CN109491816A - Knowledge based engineering method for diagnosing faults - Google Patents
Knowledge based engineering method for diagnosing faults Download PDFInfo
- Publication number
- CN109491816A CN109491816A CN201811220751.9A CN201811220751A CN109491816A CN 109491816 A CN109491816 A CN 109491816A CN 201811220751 A CN201811220751 A CN 201811220751A CN 109491816 A CN109491816 A CN 109491816A
- Authority
- CN
- China
- Prior art keywords
- fault
- knowledge
- value
- factor
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
- G06F11/079—Root cause analysis, i.e. error or fault diagnosis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/046—Forward inferencing; Production systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of Knowledge based engineering method for diagnosing faults, comprising the following steps: establishes typical fault case database；Using signal processing technology and Principle components analysis method, the extraction of fault characteristic information and the selection of key data variable are completed；Using improved neural network algorithm, fault sample is learnt；In such a way that production rule and frame combine, the integrating representation of diagnostic knowledge is realized；System failure propagation model is constructed by the relationship between each equipment in data analysis and system, specifies system failure propagation path；Fault knowledge based on acquisition constructs knowledge graph, provides inference rule using road algebraic process, realize the fault location of knowledge-based inference.The present invention is theoretical by improved neural network algorithm and knowledge reasoning, realizes the automatic acquisition and the accurate positioning of failure of knowledge, improves diagnostic accuracy, provide foundation for industrial control system fault diagnosis.
Description
Technical field
The present invention relates to a kind of industrial control system method for diagnosing faults more particularly to a kind of Knowledge based engineering fault diagnosises
Method.
Background technique
With two change fusions, industry 4.0, the landing of 2025 concept of made in China, traditional factory and industrial control system
Inevitable trend interconnection, fusion, consequent is that the maintenance cost of equipment is higher and higher, how to improve the reliability of equipment, is protected
Barrier equipment safety, stabilization, macrocyclic operation have become intelligent production equipment fault diagnosis field urgent problem to be solved.Failure
Diagnostic techniques through the entire life cycle of the design of industrial equipment, development, the manufacturing, test and working service, examine by failure
Disconnected software has become the quality assurance tool of industrial equipment design and the important means, product manufacturing developed, industrial equipment repairs
The main element of system, level have become the important mark for measuring a National modern industrial development ability and manufacturing level
One of will.
External aspect, the U.S. are the countries that research fault diagnosis technology is earliest in the world, and 1967, mechanical breakdown prevention was small
Group (MFPG) is set up in the U.S., starts to study each special topic of fault diagnosis technology respectively.Muhammet Unal et al.
Using the topological structure of genetic algorithm optimization neural network, by taking the fault diagnosis of rolling bearing as an example, fault diagnosis is demonstrated
The feasibility of scheme.Angeli et al. proposes the expert system that distributed real-time diagnosis is combined based on multi-model, to realize work
Fault diagnosis during industry.Method of the Bagheri et al. based on parameter Estimation, according to each Parameters variation in system model
Statistics feature to carry out fault diagnosis to system.
The fault diagnosis technology development of domestic aspect, China is later, substantially starts from the phase at the beginning of the eighties in last century, so far
Achieve many achievements outstanding.Pan's Chong, Chen Weigen et al. combine genetic algorithm with wavelet neural network, are calculated using heredity
The structure and weight of method optimization neural network enable network obtain preferable general Huaneng Group power so that the training process of network is accelerated,
And be applied to the algorithm in the fault diagnosis of transformer, demonstrate its feasibility.Zhang Huaqiang, Zhao Yan et al. utilize Gaussian function
The excitation function for replacing neural network, proposes a kind of fault diagnosis scheme based on adaptive probability neural network, the program
Make network that there is preferable generalization ability.In addition, Xu Min etc. has been edited monograph " equipment fault diagnosis handbook ", the chief editor such as yellow literary riddles
" equipment fault diagnosis principle, technology and application " is proposed with fuzzy diagnosis model, confidence factor model, production rule
The method for then etc. establishing fault diagnosis expert system.
Although modern most of fault diagnosis and fault prediction systems can automatically carry out measurand with cracking speed
Fault diagnosis, but with the raising of device intelligence, networking and complexity, existing system be easy to cause fault diagnosis
Ambiguity and uncertainty, and it is poor to the automaticity of Fault Knowledge Acquisition.Therefore, it is necessary to special based on history case and field
Family's knowledge makes full use of data analysis and artificial intelligence technology, effectively solves Fault Knowledge Acquisition and fault location demand, improves
The efficiency and reliability of fault diagnosis, thus powerful guarantee industrial control system stable operation.
Summary of the invention
The purpose of the present invention is to provide a kind of Knowledge based engineering method for diagnosing faults, by fully considering Industry Control system
The fault characteristic of system completes the extraction of fault characteristic information, and the acquisition of fault knowledge is realized using neural network algorithm, then sharp
Failure is completed with knowledge reasoning to be accurately positioned.
