is preceding k characteristic value and accounting for the sum of all characteristic values
It should be noted that PCA methods solve when without the concern for training set in sample classification information, no matter train
Which classification sample comes from, and can all put on an equal footing.So after PCA Feature Space Transformations, classification can not necessarily be generated
Beneficial Effect, in some instances it may even be possible to bring adverse effect.Therefore, should selection k values be improved, so as to make to reflect according to the requirement of practical problem
New k dimension datas after penetrating are lost to the greatest extent to the greatest extent mostly comprising primary data information (pdi) for effective information of classifying less.
Wherein, described " carrying out division processing to character representation data set " in step 7, the practice is as follows：To feature
Represent data set according to 3:1 ratio, is divided into training set and test set.Training set data is used to carry out SVM modelings, test set
Data are used to carry out the recognition accuracy test of degenerate state identification model.On this basis, to each fault mode characteristic
According to identical quantity is selected to enter training set and test set, accuracy rate of diagnosis, false alarm rate, false dismissed rate can in this way become and judges event
Hinder the index of diagnostic model effectiveness.If sample distribution is excessively uneven, for judging the common finger of fault diagnosis model
Mark missing meaning；
Wherein, " SVM " in step 7, refers to supporting vector machine model, which is the base in statistical theory
A kind of sorter model proposed on plinth, available for pattern classification, linear and nonlinear regression analysis.The original of support vector machines
Reason is given training sample, establishes an Optimal Separating Hyperplane as decision curved surface, makes the isolated border between sample positive example and counter-example
Edge maximizes, and completes the classification of sample.
Wherein, " GA " in step 8, refers to genetic algorithm, which has used for reference Darwinian evolutionism
The natural genetic mechanism of natural selection thought and genetic mechanisms simulates the biology in nature from rudimentary to the mistake of advanced evolution
Journey is a kind of global random searching algorithm.For complicated optimization problem, it is only necessary to select, intersect, make a variation three kinds of genetic operators
It can obtain excellent solution, and unrelated with gradient information using this algorithm solving-optimizing problem, it is only necessary to which object function is can be with
It calculates；
Wherein, " PSO " in step 8, refers to particle swarm optimization algorithm, it is a kind of theoretical based on swarm intelligence
Optimization algorithm, the algorithm have self-teaching and to other people learn two-fold advantage, can be looked in less iterations
To optimal solution；
Wherein, it is described " optimizing solution is carried out to the parameter of SVM using GA-PSO optimization algorithms " in step 8, make
Method is as follows：Using GA as basic framework, PSO optimizations are carried out on the basis of the excellent individual to selecting, algorithm is made to keep good
On the basis of good ability of searching optimum, enhance the ability of local search.From the perspective of genetic algorithm, it is equivalent to outstanding
Sample continue to optimize so that in an iteration, excellent individual is able to the optimization of algorithms of different twice so that parent with
Filial generation " common progress "；From the perspective of population, be equivalent to the information for maintaining optimum individual, select suboptimum individual into
Row intersects and the operation of heredity, not outstanding enough individual have carried out mutation operation, on the one hand saved in iterative process in this way
Good memory so that the on the other hand direction movement that the average fitness of population is become better, ensure that the diversity of population, avoid
It is absorbed in local optimum as soon as possible.The algorithm can keep good Memorability so that search speed is accelerated, while species information
It enriches constantly, avoids the one-way flow of information；
It is wherein, described " carrying out SVM model learnings using the parameter and training set sample data of optimization " in step 9,
Its practice is as follows：
Step 9.1 sets known training set：T={ (x1,y1),…,(xi,yi)}∈(X×Y)l
Wherein, xiFor feature vector, yiFor corresponding property value, xi∈X∈Rn, yi∈ Y={ -1,1 }, i=1,2 ..., l；
Step 9.2 chooses the kernel function g (x, x ') and punishment parameter C of GA-PSO optimizations, constructs and solves optimization problem：
So that
So as to obtain optimal solution
Step 9.3 chooses α*A positive component 0<α*<C, and threshold value b is calculated accordingly*：
Step 9.4 construction decision function f (x)：
Step 9.5 exports classification according to the value of decision function f (x).
Fault diagnosis model can be obtained after the completion of SVM training.If during device fails, which can lead to
Overcurrent signal positions failure, and identifies the pattern class belonging to the failure.
Wherein, it is described in step 10 " accuracy rate of diagnosis of fault diagnosis model to be tested using test set data
Card ", the practice is as follows：Model is verified by way of the injection verification of test set data, compares the diagnosis of fault mode
As a result with actual result, fault diagnosis accuracy rate is acquired, and then judges that can diagnostic model meet demand.
By above-mentioned steps, the complex device method for diagnosing faults of feature based processing can be realized, this method is with complexity
Equipment is object, first, according to the fault characteristic of equipment different faults pattern, carries out Intelligent partition to operating current signal and melts
Close character representation.Then, fusion feature is screened using Fisher feature selection approach, retains and be conducive to carry out fault diagnosis
Feature, and the feature of screening is further extracted with reference to PCA methods.Finally, optimized using the GA-PSO that GA is combined with PSO and calculated
Method obtains SVM Optimal Parameters, and the SVM methods for passing through parameter optimization realize fault diagnosis grader, completes the electricity that works equipment
The fault identification of the characteristic of flow curve and analysis.
