CN102830341A - Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller) - Google Patents

Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller) Download PDF

Info

Publication number
CN102830341A
CN102830341A CN2012103099306A CN201210309930A CN102830341A CN 102830341 A CN102830341 A CN 102830341A CN 2012103099306 A CN2012103099306 A CN 2012103099306A CN 201210309930 A CN201210309930 A CN 201210309930A CN 102830341 A CN102830341 A CN 102830341A
Authority
CN
China
Prior art keywords
circuit
performance parameters
cmac
circuit performance
power electronic
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.)
Pending
Application number
CN2012103099306A
Other languages
Chinese (zh)
Inventor
林华
王友仁
姜媛媛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN2012103099306A priority Critical patent/CN102830341A/en
Publication of CN102830341A publication Critical patent/CN102830341A/en
Pending legal-status Critical Current

Links

Images

Abstract

The invention discloses an online intelligent fault prediction method for a power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller) and belongs to the technical field of fault testing for power electronic circuits. The method includes: monitoring node signals in real time, performing wavelet denoising to obtain a fault feature sample, extracting circuit performance parameters, and building an RS-CMAC model to predict a time sequence of the circuit performance parameters in future time. According to the rough set theory and the CMAC model, input data of the CMAC model are simplified by a rough set data analysis method, and the efficiency in analyzing faults of the power electronic circuit is improved.