The technical solution for realizing the aim of the invention is as follows: a kind of Knowledge based engineering method for diagnosing faults, including following step
It is rapid:
Typical fault Database: the first step is directed to industrial control system, establishes typical fault case database；
Second step, fault characteristic information extract: the typical fault database based on foundation is mentioned using signal processing technology
Obtain time domain charactreristic parameter, time and frequency zone characteristic parameter；
Third step, Feature Selection: the data set formed according to characteristic parameter takes Principle components analysis method, by data
The key data attribute variable of collection extracts；
Neural network algorithm training study: 4th step utilizes neural network algorithm, to system failure sample set
Training is practised, the acquisition of fault knowledge is completed；
5th step, the representation of knowledge of system-oriented fault diagnosis: for the fault knowledge of acquisition, using frame and production
The mode that rule combines indicates specific malfunctioning module with fault tree framework method, indicates failure with production rule
Premise and as a result, obtaining the Rule Expression of fault diagnosis knowledge；
6th step, system failure propagation model building: firstly, capabilities map relationship of the building system task to completion task
Model；Then, mapping relations model of the structuring capacity to system comprising modules；Finally, event of the building based on module dependence
Hinder propagation model；
7th step, the fault location of knowledge-based inference: element and relationship type in analysis system fault knowledge determine
Qualified relation type, determines inference rule, constructs knowledge graph, introduces road algebra and carries out statement and operation to derivation relationship, to pushing away
Reason result credibility is calculated, and is provided confidence level foundation, is realized the positioning of the system failure.
Compared with prior art, the present invention its remarkable advantage are as follows: (1) realize failure based on improved BP neural network algorithm
The acquisition of diagnostic knowledge, also overcomes traditional neural network at the problems such as overcoming other algorithms " multiple shot array ", " infinite recursion "
The defects of algorithm " local optimum ", " convergence rate decline "；(2) it is indicated using the method that frame is combined with production rule
Diagnostic knowledge indicates specific malfunctioning module with fault tree framework method, and the premise and knot of failure are indicated with production rule
Fruit, simplifying complicated knowledge indicates step, specifies relevant fault relationship, improves representation of knowledge efficiency；(3) the building system failure is propagated
Model obtains the mapping relations of failure symptom and failure cause, passes through the state parameter of test equipment, acquisition and analysis failure sign
Million data can be predicted with position before failure generation the reason of may breaking down；(4) side of knowledge graph is used
Method, which is described knowledge and quotes the theoretical of road algebra, realizes reasoning, in fusion fault case, expert diagnosis experience and failure
On the basis of three kinds of knowledge of propagation model, allow to infer the new relation that do not mention in primary knowledge base, to be effectively improved
The limitation of industrial control system fault location.
Detailed description of the invention
Fig. 1 is Troubleshooting Flowchart of the present invention.
Fig. 2 Principle components analysis method figure.
Fig. 3 neural metwork training learning process figure.
The diagnostic knowledge expression figure that Fig. 4 frame is combined with production rule.
Fault- traverse technique schematic diagram of the Fig. 5 based on dependence.
The fault location technology scheme schematic diagram of Fig. 6 knowledge-based inference.
Specific embodiment
As shown in Figure 1, a kind of Knowledge based engineering method for diagnosing faults of the invention, comprising the following steps:
Typical fault Database: the first step is directed to industrial control system, according to related data, history case data
And expertise, establish typical fault case database, including phenomenon of the failure and failure cause.
Second step, fault characteristic information extract: the typical fault database based on foundation is mentioned using signal processing technology
Obtain time domain, time and frequency zone characteristic parameter data.
Third step, Feature Selection: the data set formed according to second step characteristic parameter takes Principle components analysis method,
The key data attribute variable of data set is extracted.
Neural network algorithm training study: 4th step utilizes neural network algorithm, to system failure sample set
Training is practised, the acquisition of fault knowledge is completed.
5th step, the representation of knowledge of system-oriented fault diagnosis: for the fault knowledge of acquisition, using frame and production
The mode that rule combines indicates specific malfunctioning module with fault tree framework method, indicates failure with production rule
Premise and as a result, obtaining the Rule Expression of fault diagnosis knowledge.
6th step, system failure propagation model building: firstly, capabilities map relationship of the building system task to completion task
Model；Then, mapping relations model of the structuring capacity to system comprising modules；Finally, event of the building based on module dependence
Hinder propagation model.
7th step, the fault location of knowledge-based inference: element and relationship type in analysis system fault knowledge determine
Qualified relation type determines inference rule, constructs knowledge graph, introduces road algebra and carries out statement and operation to derivation relationship, finally
The reasoning results confidence level is calculated, confidence level foundation is provided.
Further, time domain charactreristic parameter include root mean square, the flexure factor, the kurtosis factor, crest factor, the nargin factor,
Shape factor and the pulse factor, time and frequency zone characteristic parameter are the intrinsic mode functions value that signal passes through that empirical mode decomposition obtains.
Time and frequency zone characteristic parameter extraction method are as follows:
Determine data set x (t) local maximum and minimum in the time locating for original signal；
Make cubic spline interpolation according to maximum and minimum to determine x (t) between upper and lower envelope；
The local mean value m of x (t) is found out according to upper and lower envelopen(t), low-frequency component h representated by the envelope is removedn
(t)=x (t)-mn(t)；
With hn(t) it repeats the above steps as original signal, 2 standards until meeting IMF are denoted as c1(t) it is used as first
A IMF；
C is subtracted with original signal1(t) it is used as new signal r (t)=x (t)-c1(t), all steps before repeating are until institute
Some IMF are found out i.e.:Wherein ciIt represents in original signal and believes comprising different time scale features
Number；
Wherein, 2 standards of IMF:
(1) number of zero crossing is identical as extreme value number in entire signal or most differ is 1.
(2) at any point in time, the mean value envelope as defined in maximum envelope and minimum envelope is necessary
It is 0.