(3) advantage
The advantages of the present invention over the prior art are that：At present, most of fault diagnosis technology about complex device
Cannot to equipment acquire data carry out validity feature expression, cause input diagnostic system characteristic there are redundancies and omission
The problems such as.Also, for diagnostic data the training technique operating time is longer, step is cumbersome, practical application effect is poor.And
The present invention can be handled the operational data progress feature " fusion-selection-extraction " that equipment acquires, and obtained and effectively represented equipment event
Hinder the data set of feature.On the basis of data set, using the combinated optimization algorithm of GA and PSO, failure is fast and efficiently obtained
The Optimal Parameters of diagnostic model, and obtain fault diagnosis model by way of SVM supervised learnings.
Description of the drawings
Fig. 1 is the general steps flow of the present invention.
Fig. 2 is the codomain character representation flow of the present invention.
Fig. 3 is the genetic algorithm process flow of the present invention.
Fig. 4 is the particle cluster algorithm process flow of the present invention.
Fig. 5 is heredity-particle cluster algorithm parameter optimization flow of the present invention.
Fig. 6 is the flow of the production support vector machines fault diagnosis model of the present invention.
The method for diagnosing faults that Fig. 7 is the present invention integrally realizes flow.
Serial number, symbol, code name are described as follows in figure：
Fig. 1, " Fisher " in 7 refer to fischer special formula method, and the feature selecting of data set is represented for higher-dimension；
Fig. 1, " PCA " in 7 refer to principal component analytical method, and the feature extraction of data set is represented for higher-dimension；
Fig. 1, " GA-PSO " in 7 refers to genetic algorithm and the method that particle swarm optimization algorithm combines, for supporting vector
The optimization of machine parameter；
Fig. 1, " SVM " in 3,4,7 refers to support vector machines, for generating fault diagnosis model；
" PSO " in Fig. 5 refers to particle group optimizing method, for the optimization of support vector machines parameter.
Specific embodiment
To make the technical problem to be solved in the present invention, technical solution and advantage clearer, carried out below in conjunction with attached drawing
Detailed description.
The present invention proposes a kind of feature based processing and the complex device method for diagnosing faults of support vector machines, uses this
Method can carry out character representation to the action signal data of complex device, obtain character representation data set.In mark sheet registration
On the basis of collection, with reference to GA-PSO algorithms and SVM models, fault diagnosis model is obtained by supervised learning mechanism, for setting
Standby breakdown judge and identification.
A kind of feature based processing of the present invention and the complex device method for diagnosing faults of support vector machines, as shown in Figure 1, its
It is as follows to implement step：
Step 1：The action current signal of complex device is acquired in real time according to the suitable sampling interval, is set
Standby operation curve data, normal mode and typical fault pattern for equipment are analyzed, and summarize the event of each fault mode
Hinder type, phenomenon and producing cause occurs, find the correspondence of device action current curve form and switch breakdown type；
Step 2：Section partition is carried out to the action current curve of equipment from time domain angle, analytical equipment carries out different dynamic
The current signal data of work, the current value domain information acted, according to current field value information to the time domain (x-axis) of operation curve
Time cut-point is correspondingly arranged, can be so different sections by the temporal partitioning of operation curve, wherein each time-domain interval generation
Different action durations occur for table equipment.After the completion of section partition, usage time describes operation curve as independent variable
The physical quantity of data respectively acts the working condition of section with most basic, most intuitive expression way consersion unit.The present invention will most
Big value, minimum value, average value, variance yields, virtual value, mean square deviation, peak-to-peak value, kurtosis value, peak factor, shape factor, pulse
The factor and the nargin factor etc. are used as time domain charactreristic parameter.Each characteristic parameter is described as follows：
| Characteristic parameter | It is described in detail |
| Maximum value | Represent the distributed area of current value |
| Minimum value | Represent the distributed area of current value |
| Average value | Represent the central tendency of current value |
| Virtual value | The average energy of kinetic current value |
| Variance yields | The smoothness of current value is described |
| Mean square deviation | The smoothness of current value is described |
| Peak-to-peak value | Represent the maximum fluctuation situation of current value |
| Kurtosis value | Represent the sensitivity to rush of current signal |
| Peak factor | It represents in current signal with the presence or absence of the index of impact signal |
| Shape factor | Reflect the fluctuation tendency of current signal, and unrelated with amplitude |
| The pulse factor | Represent the sensitivity to current surge pulse class signal |
| The nargin factor | Represent the sensitivity of electric current letter shock pulse signal, which can reduce deviation difference |
Each section that time-domain is divided all carries out the acquisition of as above 12 kinds of characteristic parameters, completes section operation curve
Temporal characteristics information extraction work；
Step 3：The extraction of section partition and characteristic parameter, operation are carried out to the operation curve of equipment from codomain angle
Flow is as shown in Figure 2.First, action current curve is projected to electric current codomain (y-axis), according to device action current signal
Acquisition precision projection interval precision is set, here to avoid influence of the projection section to segmentation current value region, if
Surely projection section accuracy value is consistent with acquisition precision value.Then, scattered current value point of the projection section points less than 5 points is put
Zero, statistics is not counted in, and electric current codomain is segmented according to the current value point of the action of equipment；Finally, according to electric current codomain
Piecewise analysis extracts the non-zero current region of each section, and carries out parameter with statistical nature to these regions and represent.The present invention
