CN107884475A - A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net - Google Patents
A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net Download PDFInfo
- Publication number
- CN107884475A CN107884475A CN201710967776.4A CN201710967776A CN107884475A CN 107884475 A CN107884475 A CN 107884475A CN 201710967776 A CN201710967776 A CN 201710967776A CN 107884475 A CN107884475 A CN 107884475A
- Authority
- CN
- China
- Prior art keywords
- pipeline
- gas pipeline
- fault
- sae
- city gas
- 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
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
- G01N29/045—Analysing solids by imparting shocks to the workpiece and detecting the vibrations or the acoustic waves caused by the shocks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4445—Classification of defects
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4481—Neural networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/02—Indexing codes associated with the analysed material
- G01N2291/028—Material parameters
- G01N2291/0289—Internal structure, e.g. defects, grain size, texture
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/26—Scanned objects
- G01N2291/262—Linear objects
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Signal Processing (AREA)
- Acoustics & Sound (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Examining Or Testing Airtightness (AREA)
Abstract
The present invention provides a kind of city gas pipeline method for diagnosing faults based on deep learning neutral net, the deep learning neural network classification model being made up of sparse autocoder SAE and SOFTMAX is established, and combines pipeline fault type structure pipeline fault diagnostic model to realize the failure modes of gas pipeline.Pass through automatic learning characteristic parameter unsupervised in deep learning, the trim network for having supervision is carried out again, efficiently solve the problem of pipeline fault diagnostic characteristic parameter chooses empirical and low accuracy rate of diagnosis, city gas pipeline is carried out in the process of running, fault diagnosis is quicker, also improves the stability, accuracy, reliability of fault diagnosis.
Description
Technical field
The present invention relates to gas pipeline fault diagnosis technology field, and deep learning neutral net is based on more particularly to one kind
City gas pipeline method for diagnosing faults.
Background technology
The gas pipeline of underground is embedded in because laying work area is wide, complex circuit, after pipeline breaks down, if pipeline
Fault type judges inaccuracy, cannot adopt right measures in time to repair trouble point, so as to not only give gas user band
Carry out numerous inconvenience, also result in the waste of resource, it is also possible to bring numerous potential safety hazards to nearby residents, and cause environment
Pollution.
In the prior art, usually, domestic conventional method for diagnosing faults has artificial direct Detection Method, SVMs
(support vector machine, SVM), evidence theory (Dempster-Shafter Evidence Theory, it is hereafter simple
Claim D-S theories) fusion and reverse transmittance nerve network (Back Propagation Neural Network, BPNN) etc..First
Kind method is the most universal, but its time-consuming amount is big, high to the experience dependency degree of staff;But reverse transmittance nerve network
(BPNN) there is the problem of convergence rate is excessively slow in method, and be easily trapped into local optimum;DS fusion methods were generally identifying
Great amount of samples data can not be obtained in journey;SVM is a grader for being directed to two classification problems in fact, and is asked for more classification
Topic, then parameter be present and the shortcomings of kernel function is not easy to determine, and the construction of learner is relatively difficult, final classifying quality
Also it is bad.
The change of the conveying characteristic of pump and the change of conveying technique in being conveyed due to gas pipeline, in fuel gas transmission process
In, often failure signal is frequently influenceed by delivery pump and produces strong interference signal and the feelings of multiple repercussions signals
Condition.So as to cause the fault-signal of extraction inaccurate, cause the failure Source Type that can not accurately distinguish gas pipeline, so as to prolong
The time of the timely reparation source of trouble has been missed, has caused irremediable loss.
Therefore, decision problem of the gas pipeline to failure Source Type during conveying is resolved, avoids gas pipeline
The error that fault type judges, for ensureing that gas pipeline conveys reliable, safe operation, there is very urgent necessity.
The content of the invention
The technical problems to be solved by the invention are:In order to overcome in the prior art artificial detection speed it is slow, supporting vector
Machine (SVM), evidence theory (D-S) fusion and feature is joined in pipeline fault diagnosis the methods of reverse transmittance nerve network (BPNN)
Number chooses empirical, the relatively low deficiency of diagnostic result accuracy.The present invention provides a kind of city based on deep learning neutral net
City's gas pipeline method for diagnosing faults, establish based on deep learning neural network classification (classification deep
Learning neural network, CDLNN) pipeline fault diagnostic model, it is intended to it is accurate that city gas pipeline failure is carried out
Judge, increase applicability, the generality of gas pipeline method for diagnosing faults.
City gas pipeline method for diagnosing faults based on deep learning neutral net is one kind with sparse autocoder
(SAE) and based on the deep learning neural network classification model of SOFTMAX graders composition, and 8 kinds of gas pipeline are combined
Pipeline fault disaggregated model is established after characteristic parameter, the method that intelligent diagnostics are carried out to pipeline fault.
The present invention solves the technical scheme that its technical problem uses:A kind of city combustion based on deep learning neutral net
Feed channel method for diagnosing faults, comprises the following steps:
Step 1, CDLNN models are built, and test checking is carried out with classical machine learning data set;
Step 2, apply to pipe under test and encourage, the acoustic emission signal after collection excitation at the test node of pipeline, and
Acoustic emission signal after wavelet transform denoising is handled is as pipeline fault characteristic signal;
Step 3, pipeline fault characteristic signal is standardized, and is divided into training sample and test sample;
Step 4, encoded for different pipeline fault states, and the structure of network and relevant parameter are carried out initially
Change, typically all include the SAE networks number of plies, learning rate, feature vector dimension and iterations etc., build the city based on CDLNN
Gas pipeline fault diagnosis model;
Step 5, fault diagnosis is carried out to pipe under test by existing fault data in data set.
Further, in step 1, CDLNN models are built, and test checking is carried out with classical machine learning data set.It is
By input layer, several sparse autocoder SAE layers, layer of classifying, output layer combines, wherein several SAE layers be with
Existing for the form of stacking.In order to solve polytypic problem, so SOFTMAX is selected as grader, and because the classification
Device is capable of the probability of output category result, and can preferably be combined with sparse autocoder SAE, usually can obtain compared with
Good classifying quality.