Further, the process of third step Feature Selection are as follows:
1. calculating the average value that data matrix X is respectively arranged, the column that each numerical value in data matrix X subtracts the number column are flat
Mean value；
2. the Eigen Covariance Matrix C of data matrix X is calculated, using formula: C=X*X '；
3. carrying out singular value decomposition to covariance matrix C obtained in the previous step, formula is as follows: C=U*D0* U ', wherein U be
Unitary matrice, D0It is characterized value diagonal matrix；
4. whitening matrix M is calculated, using formula:
5. calculate the Eigen Covariance Matrix C of data matrix Z, and find out Eigen Covariance characteristic value D and feature to
Measure V；
6. the diagonal matrix D of characteristic value is converted to column vector Dn, and carried out descending arrangement；
7. calculating column vector characteristic value summation, and calculate the ratio of add up each column vector summation and column vector characteristic value summation
Value, and then the size compared with the information contribution degree of setting；If kth adds up, column vector summation is greater than with column vector characteristic value summation
Setting information contribution degree h, then using K feature vector corresponding to it as new feature vector；Conversely, continuing growing cumulative time
Number, until the ratio of add up column vector summation and column vector characteristic value summation is greater than setting information contribution degree h；
8. generating the new data matrix after reducing dimension finally, input data matrix Z is projected in new feature vector
Xnew。
Further, the process of the 4th step neural network algorithm training study are as follows:
By the calculating to model is output and input, neuron output is obtained；
It is calculated downwards since output layer, modifies the weight of each layer connection, until obtaining anticipation error precision and classification can
Reliability.
Further, the 6th step specifically:
Step 6-1, the capabilities map relational model of building system task to completion task；It is various according to industrial control system
The Activities of task propose the ability need of completion task, building system task to capabilities map relational model, extractability
Characteristic and element, capacity-building database；
Step 6-2, the mapping relations model of structuring capacity to system comprising modules；
(1) industrial control system is divided into several modules according to the granularity of minimum replaceable units, establishes system
Architectural model；
After system is divided into several modules, logical connection, this module and module are formed between modules
Between connection relationship, form the architecture of industrial control system；
(2) dependency model of industrial control system architecture is established to the mapping relations of functions of modules according to ability；
For the module in industrial control system architecture, indicated with node, the logical connection between module is with son
The transition of task or state indicate, and by status change come the dependence between comprising modules in description system；
Step 6-3 constructs the fault- traverse technique based on module dependence；
Determine the impact factor for including in the node of dependency model；
Determine the influence of the fault mode and each fault mode of each impact factor to system running state；
Construct the fault- traverse technique based on system dependence.
Present invention is further described in detail with reference to the accompanying drawings and examples.
Embodiment
In conjunction with Fig. 1, a kind of Knowledge based engineering method for diagnosing faults of the present invention is comprised the steps of
The first step, typical fault Database: be directed to industrial control system, analysis related data, history case data,
It safeguards data and expertise, establishes typical fault case database, including phenomenon of the failure and failure cause.
Second step, fault characteristic information extract: feature extraction of the data acquisition system after signal processing obtains time domain, when-
Frequency domain character supplemental characteristic.Wherein time domain charactreristic parameter includes: root mean square, the flexure factor, the kurtosis factor, crest factor, nargin
The factor, shape factor, the pulse factor.Time and frequency zone characteristic parameter are as follows: signal passes through the IMF (eigen mode that empirical mode decomposition obtains
Functional value).
It is as follows that time domain charactreristic parameter extracts principle:
Root-mean-square error:
The flexure factor:
The kurtosis factor:
Crest factor:
The nargin factor:
Shape factor:
The pulse factor:
Wherein, N is signal observation frequency, and σ is standard deviation.
Time and frequency zone characteristic parameter extraction principle is as follows:
Empirical mode decomposition:
Firstly, determining data set x (t) local maximum and minimum in the time locating for original signal；Then, according to very big
Value and minimum make cubic spline interpolation to determine x (t) between upper and lower envelope；In next step, x is found out according to upper and lower envelope
(t) local mean value mn(t), low-frequency component h representated by the envelope is removedn(t)=x (t)-mn(t)；In next step, with hn
(t) as first three step of original signal repetition, 2 standards until meeting IMF are denoted as c1(t) it is used as first IMF；It is next
Step, subtracts c with original signal1(t) it is used as new signal r (t)=x (t)-c1(t), all steps before repeating are until all
IMF is found out i.e.:Wherein ciIt represents in original signal comprising different time scale characteristic signals.
2 standards of IMF:
The number of zero crossing is identical as extreme value number in entire signal or most differ is 1.
On at any point in time, the mean value envelope as defined in maximum envelope and minimum envelope is necessary for
0.Maximum envelope is defined by local maximum, and minimum envelope is defined by local minimum.
Third step, Feature Selection: the data set formed according to second step characteristic parameter takes Principle components analysis method,
The key data attribute variable of data set is extracted.Principle components analysis method is as shown in Fig. 2, calculating process are as follows:
1. calculating the average value that data X is respectively arranged, each numerical value in matrix X subtracts the column average value of the number column.
2. the Eigen Covariance Matrix C of data matrix X is calculated, using formula: C=X*X '.
3. carrying out singular value decomposition to covariance matrix C obtained in the previous step, formula is as follows: C=U*D0* U ', wherein U be
Unitary matrice, D0It is characterized value diagonal matrix.
4. whitening matrix M is calculated, using formula:
5. calculate the Eigen Covariance Matrix C of data matrix Z, and find out Eigen Covariance characteristic value D and feature to
Measure V.
6. the diagonal matrix D of characteristic value is converted to column vector Dn, and carried out descending arrangement.
7. calculating column vector characteristic value summation, and calculate the ratio of add up each column vector summation and column vector characteristic value summation
Value, and then the size compared with the information contribution degree of setting.If kth adds up, column vector summation is greater than with column vector characteristic value summation
Setting information contribution degree h, then using K feature vector corresponding to it as new feature vector.Conversely, continuing growing cumulative time
Number, until the ratio of add up column vector summation and column vector characteristic value summation is greater than setting information contribution degree h.