Using sum, mean value, very poor, mode, variance, standard deviation, intermediate value, mean absolute error etc. as statistical nature parameter, each feature
Parameter is described as follows：
The acquisition of above-mentioned 8 kinds of characteristic parameters is carried out to each section, completes the current value characteristic information of section operation curve
Extraction work；
Step 4：The temporal characteristics information of device action and current value feature fusion are built to device action electricity
The high dimensional feature of flow curve represents data set.First, 12 characteristic informations of each time-domain interval of division are summarized, formed
Multidimensional temporal signatures vector.Then, 8 characteristic informations of each electric current codomain section of division are summarized, forms multidimensional electricity
Flow valuve characteristic of field vector.Finally, the higher-dimension that this two multidimensional characteristic vectors are carried out with fusion as device action current curve is special
Sign represents collection；
Step 5：Using Fisher methods to high dimensional feature data set carry out feature selecting, pick out it is most representative, can
The best feature of point performance, remove it is equivocal, be not easy to sentence another characteristic, so as to reduce the dimension of character representation collection.It is specific
It operates and is：First, it using two category feature back-and-forth method of Fisher criterion, calculates accurate between each typical fault pattern and normal mode
Then functional value, the d dimensional features F of the device action current curve of the i-th class fault modedFisher criterion functions be：
Here, inter-class variance SB,dWith variance within clusters SW,dIt is respectively defined as：
SB,d=(m1,d-mi,d)2I=1,2 ... n
Here, mi,dAnd σi,dIt is feature F respectivelydMean value and standard deviation in the i-th class.Then, it is determined that feature selecting mark
Standard, the present invention take the mode of " cross semi-selection ", i.e., using the half of the Fisher criterion function value maximum values of the fault mode as
Standard, the Fisher criterion functions value of each dimensional feature data is more than this standard, then is chosen, less than this standard, is then dropped.
Finally, the characteristic selected is combined together, forms the character representation set for being conducive to distinguish typical fault pattern；
Step 6：Data set, which carries out feature extraction, to be represented to the high dimensional feature of equipment using PCA methods, realizes redundancy
Elimination, further reduce character representation collection dimension.Its concrete operations is as follows：
Step 6.1 eigencenter.High dimensional feature is represented that every one-dimensional characteristic attribute data in data set A all subtracts this
The mean value of dimension, make the matrix B obtained after transformation all becomes 0 per one-dimensional mean value；
The covariance matrix C of step 6.2 calculating matrix B；
Step 6.3 calculates the characteristic value and feature vector of covariance matrix C；
Characteristic value is ranked up by step 6.4 according to sequence from big to small, maximum k is selected, then by it
Corresponding k feature vector is respectively as Column vector groups into eigenvectors matrix；
Step 6.5 projects to sample point in the feature vector of selection, obtains new k dimension data collection.Wherein, the selection of k
Foundation is ratio that is preceding k characteristic value and accounting for the sum of all characteristic values.
It should be noted that PCA methods solve when without the concern for training set in sample classification information, no matter train
Which classification sample comes from, and can all put on an equal footing training sample.After PCA Feature Space Transformations, not necessarily can to point
Class generates Beneficial Effect, in some instances it may even be possible to bring adverse effect.Therefore, it should according to the requirement of practical problem, selection k values are improved,
So as to which the new k dimension datas after mapping be made mostly comprising primary data information (pdi), to lose to the greatest extent for effective information of classifying to the greatest extent less；
Step 7：To character representation data set according to 3:1 ratio, is divided into training set and test set.Training set data
It is modeled for SVM, test set data are verified for the accuracy rate of diagnosis of fault diagnosis model.On this basis, each failure mould
Formula selects identical quantity to enter training set and test set, makes accuracy rate of diagnosis, false alarm rate, false dismissed rate as that can judge failure
The index of diagnostic model effectiveness.If sample distribution is excessively uneven, for judging the common counter of fault diagnosis model
Then lack meaning；
Step 8：On the basis of training sample set, optimizing solution is carried out to the parameter of SVM using GA-PSO algorithms.GA
Algorithm is a kind of global random searching algorithm, for complicated optimization problem, it is only necessary to select, intersect, the three kinds of heredity that make a variation are calculated
The solution that son can be optimized, and its solving-optimizing problem is unrelated with gradient information, it is only necessary to object function, the processing of GA
Flow is as shown in Figure 3；PSO is a kind of optimization method based on swarm intelligence theory, this method have self-teaching and to other people
The two-fold advantage of study, can find optimal solution in less iterations, and the process flow of PSO is as shown in Figure 4.The present invention
With reference to the advantages of GA, PSO, optimizing solution is carried out to the nuclear parameter and punishment parameter of SVM, is specifically handled as shown in figure 5, this is excellent
Change algorithm using GA as basic framework, to progress PSO optimizations on the basis of the excellent individual selected, it is intended to keep algorithm
On the basis of good ability of searching optimum, enhance the ability of local search.From the point of view of GA, it is equivalent to outstanding sample
Continue to optimize so that in an iteration, excellent individual obtains the optimization of algorithms of different twice so that parent and filial generation
" common progress "；From the perspective of PSO, the information for maintaining optimum individual is equivalent to, the individual of suboptimum is selected to be intersected
With the operation of heredity, not outstanding enough individual has carried out mutation operation, has on the one hand saved the good note in iterative process in this way
Recall so that on the other hand the direction movement that the average fitness of population is become better, ensure that the diversity of population, avoid as soon as possible
It is absorbed in local optimum.