Specifically, for ordinary circumstance, most basic autocoder AE is one three layers of neutral net, be respectively with
The input layer that x is represented, the hidden layer represented with h, the output layer represented with y, x and y typically have identical node.Typically sample
Transfer process from x to h is called coding, and h to y transfer process is called into decoding.Assuming that f is coding function, g is decoding letter
Number, then the expression formula of two functions is as follows:
H=f (x)=Sf(Wx+p) (1)
Y=g (h)=Sg(WT+q) (2)
In formula:SfAnd SgTypically take Sigmoid functions;WxWeight matrix between x and h;WTWeights between h and y
Matrix;P is h bias vector;Q is y bias vector.In order to which more accurately character representation is come out, AE parameters W, p and q
It is abbreviated as θ.
Specifically, generally assuming that training sample set S={ x1, x2 ..., xn }, pre-training is carried out to AE indeed through S
Parameter θ is trained.Therefore, generally first the object definition after training is come out, that is, the y exported after decoding should be with inputting x
Approach to greatest extent to the greatest extent, reconstructed error function L (X, Y) can be used for representing this approximation ratio, and L (X, Y) can be defined as:
Formula (4) can be expressed as being based on loss function of the formula (3) on training sample set S, and every layer of AE parameter θ is exactly will
Formula (4) carries out minimum processing to obtain.
In formula:JAE(θ) is the loss function on θ, and N is the number that training sample concentrates input sample.But for big
Most practical applications, if loss function only only to be made to minimization processing, perhaps can many times obtain an identical letter
Number.In order to avoid the appearance of such case, a kind of variant-sparse coding device SAE processing of autocoder is used.
Specifically, generally, self-encoding encoder has two kinds of variants:Sparse autocoder (Sparse
AutoEncoder, SAE) and noise reduction autocoder (Denoising AutoEncoder, DAE).In order to avoid identity
Occur, select sparse autocoder SAE to handle loss function here, sparse autocoder SAE is exactly automatic
Some regular restrictive conditions are added on the basis of encoder AE, may be sparse to make the coding of acquisition try one's best, wherein sparse
Expression way is one kind best in various expression ways.Easily there is the problem of identical, Neng Goutong for above loss function
Cross and openness limitation is carried out to it to solve, also referred to as sparse own coding.Here from it is a kind of based on the method for relative entropy come reality
Existing, formula (5) represents specific loss function.
In formula:β is weight coefficient;ρ represents openness parameter;It is xiJ-th of the average of neuron swashs on hidden layer
Activity.WhereinExpression formula such as formula (6) shown in:
From formula (6) it can be seen thatIncrease with ρ differences can makeGradually reduce, only both values
Minimum value 0 could be obtained when equal.Therefore,The process that 0 is approached with ρ difference can be realized by minimizing function.
Specifically, established CDLNN models are used to Iris, Adult, Wine, Car Evaluation this 4 classics
Machine learning data set class test, initial learning rate value is arranged to 0.1, and Gaussian Profile is deferred in network parameter θ initialization
Random smaller value, parameter renewal rate value is arranged to 0.01.Table 1 illustrates classification shapes of the CDLNN to 4 different pieces of information collection
Condition.From this column of average correct classification rate, it can be seen that CDLNN has reached satisfactory to the classifying quality of each data set
Result, the CDLNN models for illustrating to have established can be used for solve polytypic problem.
Assortments of the table 1CDLNN to 4 kinds of data sets
Further, in step 2, apply to pipe under test and encourage, the sound hair after collection excitation at the test node of pipeline
Penetrate signal.Refer to broken to tested pipeline lead, gauze friction, opening leak valve equal excitation measure, in sound emission acquisition system
Acoustic emission signal under middle collection different faults state.Due to fault diagnosis, institute directly can not be carried out to the gas pipeline of reality
Gas pipeline fault diagnostic test, the gatherer process of source of trouble acoustic emission signal are carried out in laboratory with elder generation:Sent out according to international sound
The quasi- side of detection is penetrated, disconnected lead is carried out to simulate the acoustic emission signal that pipeline fatigue crack failure is sent using 0.5mm HB pen cores, uses
Sand paper quickly rubs pipeline outer wall to simulate the pipeline acoustic emission signal that pipeline is sent when failure caused by washing away, in opening conduits
Leak valve simulate the acoustic emission signal that pipeline is sent when gas pipeline persistently leaks, and gather pipeline in the case of normal transport
The acoustic emission signal sent.The acoustic emission signal of wherein trouble point is adopted by being placed on the acoustic emission sensor of trouble point both sides
Collect, and by amplifier amplifies signals, be finally transferred in Acoustic radiating instrument.The sound gathered is sent out using Wavelet Transform
Penetrate signal carry out characteristic parameter extraction, according to the general selection rule of characteristic parameter distinguish amplitude, absolute energy, Ring-down count,
The spy of this 8 kinds of reacting pipe running statuses of rise time, duration, average signal level, RMS voltage and event count
Variable is levied, because fine setting needs to use a small amount of tape label sample, so using the data that obtain when simulating pipeline fault as finely tuning
Required sample.
Specifically, acoustic emission signal is collected by the acoustic emission sensor in data acquisition unit, and pass through preceding storing
Acoustic emission signal amplification is transferred in industrial computer by big device, so as to obtain acoustic emission signal data.In order to ensure sound emission
The integrality and accuracy of signal, the acoustic emission sensor are installed on the both sides in pipeline fault source.
Specifically, the data acquisition unit is including being provided with the industrial computer of XP systems, PCI-II bilaterals say hair
Penetrate card, model S/N2462026504, acoustic emission preamplifier, the model R15 list that filter range is 20-120KHz
Hold broadband acoustic emission sensor and corresponding processing software.
Pipe under test can produce acoustic emission phenomenon when breaking down, pipeline occurs fatigue crack, persistently leaks, washes away sternly
Elastic wave when heavy, without exception is that acoustic emission signal is propagated along the source of trouble to pipe under test both sides.Installed in pipe
After acoustic emission sensor on road monitors the signal, signal is reached and passes through a series of software processing in sound emission capture card
Data are formed afterwards to be stored in computer.