8. generating the new data matrix after reducing dimension finally, input data matrix Z is projected in new feature vector
Xnew。
4th step, neural network algorithm training study: the algorithm steps are broadly divided into two stages, calculate reality output and
Modify weight and threshold value.Calculating reality output is and to obtain neuron output by outputting and inputting the calculating of model；Modification
Weight and threshold value are calculated downwards since output layer, and the weight of each layer connection is modified, until obtaining anticipation error precision and dividing
Class confidence level.Learning training process is as shown in Figure 3.Algorithm improvement mode are as follows:
1. training process is improved
In general algorithm, learning rate η is acquired by search.Often it is defined as constant in the BP algorithm of standard, so
And a unalterable Optimal learning efficiency is not present in practice.Study or convergence speed of the learning rate to BP neural network
Degree influences very big.Learning rate η value is bigger, and the variation range of weight is bigger, when the weight of BP network in training process is distributed W
When tending towards stability, the excessive weight distribution W that will lead to of learning rate η generates oscillatory process, and system cannot restrain.
When learning rate η is too small, the modification amount of weight will reduce, and iterative process slows down that will to will lead to convergence process slack-off,
According to certain mathematical relationship existing for learning rate and training the number of iterations, it is assumed that exist according to learning rate and training the number of iterations linear
Functional relation.In the training process, it is assumed that maximum number of iterations tmax, initial learning rate is η (1), tmaxSecondary iteration junior scholar
Habit rate is η (tmax), each time after the completion of iteration, the expression formula of available learning rate is
According to convergence characteristic in convergence process, learning rate can be accelerated by designing suitable learning rate function, will not be caused
It is vibrated in learning process.Changing learning rate is that can effectively improve the number of iterations, is the main method for improving BP learning algorithm
With the important means for improving convergence speed.
2. momentum term is added
Momentum arithmetic is added in the back-propagation phase of BP neural network, therewith to this learning theory weighed value adjusting amount
Preceding adjustment amount is added, the weighed value adjusting amount as learning process next time.Such as following formula:
Wherein, n is frequency of training, mcFor momentum coefficient, work as mcWhen value is 0, weight declines according to gradient to be adjusted.Work as mc
When value is 1, new weight adjustment amount is identical as weight variable quantity size before, and gradient decline adjustment amount can be ignored.
It being added after momentum term, the corresponding network weight in error surface bottom is smaller,The decline of change of gradient amount, can obtain
To w (n) ≈ w (n+1), so that w (n+1)=0 be avoided to occur.Local Minimum can be fallen into avoid error.
3. introducing steepness factor
Because neuron output enters the saturation region of transfer function, there are flat regions on error two-dimensional surface.Such as
There is such case in fruit, needs to compress neuron and inputs only, it is made to jump out the saturation region of transfer function.Concrete thought is, in original
In transfer function introduce a steepness factor λ, make output be
Wherein dkTo export desired value, netkFor neural network load transfer function coefficient value, when E is close to zero, dk-ykStill
It is larger, illustrate that system is in flat region.When taking λ > 1, netkCoordinate has compressed λ times, so as to so that the biggish net of absolute value
Exit saturation value.When taking λ=1, transfer function restores to the original state, and has higher sensitivity for lesser net.It introduces steep
It is highly effective for the convergence rate for improving BP algorithm to spend the factor.
The representation of knowledge of system-oriented fault diagnosis: 5th step considers that industrial control system the Nomenclature Composition and Structure of Complexes is complicated, failure
Between associated with each other, the features such as knowledge quantity is big, diagnostic knowledge is indicated using the method that frame is combined with production rule, is used
Fault tree framework method indicates specific malfunctioning module, indicates with production rule the premise of failure and as a result, simplifies complicated
Representation of knowledge step specifies relevant fault relationship, improves representation of knowledge efficiency.Detailed process is as shown in Figure 4.First according to failure
Description enter Industry Control fault diagnosis system, matching rule and relevant inference mechanism according to system, by the number of input
It is matched and is compared according to the diagnostic knowledge with system database, and then judge the attribute of specific failure, in the process such as
Fruit has relevant subframe, and according to the subframe further progress diagnostic reasoning that rule starting needs, finally fault location is existed
Most basic component unit.
6th step, system failure propagation model building: firstly, capabilities map relationship of the building system task to completion task
Model.It is analyzed by the task to industrial control system, the energy of completion task is proposed according to the Activities of various tasks
Power demand, building system task to capabilities map relational model, extractability characteristic and element, capacity-building database.
Then, mapping relations model of the structuring capacity to system comprising modules.
Step 1: industrial control system is divided into several modules according to the granularity of minimum replaceable units, establishes system
The architectural model of system.
After system is divided into several modules, logic is formed in certain form in systems between modules and is connected
It connects, the connection relationship between this module and module is formed the architecture of industrial control system.
Step 2: according to ability to the mapping relations of functions of modules, the interdependent mould of industrial control system architecture is established
Type.
For the module in industrial control system architecture, indicated with node, the logical connection between module is with son
The transition of task or state indicate, and by status change come the dependence between comprising modules in description system.
Finally, fault- traverse technique of the building based on module dependence.
Step 1: it is analyzed according to industrial control system health determinants, determines the shadow for including in the node of dependency model
Ring the factor；
Step 2: the influence of the fault mode and each fault mode of each impact factor to system running state is analyzed；
Step 3: fault- traverse technique of the building based on system dependence is as shown in Figure 5.
7th step, the fault location of knowledge-based inference: core concept is that the concept of road algebra is introduced in knowledge graph, will
The process of knowledge reasoning is converted to the calculating process of road algebra, and main process is as shown in Figure 6.