Step 9：SVM supervised learnings are carried out using the parameter and training set sample data of GA-PSO algorithm optimizations, obtain event
Hinder diagnostic model, idiographic flow as shown in fig. 6, " selected training set and test set " and " data prediction " in figure at it
It is completed in preceding " step 4 " to " step 7 ".This step is substantially carried out " training pattern " and " generation fault diagnosis model " two
Point, the training process of SVM is the supervised learning process using classification accuracy rate as index, that is, finds the process of optimal classifying face.
In training, penalty c and nuclear parameter g can generate the optimal performance of SVM classifier conclusive influence, and the present invention exists
The Optimization Work that GA-PSO methods complete both parameters is employed in " step 8 ", therefore combines the parameter and training set of optimization
Sample data, the practice for carrying out SVM supervised learnings are as follows：
Step 9.1 sets known training set：T={ (x1,y1),…,(xi,yi)}∈(X×Y)l
Wherein, xiFor feature vector, yiFor corresponding property value, xi∈X∈Rn, yi∈ Y={ -1,1 }, i=1,2 ..., l；
Step 9.2 chooses the kernel function g (x, x ') and punishment parameter C of GA-PSO optimizations, constructs and solves optimization problem：
So that
So as to obtain optimal solution
Step 9.3 chooses α*A positive component 0<α*<C, and threshold value b is calculated accordingly*：
Step 9.4 construction decision function f (x)：
Step 9.5 exports classification according to the value of decision function f (x).
Fault diagnosis model can be obtained after the completion of SVM training.If during device fails, which can lead to
Overcurrent signal positions failure, and identifies the pattern class belonging to the failure.And " diagnostic result verification " meeting in Fig. 6
It is completed in " step 10 " below；
Step 10：The present invention, which adopts, to be verified to the accuracy rate of diagnosis of fault diagnosis model using test set sample data
Model is verified with the mode that test set sample data is injected, and must be paid a home visit with actual result by the diagnostic result for comparing fault mode
Disconnected accuracy rate, and then judge that can diagnostic model meet demand.
By above-mentioned steps, a kind of complex device method for diagnosing faults of feature based processing can be realized, it is whole real
Existing flow is as shown in Figure 7.This method is using complex device as object, first according to the fault characteristic of equipment different faults pattern,
The fusion feature that Intelligent partition is carried out to operating current signal represents.Then using Fisher feature selection approach to fusion feature
It is screened, retains the feature for being conducive to carry out fault diagnosis, and the feature of screening is further extracted with reference to PCA methods.Finally,
The GA-PSO optimization algorithms combined using GA with PSO obtain SVM Optimal Parameters, and the SVM methods for passing through parameter optimization realize event
Hinder diagnostic classification device, complete fault identification and the analysis of characteristic to equipment operating current curve.
Non-elaborated part of the present invention belongs to techniques well known.
The above, part specific embodiment only of the present invention, but protection scope of the present invention is not limited thereto, and is appointed
In the technical scope disclosed by the present invention, the change or replacement that can be readily occurred in should all be covered what those skilled in the art
Within protection scope of the present invention.
Claims (11)
1. a kind of complex device method for diagnosing faults of feature based processing, it is characterised in that：Its step are as follows：
Step 1 acquires the action current signal of complex device according to the suitable sampling interval in real time, obtains the dynamic of equipment
Make curve data, and analyze action current tracing pattern relationship corresponding with equipment fault type；
Step 2 carries out section partition, and extract each section operation curve data from time domain angle to the action current curve of equipment
Temporal characteristics information；
Step 3 carries out section partition, and extract the electricity of each section operation curve data from codomain angle to the operation curve of equipment
Flow valuve characteristic information；
Step 4, by the temporal characteristics information of device action and current value feature fusion, structure device action current curve
High dimensional feature represents data set；
Step 5, using Fisher methods to high dimensional feature represent data set carry out feature selecting, filter out it is most representative, can
The best feature of point performance, and reject it is equivocal, be not easy to sentence another characteristic, so as to reduce the dimension of character representation collection；
Step 6 represents that data set carries out feature extraction using PCA methods to high dimensional feature, eliminates redundancy, further reduces
The dimension of character representation collection；
Step 7 carries out division processing to character representation data set, obtains SVM training sample sets data and test sample collection data；
Step 8, on the basis of training sample set, optimizing solution is carried out to the parameter of SVM using GA-PSO optimization algorithms；
Step 9 carries out SVM supervised learnings using the parameter and training set sample data of optimization, obtains fault diagnosis model；
Step 10 verifies the accuracy rate of diagnosis of fault diagnosis model using test set data；
By above-mentioned steps, can realize the complex device method for diagnosing faults of feature based processing, this method using complex device as
Object first, according to the fault characteristic of equipment different faults pattern, carries out operating current signal the fusion feature of Intelligent partition
It represents；Then, fusion feature is screened using Fisher feature selection approach, retains the spy for being conducive to carry out fault diagnosis
Sign, and the feature of screening is further extracted with reference to PCA methods；Finally, the GA-PSO optimization algorithms combined using GA with PSO are obtained
To SVM Optimal Parameters, and the SVM methods for passing through parameter optimization realize fault diagnosis grader, complete to equipment operating current song
The fault identification of the characteristic of line and analysis.
2. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
Analysis action current tracing pattern relationship corresponding with equipment fault type, the practice are as follows in step 1：It is right
The normal mode and typical fault pattern of equipment are analyzed, and summarize the fault type of each fault mode, phenomenon and production occurs
Raw reason searches the correspondence of device action current curve form and fault type, for subsequent troubleshooting step.
3. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
It is described in step 2 that section partition is carried out, and extract each section and move to the action current curve of equipment from time domain angle
Make the temporal characteristics information of curve data, the practice is as follows：
The slave time domain angle carries out section partition to the operation curve of equipment, and the specific practice is as follows：Analytical equipment carries out
Difference action current signal data, the current value domain information acted, according to current field value information to operation curve when
Domain, that is, x-axis is correspondingly arranged time cut-point, is different sections in this way by the temporal partitioning of operation curve, each time-domain interval represents
Different action durations occur for equipment；
The temporal characteristics information of each section operation curve data of extraction, the specific practice are as follows：Usage time is used as certainly
The physical quantity of variable description operation curve data respectively acts the work of section with most basic, most intuitive expression way consersion unit
Make state；By maximum value, minimum value, average value, variance yields, virtual value, mean square deviation, peak-to-peak value, kurtosis value, peak factor, wave
Each section that time-domain is divided all is carried out maximum by the shape factor, the pulse factor and the nargin factor as time domain charactreristic parameter
Value, minimum value, average value, variance yields, virtual value, mean square deviation, peak-to-peak value, kurtosis value, peak factor, shape factor, pulse because
The acquisition of son and nargin ratio characteristics parameter.
4. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
Described in step 3 to carry out section partition to the operation curve of equipment from codomain angle, the practice is as follows：First, will
Action current curve is projected to electric current codomain, that is, y-axis, according to the acquisition precision of device action current signal to projecting section
Precision is set, and here to avoid influence of the projection section to segmentation current value region, setting projection section accuracy value is with adopting
It is consistent to collect accuracy value；Then, to scattered current value point zero setting of the projection section points less than 5 points, it is not counted in statistics；Finally, root
Electric current codomain is segmented according to the current information of the action of equipment；
The current value characteristic information of each section operation curve data of extraction, the practice are as follows in step 3：According to electric current
Codomain piecewise analysis extracts the non-zero current region of each section, and carries out parameter with statistical nature to these regions and represent；It will
Sum, mean value, very poor, mode, variance, standard deviation, intermediate value, mean absolute error are as statistical nature parameter, to each section
The acquisition of sum, mean value, very poor, mode, variance, standard deviation, intermediate value, mean absolute error characteristic parameter is carried out, to represent to set
Standby action current feature.
5. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
Described by the temporal characteristics information of device action and current value feature fusion in step 4, the practice is as follows：It is first
First, 12 characteristic informations of each time-domain interval of division are summarized, forms multidimensional temporal signatures vector；Then, to dividing
8 characteristic informations of each electric current codomain section summarized, form multidimensional current value characteristic of field vector；Finally, to this two
Multidimensional characteristic vectors carry out fusion and represent collection as the high dimensional feature of device action current curve.
6. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
The Fisher methods in steps of 5, refer to a kind of two category feature selection methods of classics, and this method is by each feature
The between class distance of expression value and the ratio of inter- object distance are as Fisher criterion functions, and functional value is bigger, then it represents that this feature
It is stronger for the discriminating performance of Classification and Identification；Using the Fisher criterion functions of single feature, to feature ordering and mirror is selected
The other stronger feature of performance can form new low-dimensional character representation collection, realize the purpose of Feature Dimension Reduction；
The use Fisher methods represent that data set carries out feature selecting to high dimensional feature in steps of 5, and the practice is as follows：
First, using two category feature back-and-forth method of Fisher criterion, criterion function between each typical fault pattern and normal mode is calculated
Value, the d dimensional features F of the device action current curve of the i-th class fault modedFisher criterion functions be：
Here, inter-class variance SB,dWith variance within clusters SW,dIt is respectively defined as：
Here, mi,dAnd σi,dIt is feature F respectivelydMean value and standard deviation in the i-th class；Then, it is determined that feature selecting standard, is adopted
Took semi-selective mode, i.e., using the half of the Fisher criterion function value maximum values of the fault mode as standard, each dimensional feature
The Fisher criterion functions value of data is more than this standard, then is chosen, less than this standard, is then dropped；Finally, it will select
Characteristic be combined together, formed be conducive to distinguish typical fault pattern character representation set.
7. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
The PCA methods in step 6, refer to a kind of feature extracting method, available for Data Dimensionality Reduction, and find in data and have
Effect and important element and structure；The main purpose of PCA methods is the big portion for going to explain using less variable in former data
Variation per minute, and ensure that the information loss of former data is minimum；
The use PCA methods represent that data set carries out feature extraction to high dimensional feature in step 6, and the practice is as follows：
Step 6.1 eigencenter；High dimensional feature is represented that every one-dimensional characteristic attribute data in data set A all subtracts the dimension
Mean value, make the matrix B obtained after transformation all becomes 0 per one-dimensional mean value；
The covariance matrix C of step 6.2 calculating matrix B；
Step 6.3 calculates the characteristic value and feature vector of covariance matrix C；
Characteristic value is ranked up by step 6.4 according to sequence from big to small, is selected maximum k, is then corresponded to
K feature vector respectively as Column vector groups into eigenvectors matrix；
Step 6.5 projects to sample point in the feature vector of selection, obtains new k dimension data collection, wherein, the selection gist of k
For ratio that is preceding k characteristic value and accounting for the sum of all characteristic values；
It should be noted that PCA methods solve when without the concern for training set in sample classification information, no matter training sample
Which, from classification, can all put on an equal footing；So after PCA Feature Space Transformations, classification can not necessarily be generated advantageous
It influences, in some instances it may even be possible to bring adverse effect；Therefore, should selection k values be improved, after making mapping according to the requirement of practical problem
New k dimension datas mostly include primary data information (pdi) to the greatest extent, lose less to the greatest extent for the effective information of classification.
8. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
Described in step 7 to carry out division processing to character representation data set, the practice is as follows：Character representation data set is pressed
According to 3:1 ratio, is divided into training set and test set；For carrying out SVM modelings, test set data are used to carry out training set data
The recognition accuracy test of degenerate state identification model；On this basis, each fault mode characteristic is selected identical
Quantity enters training set and test set, and accuracy rate of diagnosis, false alarm rate, false dismissed rate can be made to become judge fault diagnosis model in this way to be had
The index of effect degree；If sample distribution is excessively uneven, for judging the common counter of fault diagnosis model missing meaning；
The SVM in step 7, refers to supporting vector machine model, which proposes on the basis of statistical theory
A kind of sorter model, for pattern classification, linear and nonlinear regression analysis；The principle of support vector machines is given training
Sample establishes an Optimal Separating Hyperplane as decision curved surface, maximizes the isolation edge between sample positive example and counter-example, complete
The classification of sample.
9. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
Described in step 8 to carry out optimizing solution to the parameter of SVM using GA-PSO optimization algorithms, the practice is as follows：With GA
For basic framework, PSO optimizations are carried out on the basis of the excellent individual to selecting, algorithm is made to keep good global search
On the basis of ability, enhance the ability of local search；From the perspective of genetic algorithm, be equivalent to outstanding sample is continued into
Row optimization so that in an iteration, excellent individual is able to the optimization of algorithms of different twice so that parent and filial generation jointly into
Step；From the perspective of population, be equivalent to the information for maintaining optimum individual, select the individual of suboptimum intersected and heredity
Operation, not outstanding enough individual carried out mutation operation, on the one hand saved the good memory in iterative process in this way so that
The direction movement that the average fitness of population is become better, on the other hand, ensure that the diversity of population, avoids being absorbed in part as soon as possible
Optimal value；The algorithm can keep good Memorability so that search speed is accelerated, while species information is enriched constantly, avoids
The one-way flow of information.
10. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
The parameter of use optimization and training set sample data carry out SVM supervised learnings in step 9,
Its practice is as follows：
Step 9.1 sets known training set：
Wherein, xiFor feature vector, yiFor corresponding property value, xi∈X∈Rn,
Step 9.2 chooses the kernel function g and punishment parameter C of GA-PSO optimizations, constructs and solves optimization problem：
So that
So as to obtain optimal solution set
Step 9.3 chooses α*A positive component 0<α*<C, and threshold value b is calculated accordingly*：
Step 9.4 construction decision function f (x)：
Step 9.5 exports classification according to the value of decision function f (x)；
Fault diagnosis model can be obtained after the completion of SVM training；If during device fails, which can be believed by electric current
Number failure is positioned, and identify the pattern class belonging to the failure.