Further, in step 3, pipeline fault characteristic signal is standardized, and be divided into training sample and survey
Sample sheet.The data collected with reference to acoustic emission detection system, and according to the automatic study of CDLNN models, converting characteristic and divide
The ability of class, choose amplitude, absolute energy, Ring-down count, rise time, duration, average signal level, RMS voltage
With the characteristic variable of event count this 8 kinds of reacting pipe running statuses, in order that the difference between each characteristic variable contracts as far as possible
It is small, it is more accurate also for calculating is made, therefore standardization is made to each feature first with formula (7).
In formula:xnewFor characteristic value after standardization;X is characterized the original value of parameter;xmeanRepresent that each sample is concentrated shared by x
The average of all parameter value total amounts;xstdRepresent the standard deviation of this characteristic parameter in sample set.
Further, in step 4, encoded for different pipeline fault states, and by the structure and relevant parameter of network
Initialized, typically all include the SAE networks number of plies, learning rate, feature vector dimension and iterations etc., structure is based on
CDLNN city gas pipeline fault diagnosis model., can be event with reference to the failure easily occurred in the actual transport process of pipeline
The diagnostic result of barrier is summarized as 4 classes, is (0,0,0,1) by conduit running normal encoding, pipe leakage is encoded to (0,0,1,
0) pipeline fatigue crack, is encoded to (0,1,0,0), pipeline is washed and is encoded to (1,0,0,0);By the SAE nets of SAE networks
The parameters such as network layers number, learning rate, feature vector dimension and iterations are initialized in MATLAB program.By pipeline event
The training sample set for hindering acoustic emission signal carries out successively training stacking self-encoding encoder, untill reaching convergence.Using pair
Whole SAE networks are trained than diversity factor CD (Contrastive Divergence) fast algorithms.CD algorithms therein come
Gibbs sampling is come from, first calculates the conditional probability of all implicit layer units, and is sampled with Gibbs and determines hidden layer cell-like
State, then the conditional probability of output layer unit is calculated, reuse Gibbs sampling and determine output layer state, now equivalent to right
Previous step output layer is once reconstructed.Again with test sample to the city gas pipeline event based on CDLNN after the completion of training
Barrier diagnostic model is tested, and the structure of the city gas pipeline fault diagnosis model based on CDLNN is just completed after the completion of test
Build.
Further, in step 5, fault diagnosis is carried out to pipe under test by existing fault data in data set.Wherein
Data set refer to pipeline fault characteristic parameter form sample set, by sample set input the city gas pipeline based on CDLNN
Diagnostic result is exported in fault diagnosis model.
The beneficial effects of the invention are as follows:A kind of city gas pipeline based on deep learning neutral net proposed by the present invention
Method for diagnosing faults, establish the deep learning neural network classification mould being made up of sparse autocoder (SAE) and SOFTMAX
Type, and pipeline fault type structure pipeline fault diagnostic model is combined to realize the failure modes of gas pipeline.By depth
Unsupervised automatic learning characteristic parameter in habit, then the trim network for having supervision is carried out, efficiently solve pipeline fault diagnosis
Characteristic parameter chooses the problem of empirical and low accuracy rate of diagnosis, city gas pipeline is carried out fault diagnosis in the process of running
It is more quick, also improve the stability, accuracy, reliability of fault diagnosis.
Brief description of the drawings
The invention will be further described with reference to the accompanying drawings and examples.
The deep learning city gas pipeline Troubleshooting Flowchart that Fig. 1 is returned based on SAE and SOFTMAX;
Fig. 2 is the network of the own coding study (AE) in deep learning neutral net;
Fig. 3 is the mechanism figure of the own coding study (AE) in deep learning neutral net;
Fig. 4 is deep learning neural network classification model (classification deep learning neural
Network, CDLNN) model;
Fig. 5 is the structure chart that pipeline fault diagnoses CDLNN models;
Fig. 6 is the pipeline diagnostic implementing procedure figure based on CDLNN;
Fig. 7 is city gas pipeline fault diagnostic test platform;
Fig. 8 is the pipeline fault diagnosis situation based on CDLNN when the SAE numbers of plies are 0 to 10.
In figure:1st, air compressor, the 2, first ball valve, 3, surge tank, the 4, second ball valve, 5, acoustic emission sensor, 6, meter
Calculation machine, 7, first flow transmitter, 8, second flow transmitter, the 9, the 3rd flow transmitter, 10, first pressure transmitter, 11,
Second pressure transmitter, the 12, the 3rd pressure transmitter, 13, pipeline, 14, leak valve P1,15, leak valve P2,16, leak valve P3,
17th, leak valve P4,18, leak valve P5,19, leak valve P6.
Embodiment
Presently in connection with accompanying drawing, the present invention is described in detail.This figure is simplified schematic diagram, is only illustrated in a schematic way
The basic structure of the present invention, therefore it only shows the composition relevant with the present invention.
As Figure 1-Figure 8, a kind of city gas pipeline fault diagnosis based on deep learning neutral net of the invention
Method, comprise the following steps:
Step 1, CDLNN models are built, and test checking is carried out with classical machine learning data set;
CDLNN models are several SAE layers by input layer, and layer of classifying, output layer combines, wherein several SAE layers
It is existing in the form of stacking.In order to solve polytypic problem, so SOFTMAX is selected as grader, and because should
Grader is capable of the probability of output category result, and can preferably be combined with SAE, usually can obtain preferably classification effect
Fruit.
As shown in Figures 2 and 3, specifically, for ordinary circumstance, most basic AE is one three layers of neutral net, point
It is not the input layer represented with x, the hidden layer represented with h, the output layer represented with y, x and y typically have identical node.Typically
Transfer process of the sample from x to h is called coding, and h to y transfer process is called decoding.Assuming that f is coding function, g is
Decoding functions, then the expression formula of two functions is as follows:
H=f (x)=Sf(Wx+p) (1)
Y=g (h)=Sg(WT+q) (2)
In formula:SfAnd SgTypically take Sigmoid functions;WxWeight matrix between x and h;WTWeights between h and y
Matrix;P is h bias vector;Q is y bias vector.In order to which more accurately character representation is come out, AE parameters W, p and q
It is abbreviated as θ.