Step 7-1: the element and relationship type for including in analytic induction industrial control system fault diagnosis knowledge determine limit
Determine relationship type, and formalization representation is carried out to relationship；
Step 7-2: according to the description logic of industrial control system fault diagnosis knowledge, inference rule is determined；
The qualified relation type for being usually used in expert system includes:
1. the generation or variation of a CAU b:a cause the generation or variation of b；
2. a PAR b:a is a part of b
3. a AKO b:a is one kind of b
4. the reverse-power of above three relationship: a HAP b, a HAK b, a CBY b
Basic reasoning has following two situation: first is that if there is relationship between points a andb, while point b and c it
Between there are relationships, then it is reasonable that there is also relationships between point a and c, by the obtained relationship type of these serial combinations
As Reasoning；Second is that determining the obtained pass of this parallel combined if there are two or more relationships between points a andb
Set type is known as parallel inference.
Step 7-3: carrying out element decomposition to existing knowledge, construct knowledge graph, introduces road algebra and carries out table to derivation relationship
It states and operation；
In knowledge graph K (C, R), π=(c is enabled0,c1,...,ck) it is the road that a length is k, wherein ci∈ C, ri=
(ci,ci+1)∈R.One road is actually a relational sequence, each relationship either arc (ci,ci+1) or side (ci,
ci+1).If it is assumed that the sub- road for being 2 to all length allows to make multiplying, then r=(c0,ck) type (type) about road π
It can be defined as
It enablesIndicate from concept i to concept j length be at most 2 all roads, when the integrated result of K previous step relationship, when
And if only if enablingAndWhen, just obtain the figure K with r=(i, j)2(C,R2),
For empty set.In general, if the result K that n step relationship integratesnIt, can be recursively in K if expressionnA step is used on (n > 2)
Relationship integrates to reach.
Step 7-4: calculating the reasoning results confidence level, provides confidence level foundation.
Confidence level of the concept to indicate reasoning for introducing weight, to each relationship add a style weight ω (0≤
ω≤1) to indicate a possibility that relationship is set up, and indicate to push away with normalized using quadratic sum evolution in reasoning
A possibility that relationship obtained after reason.
1. parallel inference
If aR1b(ω1), aR1b(ω2), then still having relationship R between a and b1, but the weight between a and b
This inference rule explanation, parallel inference can improve the confidence level of derivation relationship.
2. Reasoning
If aR1b(ω1), bR2c(ω2), then existing relationship between a and cWeight is defined simultaneously,
In formula, symbol " * " indicates convolution；
ω=ω1*ω2
Above-mentioned inference rule explanation, Reasoning can reduce the confidence level of derivation relationship.
Using above-mentioned inference rule, we had both obtained the relationship of reasoning, had also obtained the reasoning results confidence level simultaneously.
Claims (6)
1. a kind of Knowledge based engineering method for diagnosing faults, which comprises the following steps:
Typical fault Database: the first step is directed to industrial control system, establishes typical fault case database；
Second step, fault characteristic information extract: the typical fault database based on foundation is extracted using signal processing technology
To time domain charactreristic parameter, time and frequency zone characteristic parameter；
Third step, Feature Selection: the data set formed according to characteristic parameter takes Principle components analysis method, by data set
Key data attribute variable extracts；
Neural network algorithm training study: 4th step utilizes neural network algorithm, carries out study instruction to system failure sample set
Practice, completes the acquisition of fault knowledge；
5th step, the representation of knowledge of system-oriented fault diagnosis: for the fault knowledge of acquisition, using frame and production rule
The mode combined indicates specific malfunctioning module with fault tree framework method, and the premise of failure is indicated with production rule
With as a result, obtaining the Rule Expression of fault diagnosis knowledge；
6th step, system failure propagation model building: firstly, capabilities map relationship mould of the building system task to completion task
Type；Then, mapping relations model of the structuring capacity to system comprising modules；Finally, failure of the building based on module dependence
Propagation model；
7th step, the fault location of knowledge-based inference: element and relationship type in analysis system fault knowledge are determined and are limited
Relationship type determines inference rule, constructs knowledge graph, introduces road algebra and carries out statement and operation to derivation relationship, to reasoning knot
Fruit confidence level is calculated, and confidence level foundation is provided, and realizes the positioning of the system failure.
2. Knowledge based engineering method for diagnosing faults according to claim 1, which is characterized in that in second step, temporal signatures
Parameter includes root mean square, the flexure factor, the kurtosis factor, crest factor, the nargin factor, shape factor and the pulse factor, time and frequency zone
Characteristic parameter is the intrinsic mode functions value that signal passes through that empirical mode decomposition obtains.
3. Knowledge based engineering method for diagnosing faults according to claim 2, which is characterized in that in second step, time and frequency zone is special
Levy parameter extracting method are as follows:
Determine data set x (t) local maximum and minimum in the time locating for original signal；
Make cubic spline interpolation according to maximum and minimum to determine x (t) between upper and lower envelope；
The local mean value m of x (t) is found out according to upper and lower envelopen(t), low-frequency component h representated by the envelope is removedn(t)=x
(t)-mn(t)；
With hn(t) it repeats the above steps as original signal, 2 standards until meeting IMF are denoted as c1(t) it is used as first
IMF；
C is subtracted with original signal1(t) it is used as new signal r (t)=x (t)-c1(t), all steps before repeating are until all
IMF is found out i.e.:Wherein ciIt represents in original signal comprising different time scale characteristic signals；
Wherein, 2 standards of IMF:
(1) number of zero crossing is identical as extreme value number in entire signal or most differ is 1.
(2) at any point in time, the mean value envelope as defined in maximum envelope and minimum envelope is necessary for 0.