Description

Power electronic circuit on-line intelligence failure prediction method based on RS-CMAC
Technical field
The invention discloses power electronic circuit on-line intelligence failure prediction method, belong to the technical field of power electronic circuit fault test based on RS-CMAC.
Background technology
Power circuit in the airborne power electric device all is to be made up of power component; When placing certain space environment, its element is except that suffering high frequency startup/shut-down operation and overvoltage, mistake flow operation, and its performance also very easily receives the influence of extraneous mechanical pressure (impact), EMI, environment temperature/humidity, saline and alkaline equal stress; This may cause the variation of device parameters; Surpass the scope that allows more greatly if change, then tended to cause the deterioration (for example: waveform quality variation, THD increase etc.) of circuit output performance; Even cause output function to lose efficacy, serious threat flight safety and the final smooth execution that influences aerial mission.Airborne power electric device must possess high reliability, and the situation and in time not detecting if power electric device breaks down then may cause flight control system to lose efficacy, have a strong impact on the safety of aircraft itself, and consequence is hardly imaginable.The high reliability of airborne power unit, strong viability, independently diagnosis and health control are to press for the technical barrier of researching and solving; Also received the very big attention of each military power in the world, be one of present international academic research forward position and focus always.
At present, the failure prediction method to power electronic circuit mainly divides three kinds: (1) is based on the method for fault physical model; (2) based on the method for interior building " damage scale "; (3) based on the method for data-driven.
Based on the method for model, need set up system's accurate model and understand system works mechanism in depth.Forecasting Methodology based on the damage scale; Be to one or more failure mechanisms; With the produced expected life of the identical technological process of the monitored product product shorter than monitored object; Design omen unit and main device, circuit, the system integration make it lose efficacy before main circuit or thrashing prerequisite by certain mechanism together, thereby early warning are provided for the inefficacy of host's Circuits and Systems.Based on the method for data-driven, the data that promptly on-site supervision obtained are carried out unusual and trend detects or mode detection, confirm the health status of system, and the usage trend analysis result comes the time of failure of estimating system then.This method need not to understand the inside circuit physical arrangement, is applicable to the complication system prediction, and applied range makes the real-time online prediction become possibility.
Present stage, mostly the power electronic circuit failure prediction is the prediction of key components in the circuit, selects for use the fault signature parameter to be mostly the parameter of components and parts, and to the whole failure prediction research of circuit seldom.
In recent years, artificial intelligence develops swift and violent in the prediction field.In conjunction with existing document and patented technology, in the failure prediction of electronic circuit, artificial intelligence approach commonly used comprises neural network, expert system, gray system, particle filter, regression tree, SVMs etc.The characteristics of artificial intelligence approach be learning ability strong, need not to set up precise math model; Its main deficiency is the bottleneck problem that has knowledge acquisition, and knowledge is difficult to safeguard, there are in various degree limitation in " shot array " and " infinite recurrence ", the adaptive ability of system own and the learning ability of knowledge etc. has influenced the accuracy of failure prediction greatly.Wherein, neural network there is not restriction to observation sequence, and it almost can be analyzed all time serieses.Particularly the non-linear mapping capability of neural network makes it can be widely used in the NLS prediction.
Rough set (Rough Sets; RS) theory is proposed in nineteen eighty-two by Polish scholar Pawlak; The brand-new mathematical tool that has imperfection and uncertain information as a kind of portrayal; Become a new academic focus of artificial intelligence field, utilized it to obtain many achievements abroad, domestic research still is in the starting stage.Rough Set Data Analysis ((Rough Sets Date Analysis, RSDA) be a kind of be the basis with RS, analyze correlativity and dependent a kind of notation method between the data.Utilize it can be, thereby predict and make a strategic decision from extracting data rule.
(Cerebellar Model Articulation Controller CMAC) is a kind of controller that imitates the brain connection that Albus proposes to cerebellum Model Neural CMAC.Its pace of learning is fast, and has overcome the local optimum problem of BP network.In recent years, CMAC is widely used in fields such as real-time control, pattern-recognition.CMAC is a feedforward neural network, is a kind of table lookup type adaptive neural network of expressing the complex nonlinear function, because based on part study, so each power of revising is seldom, pace of learning is fast, is applicable to real-time estimate; Have the continuous analog I/O capability, generalization ability is superior to common neural network, therefore has better non-linear approximation capability, is suitable for the non-linear characteristics of power electronic circuit complex and dynamic.Yet because the base of common CMAC model equals 1, so its generalization ability is affected.After serve the researcher and proposed Fuzzy CMAC.No matter but be basic CMAC model, or the fuzzy CMAC model, when sample input dimension was big, the calculated amount that needs was surprising.
At present, in conjunction with to adopting cerebellum neural network prediction power electronic circuit fault, have the big problem of calculated amount, the method, the cerebellum neural net prediction method that also do not have the scholar to propose the combining rough set data analysis come predicted power electronic circuit fault.
Summary of the invention
Technical matters to be solved by this invention is to the deficiency of above-mentioned background technology, and the power electronic circuit on-line intelligence failure prediction method based on RS-CMAC is provided.
The present invention adopts following technical scheme for realizing the foregoing invention purpose:
Power electronic circuit on-line intelligence failure prediction method based on RS-CMAC comprises the steps:
Step 1 is by treating that power scale selects measured node from electronic circuit, the voltage of the said measured node of on-line monitoring, current signal;
Step 2 is done wavelet threshold to voltage, the current signal of the measured node described in the step 1 and is handled, and obtains the fault signature sample;
Step 3 from the described fault signature sample extraction of step 2 circuit performance parameters, obtains the circuit performance parameters vector;
Step 4 is carried out trend prediction in conjunction with RS theory and CMAC Forecasting Methodology to the described circuit performance parameters vector of step 3, obtains the following time series of circuit performance parameters constantly;
Step 5, in sampling period of every mistake, repeating step 1 obtains the time series of circuit performance parameters to step 4 according to the circuit performance parameters vector of real-time update;
Step 6 is utilized the time series counting circuit health indicator of the said circuit performance parameters of step 5, and the circuit health indicator that relatively calculates and circuit healthy threshold value, the health status of decision circuit then.
In the said power electronic circuit on-line intelligence failure prediction method based on RS-CMAC, the practical implementation method of step 4 is: each circuit performance parameters in the circuit performance parameters vector is set up a RS-CMAC model; Obtain the mapping ruler storehouse of circuit performance parameters according to the RS theory; According to the following time series of circuit performance parameters constantly of mapping ruler storehouse prediction.