Specifically, generally assuming that training sample set S={ x1, x2 ..., xn }, pre-training is carried out to AE indeed through S
Parameter θ is trained.Therefore, generally first the object definition after training is come out, that is, the y exported after decoding should be with inputting x
Approach to greatest extent to the greatest extent, reconstructed error function L (X, Y) can be used for representing this approximation ratio, and L (X, Y) can be defined as:
Formula (4) can be expressed as being based on loss function of the formula (3) on training sample set S, and every layer of AE parameter θ is exactly will
Formula (4) carries out minimum processing to obtain.
In formula:JAE(θ) is the loss function on θ, and N is the number that training sample concentrates input sample.But for big
Most practical applications, if loss function only only to be made to minimization processing, perhaps can many times obtain an identical letter
Number.In order to avoid the appearance of such case, handled using a kind of variant-sparse coding device of autocoder.
Specifically, generally, self-encoding encoder has two kinds of variants:Sparse autocoder (Sparse
AutoEncoder, SAE) and noise reduction autocoder (Denoising AutoEncoder, DA).In order to avoid going out for identity
It is existing, select sparse autocoder SAE to handle loss function here, sparse autocoder SAE is exactly to compile automatically
Some regular restrictive conditions are added on the basis of code device, may be sparse to make the coding of acquisition try one's best, wherein sparse expression
Mode is one kind best in various expression ways.Easily there is the problem of identical for above loss function, can be by right
It carries out openness limitation to solve, also referred to as sparse own coding.Here realized from a kind of based on the method for relative entropy, formula
(5) specific loss function is represented.
In formula:β is weight coefficient;ρ represents openness parameter;It is xiJ-th of the average of neuron swashs on hidden layer
Activity.WhereinExpression formula such as formula (6) shown in:
From formula (6) it can be seen thatIncrease with ρ differences can makeGradually reduce, only both values
Minimum value 0 could be obtained when equal.Therefore,The process that 0 is approached with ρ difference can be realized by minimizing function.
Specifically, established CDLNN models are used to Iris, Adult, Wine, Car Evaluation this 4 classics
Machine learning data set class test, initial learning rate value is arranged to 0.1, and Gaussian Profile is deferred in network parameter θ initialization
Random smaller value, parameter renewal rate value is arranged to 0.01.Table 1 illustrates classification shapes of the CDLNN to 4 different pieces of information collection
Condition.From this column of average correct classification rate, it can be seen that CDLNN has reached satisfactory to the classifying quality of each data set
Result, the CDLNN models for illustrating to have established can be used for solve polytypic problem.
Assortments of the table 1CDLNN to 4 kinds of data sets
Step 2, apply to pipe under test and encourage, the acoustic emission signal after collection excitation at the test node of pipeline, and
Acoustic emission signal after wavelet transform denoising is handled is as pipeline fault characteristic signal;
Apply to pipe under test and encourage, the acoustic emission signal after collection excitation at the test node of pipeline.Refer to quilt
Test tube road broken lead, gauze friction, open leak valve equal excitation measure, in sound emission acquisition system gather different faults shape
Acoustic emission signal under state.Due to fault diagnosis directly can not be carried out to the gas pipeline of reality, so first being carried out in laboratory
Gas pipeline fault diagnostic test, the gatherer process of source of trouble acoustic emission signal:According to the quasi- side of international acoustic emission detection, utilize
0.5mm HB pen cores carry out disconnected lead to simulate the acoustic emission signal that pipeline fatigue crack failure is sent, with the quick friction tube of sand paper
Pipeline outer wall simulates the pipeline acoustic emission signal that pipeline is sent when failure caused by washing away, and leak valve in opening conduits is simulated
The acoustic emission signal that pipeline is sent when gas pipeline persistently leaks, and gather the sound emission letter that pipeline in the case of normal transport is sent
Number.The acoustic emission signal of wherein trouble point is collected by being placed on the acoustic emission sensor of trouble point both sides, and is passed through
Amplifier amplifies signals, finally it is transferred in Acoustic radiating instrument.The acoustic emission signal gathered is carried out using Wavelet Transform special
Levy parameter extraction, according to the general selection rule of characteristic parameter distinguish amplitude, absolute energy, Ring-down count, the rise time, continue
The characteristic variable of this 8 kinds of reacting pipe running statuses of time, average signal level, RMS voltage and event count, due to micro-
Tune needs to use a small amount of tape label sample, so using the data that obtain when simulating pipeline fault as sample needed for finely tuning.
Specifically, acoustic emission signal is collected by the acoustic emission sensor in data acquisition unit, and pass through preceding storing
Acoustic emission signal amplification is transferred in industrial computer by big device, so as to obtain acoustic emission signal data.In order to ensure sound emission
The integrality and accuracy of signal, the acoustic emission sensor are installed on the both sides in pipeline fault source.
Specifically, the data acquisition unit is including being provided with the industrial computer of XP systems, PCI-II bilaterals say hair
Penetrate card, model S/N2462026504, acoustic emission preamplifier, the model R15 list that filter range is 20-120KHz
Hold broadband acoustic emission sensor and corresponding processing software.
Pipe under test can produce acoustic emission phenomenon when breaking down, pipeline occurs fatigue crack, persistently leaks, washes away sternly
Elastic wave when heavy, without exception is that acoustic emission signal is propagated along the source of trouble to pipe under test both sides.Installed in pipe
After acoustic emission sensor on road monitors the signal, signal is reached and passes through a series of software processing in sound emission capture card
Data are formed afterwards to be stored in computer.
Step 3, pipeline fault characteristic signal is standardized, and is divided into training sample and test sample;
The data collected with reference to acoustic emission detection system, and according to the automatic study of CDLNN models, converting characteristic and divide
The ability of class, choose amplitude, absolute energy, Ring-down count, rise time, duration, average signal level, RMS voltage
With the characteristic variable of event count this 8 kinds of reacting pipe running statuses, in order that the difference between each characteristic variable contracts as far as possible
It is small, it is more accurate also for calculating is made, therefore standardization is made to each feature first with formula (7).
In formula:xnewFor characteristic value after standardization;X is characterized the original value of parameter;xmeanRepresent that each sample is concentrated shared by x
The average of all parameter value total amounts;xstdRepresent the standard deviation of this characteristic parameter in sample set.