4. Knowledge based engineering method for diagnosing faults according to claim 1, which is characterized in that the mistake of third step Feature Selection
Journey are as follows:
1. calculating the average value that data matrix X is respectively arranged, each numerical value in data matrix X subtracts the column average of the number column
Value；
2. the Eigen Covariance Matrix C of data matrix X is calculated, using formula: C=X*X '；
3. carrying out singular value decomposition to covariance matrix C obtained in the previous step, formula is as follows: C=U*D0* U ', wherein U is tenth of the twelve Earthly Branches square
Battle array, D0It is characterized value diagonal matrix；
4. whitening matrix M is calculated, using formula:
5. calculating the Eigen Covariance Matrix C of data matrix Z, and find out the characteristic value D and feature vector V of Eigen Covariance；
6. the diagonal matrix D of characteristic value is converted to column vector Dn, and carried out descending arrangement；
7. calculating column vector characteristic value summation, and the ratio of add up each column vector summation and column vector characteristic value summation is calculated,
And then the size compared with the information contribution degree of setting；If kth adds up, column vector summation is greater than with column vector characteristic value summation sets
Confidence ceases contribution degree h, then using K feature vector corresponding to it as new feature vector；Conversely, continuing growing cumulative time
Number, until the ratio of add up column vector summation and column vector characteristic value summation is greater than setting information contribution degree h；
8. generating the new data matrix X after reducing dimension finally, input data matrix Z is projected in new feature vectornew。
5. Knowledge based engineering method for diagnosing faults according to claim 1, which is characterized in that the 4th step neural network algorithm
The process of training study are as follows:
By the calculating to model is output and input, neuron output is obtained；
It is calculated downwards since output layer, modifies the weight of each layer connection, until obtaining anticipation error precision and classification confidence level.
6. Knowledge based engineering method for diagnosing faults according to claim 1, which is characterized in that the 6th step specifically:
Step 6-1, the capabilities map relational model of building system task to completion task；According to the various tasks of industrial control system
Activities propose the ability need of completion task, building system task is to capabilities map relational model, extractability characteristic
And element, capacity-building database；
Step 6-2, the mapping relations model of structuring capacity to system comprising modules；
(1) industrial control system is divided into several modules according to the granularity of minimum replaceable units, establishes the system of system
Structural model；
After system is divided into several modules, logical connection is formed between modules, between this module and module
Connection relationship, form the architecture of industrial control system；
(2) dependency model of industrial control system architecture is established to the mapping relations of functions of modules according to ability；
For the module in industrial control system architecture, indicated with node, the logical connection between module is with subtask
Or the transition of state indicate, and by status change come the dependence between comprising modules in description system；
Step 6-3 constructs the fault- traverse technique based on module dependence；
Determine the impact factor for including in the node of dependency model；
Determine the influence of the fault mode and each fault mode of each impact factor to system running state；
Construct the fault- traverse technique based on system dependence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811220751.9A CN109491816B (en) | 2018-10-19 | 2018-10-19 | Knowledge-based fault diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811220751.9A CN109491816B (en) | 2018-10-19 | 2018-10-19 | Knowledge-based fault diagnosis method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109491816A true CN109491816A (en) | 2019-03-19 |
| CN109491816B CN109491816B (en) | 2022-05-27 |
Family
ID=65692084
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811220751.9A Active CN109491816B (en) | 2018-10-19 | 2018-10-19 | Knowledge-based fault diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109491816B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674893A (en) * | 2019-10-30 | 2020-01-10 | 江苏方天电力技术有限公司 | Self-adaptive correction method for diagnosis experience in rotary machine fault diagnosis knowledge base |
| CN110782036A (en) * | 2019-07-01 | 2020-02-11 | 烟台宏远氧业股份有限公司 | Big data analysis system of hyperbaric oxygen chamber |
| CN110836783A (en) * | 2019-10-29 | 2020-02-25 | 杭州电子科技大学 | GA-BP magnetic suspension train fault detection method based on threshold judgment |
| CN110989558A (en) * | 2019-12-19 | 2020-04-10 | 浙江中控技术股份有限公司 | Fault diagnosis item processing method and system |
| CN111650898A (en) * | 2020-05-13 | 2020-09-11 | 大唐七台河发电有限责任公司 | Distributed control system and method with high fault tolerance performance |
| CN111682960A (en) * | 2020-05-14 | 2020-09-18 | 深圳市有方科技股份有限公司 | Fault diagnosis method and device for Internet of things network and equipment |
| CN112036568A (en) * | 2020-07-09 | 2020-12-04 | 中国人民解放军海军工程大学 | Intelligent diagnosis method for damage fault of primary loop coolant system of nuclear power plant |
| CN112345276A (en) * | 2020-11-10 | 2021-02-09 | 北京交通大学 | State evaluation and prediction system for key components of medium-speed maglev train |