The present invention adopts technique scheme, has following beneficial effect: combined rough set theory, and the CMAC model, utilize the Rough Set Data Analysis method to simplify the input data of CMAC model, improved the efficient of power electronic circuit fault analysis.
Description of drawings
Fig. 1 is the schematic flow sheet based on the power electronic circuit on-line intelligence failure prediction method of RS-CMAC.
Fig. 2 is the process flow diagram of wavelet threshold denoising.
Fig. 3 is RS-CMAC model mapping synoptic diagram.
Fig. 4 is the synoptic diagram of buck circuit.
Embodiment
Be elaborated below in conjunction with the technical scheme of accompanying drawing to invention:
As shown in Figure 1, utilize power electronic circuit on-line intelligence failure prediction method prediction buck fault as shown in Figure 4 based on RS-CMAC.Comprise the steps:
Step 1 is by treating that power scale selects measured node from electronic circuit, the voltage of the said measured node of on-line monitoring (input voltage U i, output voltage U o) signal, electric current (output current I o) signal.
Step 2 is utilized wavelet threshold denoising method as shown in Figure 2, to the U of input voltage described in the step 1 i, output voltage U o, output current I oDo wavelet threshold and handle, obtain fault signature sample U ' i, U ' o, I ' o
Step 3 from the described fault signature sample extraction of step 2 circuit performance parameters, obtains circuit performance parameters vector { x 1, x 2..., x i..., x p, i=1,2 ..., p, p are the number of circuit performance parameters.The circuit performance parameters of extracting is the parameter of reflection circuit health status; Comprise: output average voltage, output average current, output power, power input, efficient, output voltage ripple, output voltage ripple compare etc., and the number of physical circuit performance parameter depends on concrete power electronic circuit.
Be extracted in this example in the circuit malfunction process change tangible output voltage ripple than δ as buck circuit performance parameter.Wherein,
Figure BDA00002063014300041
U PPBe output voltage ripple.By fault signature sample U ' i, U ' o, I ' oObtain the time series δ of output voltage ripple than δ 1, δ 2... δ n
Step 4 is carried out trend prediction in conjunction with RS theory and CMAC Forecasting Methodology to the described circuit performance parameters vector of step 3, obtains following circuit performance parameters vector constantly.As shown in Figure 3, to circuit performance parameters vector { x 1, x 2..., x i..., x pIn each circuit performance parameters set up a RS-CMAC model.Each RS-CMAC model comprises 4 layers of structure, and its mapping algorithm is following:
The 1st layer is input layer: a period of time sequence of input circuit performance parameter xi
Figure BDA00002063014300042
; J=l; 2;, m.If ground floor has m node, j time series component x of j the corresponding input vector of node Ij
In the 2nd layer, each node is represented a rule, these rules are based on rough set theory data set is carried out yojan after, carry out then that Rule Extraction obtains.Be connected the rule that depends on that node is represented between this node layer and ground floor and the 3rd node layer, that is: each node of the second layer and represent the condition part of rule being connected of ground floor node; With the conclusion part of representing rule being connected of the 3rd node layer.If this node is represented complex rule, it possibly link to each other with n node in the ground floor so, and n < m (m is the total node number of ground floor).The action function of this node layer is α i=x I1X I2∧ x Ic(1≤i≤k).Wherein, x I1X I2∧ x IcThe input value of the ground floor node that is connected with this node of expression, c is the number of the ground floor node that is connected with this node, k is regular bar number.
The 3rd layer be output can not differentiate division because the rule of second layer node after representing to simplify, therefore the node of this layer can be corresponding with the 3rd layer, and 0 or a plurality of node are continuous, the neuronic effect formula of this layer is:
y s &prime; = &Sigma; i = 1 q w si &alpha; i ( 1 &le; s &le; l ) - - - ( 1 ) ,
Wherein, l is the 3rd layer a node number, and q is the node number of the second layer, w SiFor connecting weights, its initial value is redefined for each regular degree of confidence.
The 4th layer, the 4th layer is output layer.This node layer is output as:
y=∑w iy′ i （2）。
In this one deck, the k that only is activated neuronic connection power w iObtain revising (i=1,2 ..., k).Learning algorithm adopts the BP algorithm.When network was input as x, establishing y was the actual output of CMAC, y dBe corresponding desired output, the error objective function does
Figure BDA00002063014300052
Thereby
&PartialD; E p &PartialD; w i = - ( y - y d ) v i , w i ( k + 1 ) = w i ( k ) - &beta; &PartialD; E p &PartialD; w i - - - ( 3 )
Wherein, β is a learning rate, when training, is a constant, and value (scope 0～1) is different and different with training speed.
With circuit performance parameters x i(i=1,2 ..., p) continuous m time series { x constantly I1, x I2..., x ImThe input of RS-CMAC model; Utilize RS-CMAC to { x I1, x I2..., x ImCarry out forward direction k step prediction, obtain the performance parameter { x of following a period of time circuit I, m+1, x I, m+2..., x I, m+k.
And in the present example, with 0 ~ n δ constantly 1, δ 2... δ nAs the input of RS-CMAC network, be test data.The RS-CMAC network has four layers, and ground floor is an input layer, and the node number is made as n.The second layer is divided according to the relation of can not distinguishing respectively with after n the input quantity discretize, obtains r iIndividual different value.Defining the neuronic action function of this layer is the Gauss function
Figure BDA00002063014300061
Wherein, i=1,2 ..., n; J=1,2 ..., r iThe output number of this layer does
Figure BDA00002063014300062
The 3rd layer of relevance grade μ that calculates every rule lIf the relevance grade μ of rule l>=γ, then this node is output as: v i=α iμ l, wherein, α iThe action function α that representes this node layer i=x I1X I2Λ x Ic(1≤i≤k), γ is selected constant.Otherwise delete this node.The neuronic number of this layer does
Figure BDA00002063014300063
The 4th layer of weight space w that is connected for selected l bar rule activation i, the l that is activated neuronic connection power w iObtain revising, learning algorithm adopts the BP algorithm.
Figure BDA00002063014300064
J=1,2 ..., k is forward direction k step prediction result, i.e. δ N+1, δ N+2... δ N+k
Step 5, in sampling period of every mistake, repeating step 1 obtains the time series of circuit performance parameters to step 4 according to the circuit performance parameters vector of real-time update.Off-line training RS-CMAC network, training data are fault simulation data or fault experience data.
Step 6 is utilized the time series counting circuit health indicator of the said circuit performance parameters of step 5, and circuit health indicator that relatively calculates and the healthy threshold value of circuit, judges then, the health status of output circuit.If the output voltage ripple that allows during circuit operate as normal ratio is 5%, then can be by δ constantly in future N+1, δ N+2... δ N+kCompare with threshold value 5%, work as δ N+1, δ N+2... δ N+kDuring greater than threshold value 5%, expression buck fault; Otherwise the buck circuit is normal.
In sum, the present invention has combined rough set theory, and the CMAC model, utilizes the Rough Set Data Analysis method to simplify the input data of CMAC model, has improved the efficient of power electronic circuit fault analysis.Above-mentioned buck embodiment of circuit is merely a specific embodiment of the present invention, and according to the complexity of power electronic circuit in the practice, the circuit performance parameters number of selection is different, but the method for prediction fault is identical.Therefore, the described power electronic circuit in every this area all can carry out failure prediction with method of the present invention.