Step 4, encoded for different pipeline fault states, and the structure of network and relevant parameter are carried out initially
Change, typically all include the SAE networks number of plies, learning rate, feature vector dimension and iterations etc., build the city based on CDLNN
Gas pipeline fault diagnosis model;
With reference to the failure easily occurred in the actual transport process of pipeline, the diagnostic result of failure can be summarized as 4 classes, will
Conduit running normal encoding is (0,0,0,1), and pipe leakage is encoded into (0,0,1,0), pipeline fatigue crack is encoded to (0,
1,0,0), pipeline is washed and is encoded to (1,0,0,0);It typically will all include the SAE networks number of plies, learning rate, characteristic vector dimension
The parameter such as number and iterations is initialized in MATLAB program.By the training sample set of pipeline fault acoustic emission signal
Carry out successively training and stack self-encoding encoder, untill reaching convergence.Use contrast difference degree CD (Contrastive
Divergence) fast algorithm trains whole network.CD algorithms therein are derived from Gibbs sampling, first calculate all
The conditional probability of implicit layer unit, and sampled with Gibbs and determine hidden layer location mode, then calculate the condition of output layer unit
Probability, reuse Gibbs sampling and determine output layer state, now once reconstructed equivalent to previous step output layer.
The city gas pipeline fault diagnosis model based on CDLNN is tested with test sample again after the completion of training, test is completed
The structure of the city gas pipeline fault diagnosis model based on CDLNN is just completed afterwards.
Step 5, fault diagnosis is carried out to pipe under test by existing fault data in data set.Data set therein is
Refer to the sample set that pipeline fault characteristic parameter is formed, sample set is inputted into the city gas pipeline fault diagnosis mould based on CDLNN
Diagnostic result is exported in type.
Embodiment:
In the present embodiment, effectiveness of the invention is verified by following steps:
As shown in fig. 7, city gas pipeline fault diagnostic test platform includes laboratory pipe leakage acoustic emission detection system
System and the acoustic emission sensor of trouble point both sides, laboratory pipe leakage acoustic emission detection system be by data acquisition and processing,
Oil pipeline and storage unit and measuring instrument instrument 3 units composition, measuring instrument are the sound emission cards of PCI- II, S/N2462026504
Amplifier, the single-ended broadband acoustic emission sensors of R15.
Pipe under test is specifically included, pipe under test is stacked around forming three layers, and setting two in every layer of pipe under test lets out
Valve, a flow transmitter and a pressure transmitter are leaked, therefore, six leak valves is shared, respectively leak valve P1 14, lets out
Leak valve P2 15, leak valve P3 16, leak valve P4 17, leak valve P5 18 and leak valve P6 19;Share three flow pick-ups
Device, respectively first flow transmitter 7, the flow transmitter 9 of second flow transmitter 8 and the 3rd;Three pressure transmitters are shared,
Respectively first pressure transmitter 10, the pressure transmitter 12 of second pressure transmitter 11 and the 3rd;Pipeline input also according to
It is secondary to be provided with air compressor 1, the conveying of combustion gas in pipeline is simulated, keeps manifold pressure;The He of the first ball valve 2 as switch
Second ball valve 4, surge tank 3 is additionally provided between the first ball valve 2 and the second ball valve 4;The voice sending sensor being connected with computer
Device 5.The present embodiment includes the single-ended broadband acoustic emission sensors of two R15, and the two sensors are placed on pipeline fault point
Both sides.
The first step:Apply to pipe under test and encourage, the acoustic emission signal after collection excitation at the test node of pipeline, and
Acoustic emission signal after wavelet transform denoising is handled is as pipeline fault characteristic signal.
As shown in Figure 5 and Figure 6, experimental data gatherer process is as follows:By air compressor air simulation combustion is provided for pipeline
Feed channel, disconnected lead, gauze friction are carried out to tested pipeline, opens operation simulation pipeline break, crackle, the leakages such as leak valve, and
And any operation simulation pipeline normal operation is not carried out to pipeline, table 2 is illustrated and encoded corresponding to fault type.Pass through two
The single-ended broadband acoustic emission sensors of R15 carry out the collection of sound emission data, then using wavelet analysis method to two sensors
Signal data carries out feature extraction, selects amplitude, absolute energy, Ring-down count, rise time, duration, average signal electricity
Flat, RMS voltage and event count this initial data of 8 kinds of characteristic parameter as input model.
The conduit running state encoding of table 2
Normal, failure is selected in this experiment and approximate fault sample goes label data to be used as pre-training collection for totally 2500 groups, by event
Barrier, normal condition tape label data are used as fine setting collection and test set for totally 600 groups, and its ratio is 2:1.With the CDLNN moulds established
Type makees following test for pipeline fault diagnosis.
Second step:The determination of the SAE numbers of plies.
In a practical situation, the SAE number of plies is not changeless, but to solve the problems, such as to carry out not according to current
It is disconnected attempt after choose, therefore, the problem of diagnose for pipeline fault, chooses 0 to 10 SAE numbers of plies progress diagnostic test successively,
Diagnosis effect is as shown in Figure 8.It is can be seen that from " the SAE numbers of plies-diagnosis average accuracy " curve in Fig. 8 when the SAE numbers of plies reach
At 4 layers, the average rate of correct diagnosis of pipeline fault is very high, and as the gradual increase of the SAE numbers of plies, average diagnosis are correct
The growth of rate drastically slows down, and the training time of CDLNN models also can be elongated, therefore in order to be obtained with the minimum training time
Optimal diagnosis effect, the SAE number of plies is arranged to 4 layers in experiment.
3rd step:CDLNN diagnoses situation for the pipeline fault of different pre-training collection.
It is as shown in table 3 that CDLNN diagnoses situation for the pipeline fault of different pre-training collection.
CDLNN pipeline fault diagnosis situation during the different pre-training collection of table 3
, it can be seen that CDLNN pipeline fault average accuracy is continuous with the increase of pre-training collection quantity from table 3
Lifting.When pre-training collection reaches 2500, pipeline fault average accuracy has reached more than 90%, and this shows CDLNN
There is preferably performance of fault diagnosis, and suitable for the diagnosis of different pipeline fault situations.
Meanwhile for comparative analysis, the method applied to pipeline fault diagnosis is relatively more, and conventional method has BPNN in addition
Algorithm and SVM algorithm.Now both approaches are tested under identical experiment condition, and contrast ratio is carried out with this paper algorithms
Compared with test result is as shown in table 4.Wherein BPNN maximum training iterations cepochs takes 1500, and learning rate vIr takes
0.01;SVM regularization coefficient C takes 2048, kernel functional parameter γ to take 0.03, and vector machine kernel function selects RBF functions.