| CN112418450A (en) * | 2020-10-30 | 2021-02-26 | 济南浪潮高新科技投资发展有限公司 | Equipment predictive maintenance method based on multi-mode machine learning |
| CN112631235A (en) * | 2020-11-24 | 2021-04-09 | 北京妙微科技有限公司 | Iron tower remote monitoring and fault diagnosis system based on improved SOM network |
| CN112990275A (en) * | 2021-02-20 | 2021-06-18 | 长春工业大学 | High-speed train running gear system fault diagnosis method based on semi-quantitative information fusion |
| CN112988843A (en) * | 2021-03-26 | 2021-06-18 | 桂林电子科技大学 | SMT chip mounter fault management and diagnosis system based on SQL Server database |
| CN113110402A (en) * | 2021-05-24 | 2021-07-13 | 浙江大学 | Knowledge and data driven large-scale industrial system distributed state monitoring method |
| CN113393211A (en) * | 2021-06-22 | 2021-09-14 | 柳州市太启机电工程有限公司 | Method and system for intelligently improving automatic production efficiency |
| CN113640699A (en) * | 2021-10-14 | 2021-11-12 | 南京国铁电气有限责任公司 | Fault judgment method, system and equipment for microcomputer control type alternating current and direct current power supply system |
| CN113837161A (en) * | 2021-11-29 | 2021-12-24 | 广东东软学院 | Identity recognition method, device and equipment based on image recognition |
| CN115563549A (en) * | 2022-10-27 | 2023-01-03 | 武汉理工大学 | Welding defect cause diagnosis method and system and electronic equipment |
| CN116360388A (en) * | 2023-01-18 | 2023-06-30 | 北京控制工程研究所 | Reasoning method and device of performance-fault relation map based on graph neural network |
| CN117668751A (en) * | 2023-11-30 | 2024-03-08 | 广东一业建设股份有限公司 | High-low voltage power system fault diagnosis method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7720779B1 (en) * | 2006-01-23 | 2010-05-18 | Quantum Leap Research, Inc. | Extensible bayesian network editor with inferencing capabilities |
| CN104331543A (en) * | 2014-10-17 | 2015-02-04 | 中国船舶重工集团公司第七一二研究所 | Fault diagnostic expert system for marine electrical propulsion system and establishing method thereof |
| CN107657250A (en) * | 2017-10-30 | 2018-02-02 | 四川理工学院 | Bearing fault detection and localization method and detection location model realize system and method |
| CN108549734A (en) * | 2018-01-30 | 2018-09-18 | 南京航空航天大学 | TFM three-dimensional information stream modeling methods based on systematic functional structrue |
-
2018
- 2018-10-19 CN CN201811220751.9A patent/CN109491816B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7720779B1 (en) * | 2006-01-23 | 2010-05-18 | Quantum Leap Research, Inc. | Extensible bayesian network editor with inferencing capabilities |
| CN104331543A (en) * | 2014-10-17 | 2015-02-04 | 中国船舶重工集团公司第七一二研究所 | Fault diagnostic expert system for marine electrical propulsion system and establishing method thereof |
| CN107657250A (en) * | 2017-10-30 | 2018-02-02 | 四川理工学院 | Bearing fault detection and localization method and detection location model realize system and method |
| CN108549734A (en) * | 2018-01-30 | 2018-09-18 | 南京航空航天大学 | TFM three-dimensional information stream modeling methods based on systematic functional structrue |
Non-Patent Citations (1)
| Title |
|---|
| 刘兴超: "电液伺服系统的智能故障检测与诊断的研究", 《中国优秀硕士学位论文全文数据库（工程科技Ⅱ专辑）》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782036A (en) * | 2019-07-01 | 2020-02-11 | 烟台宏远氧业股份有限公司 | Big data analysis system of hyperbaric oxygen chamber |
| CN110836783A (en) * | 2019-10-29 | 2020-02-25 | 杭州电子科技大学 | GA-BP magnetic suspension train fault detection method based on threshold judgment |
| CN110674893B (en) * | 2019-10-30 | 2022-07-08 | 江苏方天电力技术有限公司 | Self-adaptive correction method for diagnosis experience in rotary machine fault diagnosis knowledge base |
| CN110674893A (en) * | 2019-10-30 | 2020-01-10 | 江苏方天电力技术有限公司 | Self-adaptive correction method for diagnosis experience in rotary machine fault diagnosis knowledge base |
| CN110989558A (en) * | 2019-12-19 | 2020-04-10 | 浙江中控技术股份有限公司 | Fault diagnosis item processing method and system |
| CN111650898B (en) * | 2020-05-13 | 2023-10-20 | 大唐七台河发电有限责任公司 | Distributed control system and method with high fault tolerance performance |
| CN111650898A (en) * | 2020-05-13 | 2020-09-11 | 大唐七台河发电有限责任公司 | Distributed control system and method with high fault tolerance performance |
| CN111682960A (en) * | 2020-05-14 | 2020-09-18 | 深圳市有方科技股份有限公司 | Fault diagnosis method and device for Internet of things network and equipment |
| CN112036568A (en) * | 2020-07-09 | 2020-12-04 | 中国人民解放军海军工程大学 | Intelligent diagnosis method for damage fault of primary loop coolant system of nuclear power plant |
| CN112036568B (en) * | 2020-07-09 | 2023-10-17 | 中国人民解放军海军工程大学 | Intelligent diagnosis method for damage faults of primary loop coolant system of nuclear power device |
| CN112418450A (en) * | 2020-10-30 | 2021-02-26 | 济南浪潮高新科技投资发展有限公司 | Equipment predictive maintenance method based on multi-mode machine learning |
| CN112345276A (en) * | 2020-11-10 | 2021-02-09 | 北京交通大学 | State evaluation and prediction system for key components of medium-speed maglev train |
| CN112631235A (en) * | 2020-11-24 | 2021-04-09 | 北京妙微科技有限公司 | Iron tower remote monitoring and fault diagnosis system based on improved SOM network |
| CN112990275A (en) * | 2021-02-20 | 2021-06-18 | 长春工业大学 | High-speed train running gear system fault diagnosis method based on semi-quantitative information fusion |
| CN112990275B (en) * | 2021-02-20 | 2022-08-26 | 长春工业大学 | High-speed train running gear system fault diagnosis method based on semi-quantitative information fusion |
| CN112988843A (en) * | 2021-03-26 | 2021-06-18 | 桂林电子科技大学 | SMT chip mounter fault management and diagnosis system based on SQL Server database |
| CN113110402A (en) * | 2021-05-24 | 2021-07-13 | 浙江大学 | Knowledge and data driven large-scale industrial system distributed state monitoring method |