Claims (2)

1. based on the power electronic circuit on-line intelligence failure prediction method of RS-CMAC, it is characterized in that comprising the steps:
Step 1 is by treating that power scale selects measured node from electronic circuit, the voltage of the said measured node of on-line monitoring, current signal;
Step 2 is done wavelet threshold to voltage, the current signal of the measured node described in the step 1 and is handled, and obtains the fault signature sample;
Step 3 from the described fault signature sample extraction of step 2 circuit performance parameters, obtains the circuit performance parameters vector;
Step 4 is carried out trend prediction in conjunction with RS theory and CMAC Forecasting Methodology to the described circuit performance parameters vector of step 3, obtains the following time series of circuit performance parameters constantly;
Step 5, in sampling period of every mistake, repeating step 1 obtains the time series of circuit performance parameters to step 4 according to the circuit performance parameters vector of real-time update;
Step 6 is utilized the time series counting circuit health indicator of the said circuit performance parameters of step 5, and the circuit health indicator that relatively calculates and circuit healthy threshold value, the health status of decision circuit then.
2. the power electronic circuit on-line intelligence failure prediction method based on RS-CMAC according to claim 1 is characterized in that the practical implementation method of said step 4 is: each circuit performance parameters in the circuit performance parameters vector is set up a RS-CMAC model; Obtain the mapping ruler storehouse of circuit performance parameters according to the RS theory; According to the following time series of circuit performance parameters constantly of mapping ruler storehouse prediction.
CN2012103099306A 2012-08-28 2012-08-28 Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller) Pending CN102830341A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2012103099306A CN102830341A (en) 2012-08-28 2012-08-28 Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2012103099306A CN102830341A (en) 2012-08-28 2012-08-28 Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller)

Publications (1)

Publication Number Publication Date
CN102830341A true CN102830341A (en) 2012-12-19

Family

ID=47333543

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2012103099306A Pending CN102830341A (en) 2012-08-28 2012-08-28 Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller)

Country Status (1)