Pipeline fault diagnosis situation based on SVM, BPNN during 4 different training sets of table
Pass through the pipeline fault diagnostic result in comparison sheet 3 and table 4, the mean failure rate diagnosis of this paper research methods
Highest, SVM effect is better than traditional BPNN effect, but SVM applies in general to two classification problems, and for more classification
Problem, then there is the deficiency in terms of find optimized parameter and it is necessary to needs that by many experiments ideal could be obtained
Differentiation effect;And the diagnosis effect of this paper algorithms is better than SVM, and avoids in BPNN, SVM method and manually extract and select
The process of feature is selected, automatically Level by level learning feature, extraction reconstruct the feature of representing fault signal essence from input data,
And it is input to and is classified suitable for the grader SOFTMAX of more classification problems.Therefore, this paper research method compared to
BPNN, SVM have more preferable diagnosis effect on pipeline diagnostic, can be rob, maintainer for judge pipeline whether failure
More scientific reference frame is provided.The present invention can realize under different operating modes to the extracted in self-adaptive of pipeline fault degree feature and
Intelligent diagnostics, the feasibility and validity of the present invention are absolutely proved.
The present invention first passes through Wavelet Transform and carries out denoising Processing to the acoustic emission signal of pipeline, and it is special to obtain original sound emission
Parameter is levied, because characteristics of Acoustic Emission parameter is more, the ginseng being had a great influence to diagnostic result is chosen typically by experience
Count to be diagnosed to pipeline, empirically selected characteristic parameter does not have scientific basis, wherein artificial selected characteristic parameter has one
Fixed uncertainty and complexity, so result in the pipeline event of conventional pipelines fault detection method and shallow-layer learning neural network
It is relatively low to hinder discrimination.Original characteristics of Acoustic Emission parameter is changed by sparse autocoder (SAE), and reconstructs reaction
The characteristic parameter of pipeline actual operating state to carry out pipeline fault diagnosis, and the effect of diagnosis is that solve pipeline sound emission spy
The problem of parameter chooses empirical is levied, failure modes discrimination has been reached more than 90%, has reached a preferably diagnosis effect
Fruit.
Generally, the fault diagnosis of pipeline simply judges that pipeline is normal or leaks, the problem of belonging to two classification;And
What the present invention was carried out is the diagnosis of the multistream heat exchanger situation such as pipeline break, leakage, normal, and passes through verification experimental verification pipeline fault
Polytypic rate of correct diagnosis has reached more than 90%, in perfect condition.
Using the above-mentioned desirable embodiment according to the present invention as enlightenment, by above-mentioned description, related staff
Various changes and amendments can be carried out in without departing from the scope of the present invention completely.The technical scope of this invention is not
The content being confined on specification, it is necessary to which its technical scope is determined according to right.
Claims (6)
- A kind of 1. city gas pipeline method for diagnosing faults based on deep learning neutral net, it is characterised in that:With it is sparse from Based on the deep learning neural network classification model that dynamic encoder SAE and SOFTMAX grader is formed, and combine gas pipeline Various features parameter, establish pipeline fault disaggregated model, with to pipeline fault carry out intelligent diagnostics method;Specifically include with Lower step:Step 1, CDLNN models are built, and test checking is carried out with classical machine learning data set;Step 2, city gas pipeline fault diagnostic test platform is built, applies to pipe under test and encourages, in the test section of pipeline The collection of point place encourage after acoustic emission signal, and the acoustic emission signal after wavelet transform denoising is handled is as pipeline fault feature Signal;Step 3, pipeline fault characteristic signal is standardized, and is divided into training sample and test sample;Step 4, encoded for different pipeline fault states, and the structure of SAE networks and relevant parameter initialized, The structure of SAE networks comprises at least the SAE networks number of plies, input characteristic parameter number, output characteristic number of parameters, SAE networks Relevant parameter comprises at least learning rate, feature vector dimension and iterations, using the training sample of pipeline fault acoustic emission signal Sheet and test sample, build the CDLNN city gas pipeline fault diagnosis models based on SAE and SOFTMAX recurrence;Step 5, fault diagnosis is carried out to pipe under test by existing fault data in data set.
- 2. the city gas pipeline method for diagnosing faults based on deep learning neutral net as claimed in claim 1, its feature It is:CDLNN models are several sparse autocoder SAE layers by input layer in step 1, classify layer and output layer combination Form, wherein several SAE layers are existing in the form of stacking, and classification layer choosing selects SOFTMAX as grader;Using being built Vertical CDLNN models are to this 4 Iris, Adult, Wine, Car Evaluation classical machine learning data set class tests.
- 3. the city gas pipeline method for diagnosing faults based on deep learning neutral net as claimed in claim 1, its feature It is:In step 2, apply excitation to pipe under test and refer to break to tested pipeline in lead, gauze friction and opening leak valve The incentive measure of one or more modes, the acoustic emission signal under different faults state is gathered in sound emission acquisition system.
- 4. the city gas pipeline method for diagnosing faults based on deep learning neutral net as claimed in claim 1, its feature It is:In step 3, the data that are collected with reference to acoustic emission detection system, and it is special according to the automatic study of CDLNN models, conversion Seek peace the ability of classification, choose amplitude, absolute energy, Ring-down count, rise time, duration, average signal level, effectively The characteristic variable of this 8 kinds of reacting pipe running statuses of threshold voltage and event count, in order that the difference between each characteristic variable is use up It may reduce, it is more accurate also for calculating is made, therefore standardization is made to each feature first with formula (7):<mrow> <msub> <mi>x</mi> <mrow> <mi>n</mi> <mi>e</mi> <mi>w</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mrow> <mi>x</mi> <mo>-</mo> <msub> <mi>x</mi> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> </mrow> </msub> </mrow> <msub> <mi>x</mi> <mrow> <mi>s</mi> <mi>t</mi> <mi>d</mi> </mrow> </msub> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> </mrow>In formula:xnewFor characteristic value after standardization;X is characterized the original value of parameter;xmeanRepresent that each sample concentrates all ginsengs shared by x The average of numerical value total amount;xstdRepresent the standard deviation of this characteristic parameter in sample set.