| CN113110402B (en) * | 2021-05-24 | 2022-04-01 | 浙江大学 | Knowledge and data driven large-scale industrial system distributed state monitoring method |
| CN113393211B (en) * | 2021-06-22 | 2022-12-09 | 柳州市太启机电工程有限公司 | Method and system for intelligently improving automatic production efficiency |
| CN113393211A (en) * | 2021-06-22 | 2021-09-14 | 柳州市太启机电工程有限公司 | Method and system for intelligently improving automatic production efficiency |
| CN113640699B (en) * | 2021-10-14 | 2021-12-24 | 南京国铁电气有限责任公司 | Fault judgment method, system and equipment for microcomputer control type alternating current and direct current power supply system |
| CN113640699A (en) * | 2021-10-14 | 2021-11-12 | 南京国铁电气有限责任公司 | Fault judgment method, system and equipment for microcomputer control type alternating current and direct current power supply system |
| CN113837161B (en) * | 2021-11-29 | 2022-02-22 | 广东东软学院 | Identity recognition method, device and equipment based on image recognition |
| CN113837161A (en) * | 2021-11-29 | 2021-12-24 | 广东东软学院 | Identity recognition method, device and equipment based on image recognition |
| CN115563549A (en) * | 2022-10-27 | 2023-01-03 | 武汉理工大学 | Welding defect cause diagnosis method and system and electronic equipment |
| CN115563549B (en) * | 2022-10-27 | 2023-10-20 | 武汉理工大学 | Welding defect cause diagnosis method and system and electronic equipment |
| CN116360388A (en) * | 2023-01-18 | 2023-06-30 | 北京控制工程研究所 | Reasoning method and device of performance-fault relation map based on graph neural network |
| CN116360388B (en) * | 2023-01-18 | 2023-09-08 | 北京控制工程研究所 | Reasoning method and device of performance-fault relation map based on graph neural network |
| CN117668751A (en) * | 2023-11-30 | 2024-03-08 | 广东一业建设股份有限公司 | High-low voltage power system fault diagnosis method and device |
| CN117668751B (en) * | 2023-11-30 | 2024-04-26 | 广东一业建设股份有限公司 | High-low voltage power system fault diagnosis method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109491816B (en) | 2022-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109491816A (en) | Knowledge based engineering method for diagnosing faults | |
| CN108448610B (en) | Short-term wind power prediction method based on deep learning | |
| CN104536412B (en) | Photoetching procedure dynamic scheduling method based on index forecasting and solution similarity analysis | |
| CN107220734A (en) | CNC Lathe Turning process Energy Consumption Prediction System based on decision tree | |
| CN103778304B (en) | A kind of method for designing of driving bridge for motor vehicle | |
| CN103745273B (en) | Semiconductor fabrication process multi-performance prediction method | |
| CN110135637A (en) | Micro-capacitance sensor short-term load forecasting method based on shot and long term memory and adaptive boosting | |
| CN103886203B (en) | Automatic modeling system and method based on index prediction | |
| CN109407654A (en) | A kind of non-linear causality analysis method of industrial data based on sparse depth neural network | |
| CN112086958B (en) | Power transmission network extension planning method based on multi-step backtracking reinforcement learning algorithm | |
| Lin et al. | Machine learning templates for QCD factorization in the search for physics beyond the standard model | |
| CN111127246A (en) | Intelligent prediction method for transmission line engineering cost | |
| CN108346293A (en) | A kind of arithmetic for real-time traffic flow Forecasting Approach for Short-term | |
| CN109101712B (en) | Product model design system and method based on graph network | |
| CN107451230A (en) | A kind of answering method and question answering system | |
| CN107480141A (en) | It is a kind of that allocating method is aided in based on the software defect of text and developer's liveness | |
| CN110826237B (en) | Wind power equipment reliability analysis method and device based on Bayesian belief network | |
| CN106649479A (en) | Probability graph-based transformer state association rule mining method | |
| CN106779346A (en) | A kind of Forecasting Methodology of monthly power consumption | |
| CN104732067A (en) | Industrial process modeling forecasting method oriented at flow object | |
| El Bourakadi et al. | Multi-agent system based sequential energy management strategy for Micro-Grid using optimal weighted regularized extreme learning machine and decision tree | |
| Du et al. | Applying deep convolutional neural network for fast security assessment with N-1 contingency | |
| Kutschenreiter-Praszkiewicz | Application of artificial neural network for determination of standard time in machining | |
| Abiyev | Fuzzy wavelet neural network for prediction of electricity consumption | |
| CN104182854A (en) | Mixed energy consumption measuring method for enterprise energy management system |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Wang Kai Inventor after: Lu Zhongyang Inventor after: Jiang Tingting Inventor after: Yang Zhao Inventor after: Fang Jianyong Inventor after: Zhang Junwei Inventor after: Zhou Bin Inventor after: Wang Li Inventor after: Li Yin Inventor after: Chen Shanhao Inventor before: Jiang Tingting Inventor before: Lu Zhongyang Inventor before: Fang Jianyong Inventor before: Zhou Bin Inventor before: Li Yin Inventor before: Yang Zhao Inventor before: Wang Li Inventor before: Zhang Junwei Inventor before: Chen Shanhao Inventor before: Wang Kai |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 222001 No.18 Shenghu Road, Lianyungang City, Jiangsu Province Patentee after: The 716th Research Institute of China Shipbuilding Corp. Address before: 222001 No.18 Shenghu Road, Lianyungang City, Jiangsu Province Patentee before: 716TH RESEARCH INSTITUTE OF CHINA SHIPBUILDING INDUSTRY Corp. |