11. a kind of complex device method for diagnosing faults of feature based processing according to claim 1, it is characterised in that：
Described in step 10 that the accuracy rate of diagnosis of fault diagnosis model is verified using test set data, the practice is such as
Under：Model is verified by way of the injection verification of test set data, the diagnostic result for comparing fault mode is tied with practical
Fruit acquires fault diagnosis accuracy rate, and then judges that can diagnostic model meet demand.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710706030.8A CN107590506B (en) | 2017-08-17 | 2017-08-17 | A kind of complex device method for diagnosing faults of feature based processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710706030.8A CN107590506B (en) | 2017-08-17 | 2017-08-17 | A kind of complex device method for diagnosing faults of feature based processing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107590506A CN107590506A (en) | 2018-01-16 |
| CN107590506B true CN107590506B (en) | 2018-06-15 |
Family
ID=61042214
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710706030.8A Active CN107590506B (en) | 2017-08-17 | 2017-08-17 | A kind of complex device method for diagnosing faults of feature based processing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107590506B (en) |
Families Citing this family (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345284B (en) * | 2018-03-06 | 2020-06-16 | 宁波大学 | Quality-related fault detection method based on two variable blocks |
| CN108549741B (en) * | 2018-03-07 | 2021-06-11 | 上海交通大学 | Shield tunneling machine main bearing sealing performance health assessment method and system |
| CN110363208B (en) * | 2018-04-09 | 2022-08-12 | 南瑞集团有限公司 | Multi-parameter fusion stacker track fault detection and early warning method |
| CN108985159A (en) * | 2018-06-08 | 2018-12-11 | 平安科技（深圳）有限公司 | Human-eye model training method, eye recognition method, apparatus, equipment and medium |
| CN108985279B (en) * | 2018-08-28 | 2020-11-03 | 上海仁童电子科技有限公司 | Fault diagnosis method and device for MVB waveform of multifunctional vehicle bus |
| TWI684882B (en) * | 2018-12-21 | 2020-02-11 | 帆宣系統科技股份有限公司 | Automatic optimization method for generating fault characteristics |
| CN109947080B (en) * | 2019-03-21 | 2022-06-07 | 北京明略软件系统有限公司 | Fault diagnosis method and device, computer storage medium and terminal |
| CN110349289A (en) * | 2019-06-21 | 2019-10-18 | 东软集团股份有限公司 | Failure prediction method, device, storage medium and electronic equipment |
| CN110441629B (en) * | 2019-07-29 | 2022-02-15 | 交控科技股份有限公司 | Fault diagnosis method and device based on turnout switch machine operating current curve |
| CN110503004B (en) * | 2019-07-29 | 2022-03-22 | 七彩安科智慧科技有限公司 | On-line judging method for operating state of switching power supply |
| CN110579709B (en) * | 2019-08-30 | 2021-04-13 | 西南交通大学 | Fault diagnosis method for proton exchange membrane fuel cell for tramcar |
| CN110717301B (en) * | 2019-09-19 | 2024-06-28 | 中国石油大学(华东) | Flow unit information classification and identification method based on support vector machine algorithm |
| CN110674725B (en) * | 2019-09-20 | 2022-06-03 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN110674991A (en) * | 2019-09-25 | 2020-01-10 | 国家能源集团谏壁发电厂 | OCSVM (online charging management system VM) -based method for detecting abnormality of primary fan of thermal power plant |
| CN110824280B (en) * | 2019-10-08 | 2021-08-17 | 西南交通大学 | Diagnosis method and visualization method of turnout health state based on feature similarity |
| CN110866502B (en) * | 2019-11-19 | 2023-02-14 | 安徽工业大学 | Fault diagnosis method based on linear discriminant analysis and particle swarm optimization support vector machine |
| CN110927243A (en) * | 2019-12-09 | 2020-03-27 | 金陵科技学院 | Cold-rolled strip steel fault diagnosis optimization method based on PCA-Kmeans algorithm |
| CN111307453B (en) * | 2020-03-20 | 2021-11-12 | 朗斯顿科技(北京)有限公司 | Transmission system fault diagnosis method based on multi-information fusion |
| CN111474475B (en) * | 2020-03-22 | 2021-06-08 | 华南理工大学 | Motor fault diagnosis system and method |
| CN113537519B (en) * | 2020-04-10 | 2024-05-24 | 北京京东乾石科技有限公司 | Method and device for identifying abnormal equipment |
| CN111641384B (en) * | 2020-04-28 | 2021-10-22 | 特变电工新疆新能源股份有限公司 | Photovoltaic power station string fault diagnosis method, device, equipment and readable storage medium |
| CN111667355A (en) * | 2020-05-29 | 2020-09-15 | 北京睿知图远科技有限公司 | Big data analysis method and system |
| CN111928925B (en) * | 2020-07-16 | 2022-12-09 | 中广核核电运营有限公司 | Venturi tube fault correction method and device, computer equipment and storage medium |
| CN112084910A (en) * | 2020-08-28 | 2020-12-15 | 北京旋极信息技术股份有限公司 | Fault diagnosis method and system |
| CN112633345A (en) * | 2020-12-17 | 2021-04-09 | 鄂尔多斯应用技术学院 | Method for establishing coal mining machine digital dictionary based on multi-view method |
| CN112650200B (en) * | 2020-12-25 | 2022-08-30 | 中能融安(北京)科技有限公司 | Method and device for diagnosing plant station equipment faults |