- 5. the city gas pipeline method for diagnosing faults based on deep learning neutral net as claimed in claim 1, its feature It is:In step 4, with reference to the failure easily occurred in the actual transport process of pipeline, the diagnostic result of failure is summarized as 4 classes, It is (0,0,0,1) by conduit running normal encoding, pipe leakage is encoded to (0,0,1,0), pipeline fatigue crack is encoded to (0,1,0,0), pipeline is washed and is encoded to (1,0,0,0);The parameter of SAE networks is carried out initially in MATLAB program Change, the parameter includes the SAE networks number of plies, learning rate, feature vector dimension and iterations;By pipeline fault acoustic emission signal Training sample set carry out successively training stack self-encoding encoder, untill reaching convergence;It is fast using contrast difference's degree CD The short-cut counting method trains whole SAE networks;CD algorithms therein are derived from Gibbs sampling, first calculate all implicit layer units Conditional probability, and sampled with Gibbs and determine hidden layer location mode, then calculate the conditional probability of output layer unit, again Sampled using Gibbs and determine output layer state, now once reconstructed equivalent to previous step output layer;After the completion of training The city gas pipeline fault diagnosis model based on CDLNN is tested with test sample again, just completed after the completion of test The structure of city gas pipeline fault diagnosis model based on CDLNN.
- 6. the city gas pipeline method for diagnosing faults based on deep learning neutral net as claimed in claim 1, its feature It is:In step 5, data set refers to the sample set that pipeline fault characteristic parameter is formed, and sample set is inputted into the city based on CDLNN Diagnostic result is exported in city's gas pipeline fault diagnosis model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710967776.4A CN107884475A (en) | 2017-10-18 | 2017-10-18 | A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710967776.4A CN107884475A (en) | 2017-10-18 | 2017-10-18 | A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107884475A true CN107884475A (en) | 2018-04-06 |
Family
ID=61781667
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710967776.4A Pending CN107884475A (en) | 2017-10-18 | 2017-10-18 | A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107884475A (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108875281A (en) * | 2018-08-08 | 2018-11-23 | 佛山科学技术学院 | A kind of hybrid intelligent health status diagnostic method and device based on deep learning |
CN109145886A (en) * | 2018-10-12 | 2019-01-04 | 西安交通大学 | A kind of asynchronous machine method for diagnosing faults of Multi-source Information Fusion |
CN109254577A (en) * | 2018-08-08 | 2019-01-22 | 佛山科学技术学院 | A kind of intelligence manufacture procedure fault classification method and device based on deep learning |
CN109555979A (en) * | 2018-12-10 | 2019-04-02 | 清华大学 | A kind of water supply network leakage monitoring method |
CN109635864A (en) * | 2018-12-06 | 2019-04-16 | 佛山科学技术学院 | A kind of fault tolerant control method and device based on data |
CN109838696A (en) * | 2019-01-09 | 2019-06-04 | 常州大学 | Pipeline fault diagnostic method based on convolutional neural networks |
CN109886433A (en) * | 2019-02-01 | 2019-06-14 | 常州大学 | The method of intelligent recognition city gas pipeline defect |
CN109902617A (en) * | 2019-02-25 | 2019-06-18 | 百度在线网络技术(北京)有限公司 | A kind of image identification method, device, computer equipment and medium |
CN110044554A (en) * | 2019-04-16 | 2019-07-23 | 重庆大学 | A kind of online test method of the metal pressure container leakage based on acoustic emission signal |
CN110083125A (en) * | 2019-05-19 | 2019-08-02 | 重庆理工大学 | A kind of machine tool thermal error modeling method based on deep learning |
CN110568082A (en) * | 2019-09-02 | 2019-12-13 | 北京理工大学 | cable wire breakage distinguishing method based on acoustic emission signals |
CN111024158A (en) * | 2019-12-23 | 2020-04-17 | 广东工业大学 | Kitchen electrical hazard intelligent monitoring method combined with edge calculation |
WO2020119169A1 (en) * | 2018-12-13 | 2020-06-18 | 数优(苏州)人工智能科技有限公司 | Computer readable storage medium, input data checking method and computing device |
US20200210810A1 (en) * | 2019-01-02 | 2020-07-02 | The Aerospace Corporation | Fusing output of artificial intelligence networks |
WO2021000061A1 (en) * | 2019-07-02 | 2021-01-07 | 北京理工大学 | Method for determining leakage level of gas pipe network based on improved evidence fusion algorithm |
CN112597692A (en) * | 2020-09-21 | 2021-04-02 | 南京航空航天大学 | Method for monitoring flow state of supersonic air inlet channel in real time |
CN113484054A (en) * | 2021-07-06 | 2021-10-08 | 座头鲸通信技术(武汉)有限公司 | Intelligent furniture fresh air system operation fault intelligent diagnosis and analysis method and cloud analysis platform |
CN115588166A (en) * | 2022-11-10 | 2023-01-10 | 新乡市诚德能源科技装备有限公司 | Prevent leaking marine LNG fuel jar |
CN116625438A (en) * | 2023-07-25 | 2023-08-22 | 克拉玛依市燃气有限责任公司 | Gas pipe network safety on-line monitoring system and method thereof |
CN116772944A (en) * | 2023-08-25 | 2023-09-19 | 克拉玛依市燃气有限责任公司 | Intelligent monitoring system and method for gas distribution station |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104792530A (en) * | 2015-04-15 | 2015-07-22 | 北京航空航天大学 | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression |
US20170004397A1 (en) * | 2015-06-30 | 2017-01-05 | Adobe Systems Incorporated | Procedural modeling using autoencoder neural networks |