Country Link
CN (1) CN102830341A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109615003A (en) * 2018-12-06 2019-04-12 哈尔滨工业大学 A kind of power source failure prediction method based on ELM-CHMM
CN112348078A (en) * 2020-11-09 2021-02-09 南京工程学院 Gate machine controller with sub-health pre-diagnosis and fault type clustering functions
CN112505531A (en) * 2021-02-04 2021-03-16 湖南遥光科技有限公司 Circuit fault diagnosis method and device based on support vector optimization

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4193115A (en) * 1977-12-15 1980-03-11 The United States Of America As Represented By The Secretary Of Commerce Method and apparatus for implementation of the CMAC mapping algorithm
KR20000051612A (en) * 1999-01-25 2000-08-16 권성규 Parameter mapping calculation method in CMAC(Cerebellar Model Articulation Controller)
CN101799367A (en) * 2010-01-27 2010-08-11 北京信息科技大学 Electromechanical device neural network failure trend prediction method
CN102636740A (en) * 2012-04-18 2012-08-15 南京航空航天大学 Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine)

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4193115A (en) * 1977-12-15 1980-03-11 The United States Of America As Represented By The Secretary Of Commerce Method and apparatus for implementation of the CMAC mapping algorithm
KR20000051612A (en) * 1999-01-25 2000-08-16 권성규 Parameter mapping calculation method in CMAC(Cerebellar Model Articulation Controller)
CN101799367A (en) * 2010-01-27 2010-08-11 北京信息科技大学 Electromechanical device neural network failure trend prediction method
CN102636740A (en) * 2012-04-18 2012-08-15 南京航空航天大学 Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine)

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
冯远静等: "粗糙集CMAC神经网络故障诊断策略", 《华侨大学学报（自然科学版）》, vol. 25, no. 3, 31 July 2004 (2004-07-31) *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109615003A (en) * 2018-12-06 2019-04-12 哈尔滨工业大学 A kind of power source failure prediction method based on ELM-CHMM
CN109615003B (en) * 2018-12-06 2022-11-08 哈尔滨工业大学 Power failure prediction method based on ELM-CHMM
CN112348078A (en) * 2020-11-09 2021-02-09 南京工程学院 Gate machine controller with sub-health pre-diagnosis and fault type clustering functions
CN112505531A (en) * 2021-02-04 2021-03-16 湖南遥光科技有限公司 Circuit fault diagnosis method and device based on support vector optimization

Similar Documents

Publication Publication Date Title
CN106872657B (en) A kind of multivariable water quality parameter time series data accident detection method
CN102749572B (en) IECMAC parameter identification-based power electronic circuit failure predicting method
CN105300692B (en) A kind of bearing failure diagnosis and Forecasting Methodology based on expanded Kalman filtration algorithm
CN109543210A (en) A kind of Wind turbines failure prediction system based on machine learning algorithm platform
CN102636740B (en) Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine)
CN105095963A (en) Method for accurately diagnosing and predicting fault of wind tunnel equipment
CN107797537A (en) A kind of prognostic and health management method applied to automatic production line
CN102340811A (en) Method for carrying out fault diagnosis on wireless sensor networks
CN108764601A (en) A kind of monitoring structural health conditions abnormal data diagnostic method based on computer vision and depth learning technology
CN102013148B (en) Multi-information fusion fire hazard detection method
CN109492790A (en) Wind turbines health control method based on neural network and data mining
CN105891794B (en) Radar health control method and system based on fuzzy criterion
CN103810328A (en) Transformer maintenance decision method based on hybrid model
CN101799674A (en) Method for analyzing service state of numerical control equipment
EP2682836B1 (en) Method for performing diagnostics of a structure subject to loads and system for implementing said method
CN104615122B (en) A kind of industry control signal detection system and detection method
CN102779230A (en) State analysis and maintenance decision judging method of power transformer system
CN104268375A (en) Ship electric power station fault diagnosing method based on knowledge petri network
CN105512483A (en) Residual life predication method applied to aircraft engines and based on damage transmission models and data analysis
CN108090606A (en) Equipment fault finds method and system
CN103455658A (en) Weighted grey target theory based fault-tolerant motor health status assessment method
Xie et al. Neural-network based structural health monitoring with wireless sensor networks
CN102749584B (en) Prediction method for residual service life of turbine generator based on ESN (echo state network) of Kalman filtering
CN102279928A (en) Product performance degradation interval prediction method based on support vector machine and fuzzy information granulation
CN102830341A (en) Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller)

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20121219