| CN113049749B (en) * | 2021-03-12 | 2022-05-10 | 中国农业大学 | Electronic nose detection system |
| CN113361324B (en) * | 2021-04-25 | 2023-06-30 | 杭州玖欣物联科技有限公司 | Lstm-based motor current anomaly detection method |
| CN113484662A (en) * | 2021-06-02 | 2021-10-08 | 青岛鼎信通讯股份有限公司 | Fault indicator starting wave recording method |
| CN113641990A (en) * | 2021-06-21 | 2021-11-12 | 上海电力大学 | Intrusion detection method based on multi-innovation extended Kalman filtering |
| CN114021602A (en) * | 2021-10-14 | 2022-02-08 | 北京和利时系统集成有限公司 | Method and system for processing data in point switch fault diagnosis model |
| CN114021671B (en) * | 2022-01-05 | 2022-04-05 | 中国电力科学研究院有限公司 | Active power distribution network online fault processing method and system |
| CN116881700A (en) * | 2023-07-19 | 2023-10-13 | 中国人民解放军总参谋部第六十研究所 | Fault diagnosis method based on random walk and machine learning |
| CN117871994B (en) * | 2023-12-22 | 2024-07-23 | 湖南奕坤科技有限公司 | Rapid fault detection method and system for PLC (programmable logic controller) electric cabinet |
| CN117932322B (en) * | 2024-03-22 | 2024-06-04 | 山东冠县鑫恒祥面业有限公司 | Flour equipment fault diagnosis method and system |
| CN118332477B (en) * | 2024-06-13 | 2024-08-23 | 中国人民解放军海军航空大学 | Equipment fault detection model generation method based on state data analysis |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7693643B2 (en) * | 2005-02-14 | 2010-04-06 | Honeywell International Inc. | Fault detection system and method for turbine engine fuel systems |
| CN101814114B (en) * | 2010-01-07 | 2012-03-21 | 北京航空航天大学 | Method for performing fault diagnosis by using model conversion |
| US9824060B2 (en) * | 2010-07-23 | 2017-11-21 | Polyvalor, Limited Partnership | Tool and method for fault detection of devices by condition based maintenance |
| CN104506137A (en) * | 2014-11-26 | 2015-04-08 | 北京金鸿泰科技有限公司 | Equipment fault diagnosis method and apparatus |
| CN105550700B (en) * | 2015-12-08 | 2019-04-09 | 国网山东省电力公司电力科学研究院 | A kind of time series data cleaning method based on association analysis and principal component analysis |
| CN106844194B (en) * | 2016-12-21 | 2018-06-15 | 北京航空航天大学 | A kind of construction method of multi-level software fault diagnosis expert system |
-
2017
- 2017-08-17 CN CN201710706030.8A patent/CN107590506B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107590506A (en) | 2018-01-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107590506B (en) | A kind of complex device method for diagnosing faults of feature based processing | |
| CN109102005A (en) | Small sample deep learning method based on shallow Model knowledge migration | |
| US10706332B2 (en) | Analog circuit fault mode classification method | |
| CN106201871B (en) | Based on the Software Defects Predict Methods that cost-sensitive is semi-supervised | |
| CN104598813B (en) | Computer intrusion detection method based on integrated study and semi-supervised SVM | |
| CN111126820B (en) | Method and system for preventing electricity stealing | |
| CN110378869A (en) | A kind of rail fastening method for detecting abnormality of sample automatic marking | |
| CN106570513A (en) | Fault diagnosis method and apparatus for big data network system | |
| CN109582003A (en) | Based on pseudo label semi-supervised kernel part Fei Sheer discriminant analysis bearing failure diagnosis | |
| CN101738998B (en) | System and method for monitoring industrial process based on local discriminatory analysis | |
| CN102291392B (en) | Hybrid intrusion detection method based on Bagging algorithm | |
| CN106202952A (en) | A kind of Parkinson disease diagnostic method based on machine learning | |
| CN110503004A (en) | A kind of online method of discrimination of Switching Power Supply operating status | |
| CN103390154A (en) | Face recognition method based on extraction of multiple evolution features | |
| CN106482967A (en) | A kind of Cost Sensitive Support Vector Machines locomotive wheel detecting system and method | |
| CN114707571B (en) | Credit data anomaly detection method based on enhanced isolation forest | |
| CN113376516A (en) | Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning | |
| CN108664538A (en) | A kind of automatic identification method and system of the doubtful familial defect of power transmission and transforming equipment | |
| CN108375729B (en) | Degenerate state appraisal procedure is clustered based on the motor SOM that Fisher differentiates | |
| CN110135487A (en) | A kind of computer user mouse Behavior modeling method | |
| CN108154924A (en) | Alzheimer's disease tagsort method and system based on support vector machines | |
| CN110263867A (en) | A kind of rail defects and failures classification method | |
| Zhang et al. | Trajectory series analysis based event rule induction for visual surveillance | |
| CN110288028A (en) | ECG detecting method, system, equipment and computer readable storage medium | |
| CN109800790A (en) | A kind of feature selection approach towards high dimensional data |
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 |