CN106323636A (en) * | 2016-08-16 | 2017-01-11 | 重庆交通大学 | Adaptive extraction and diagnosis method for degree features of mechanical fault through stack-type sparse automatic coding depth neural network |
CN106408088A (en) * | 2016-11-22 | 2017-02-15 | 北京六合智汇技术有限责任公司 | Depth-learning-theory-based fault diagnosis method for rotating machinery |
CN106845437A (en) * | 2017-02-13 | 2017-06-13 | 常州大学 | Leakage of City Gas Pipeline localization method based on Support vector regression |
CN106895975A (en) * | 2017-01-17 | 2017-06-27 | 苏州大学 | Method for Bearing Fault Diagnosis based on Stacked SAE deep neural networks |
-
2017
- 2017-10-18 CN CN201710967776.4A patent/CN107884475A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104792530A (en) * | 2015-04-15 | 2015-07-22 | 北京航空航天大学 | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression |
US20170004397A1 (en) * | 2015-06-30 | 2017-01-05 | Adobe Systems Incorporated | Procedural modeling using autoencoder neural networks |
CN106323636A (en) * | 2016-08-16 | 2017-01-11 | 重庆交通大学 | Adaptive extraction and diagnosis method for degree features of mechanical fault through stack-type sparse automatic coding depth neural network |
CN106408088A (en) * | 2016-11-22 | 2017-02-15 | 北京六合智汇技术有限责任公司 | Depth-learning-theory-based fault diagnosis method for rotating machinery |
CN106895975A (en) * | 2017-01-17 | 2017-06-27 | 苏州大学 | Method for Bearing Fault Diagnosis based on Stacked SAE deep neural networks |
CN106845437A (en) * | 2017-02-13 | 2017-06-13 | 常州大学 | Leakage of City Gas Pipeline localization method based on Support vector regression |
Non-Patent Citations (2)
Title |
---|
王新颖 等: "基于SVR的城市燃气管道泄漏定位研究", 《消防管理研究》 * |
石鑫 等: "深度学习神经网络在电力变压器故障诊断中的应用", 《电力建设》 * |
Cited By (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109254577A (en) * | 2018-08-08 | 2019-01-22 | 佛山科学技术学院 | A kind of intelligence manufacture procedure fault classification method and device based on deep learning |
CN108875281A (en) * | 2018-08-08 | 2018-11-23 | 佛山科学技术学院 | A kind of hybrid intelligent health status diagnostic method and device based on deep learning |
CN109145886A (en) * | 2018-10-12 | 2019-01-04 | 西安交通大学 | A kind of asynchronous machine method for diagnosing faults of Multi-source Information Fusion |
CN109635864A (en) * | 2018-12-06 | 2019-04-16 | 佛山科学技术学院 | A kind of fault tolerant control method and device based on data |
CN109555979A (en) * | 2018-12-10 | 2019-04-02 | 清华大学 | A kind of water supply network leakage monitoring method |
WO2020119169A1 (en) * | 2018-12-13 | 2020-06-18 | 数优(苏州)人工智能科技有限公司 | Computer readable storage medium, input data checking method and computing device |
US11494613B2 (en) * | 2019-01-02 | 2022-11-08 | The Aerospace Corporation | Fusing output of artificial intelligence networks |
US20200210810A1 (en) * | 2019-01-02 | 2020-07-02 | The Aerospace Corporation | Fusing output of artificial intelligence networks |
CN109838696A (en) * | 2019-01-09 | 2019-06-04 | 常州大学 | Pipeline fault diagnostic method based on convolutional neural networks |
CN109886433A (en) * | 2019-02-01 | 2019-06-14 | 常州大学 | The method of intelligent recognition city gas pipeline defect |
CN109902617A (en) * | 2019-02-25 | 2019-06-18 | 百度在线网络技术(北京)有限公司 | A kind of image identification method, device, computer equipment and medium |
CN110044554A (en) * | 2019-04-16 | 2019-07-23 | 重庆大学 | A kind of online test method of the metal pressure container leakage based on acoustic emission signal |
CN110083125B (en) * | 2019-05-19 | 2021-05-07 | 重庆理工大学 | Machine tool thermal error modeling method based on deep learning |
CN110083125A (en) * | 2019-05-19 | 2019-08-02 | 重庆理工大学 | A kind of machine tool thermal error modeling method based on deep learning |
WO2021000061A1 (en) * | 2019-07-02 | 2021-01-07 | 北京理工大学 | Method for determining leakage level of gas pipe network based on improved evidence fusion algorithm |
CN110568082A (en) * | 2019-09-02 | 2019-12-13 | 北京理工大学 | cable wire breakage distinguishing method based on acoustic emission signals |
CN111024158B (en) * | 2019-12-23 | 2021-11-19 | 广东工业大学 | Kitchen electrical hazard intelligent monitoring method combined with edge calculation |
CN111024158A (en) * | 2019-12-23 | 2020-04-17 | 广东工业大学 | Kitchen electrical hazard intelligent monitoring method combined with edge calculation |
CN112597692A (en) * | 2020-09-21 | 2021-04-02 | 南京航空航天大学 | Method for monitoring flow state of supersonic air inlet channel in real time |
CN113484054A (en) * | 2021-07-06 | 2021-10-08 | 座头鲸通信技术(武汉)有限公司 | Intelligent furniture fresh air system operation fault intelligent diagnosis and analysis method and cloud analysis platform |
CN115588166A (en) * | 2022-11-10 | 2023-01-10 | 新乡市诚德能源科技装备有限公司 | Prevent leaking marine LNG fuel jar |
CN115588166B (en) * | 2022-11-10 | 2023-02-17 | 新乡市诚德能源科技装备有限公司 | Prevent leaking marine LNG fuel jar |
CN116625438A (en) * | 2023-07-25 | 2023-08-22 | 克拉玛依市燃气有限责任公司 | Gas pipe network safety on-line monitoring system and method thereof |
CN116625438B (en) * | 2023-07-25 | 2023-12-01 | 克拉玛依市燃气有限责任公司 | Gas pipe network safety on-line monitoring system and method thereof |
CN116772944A (en) * | 2023-08-25 | 2023-09-19 | 克拉玛依市燃气有限责任公司 | Intelligent monitoring system and method for gas distribution station |
CN116772944B (en) * | 2023-08-25 | 2023-12-01 | 克拉玛依市燃气有限责任公司 | Intelligent monitoring system and method for gas distribution station |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107884475A (en) | A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net | |
CN105372087B (en) | Method for diagnosing faults based on multiple sensor signals analysis | |
CN110044554A (en) | A kind of online test method of the metal pressure container leakage based on acoustic emission signal | |
CN103822793B (en) | A kind of complex apparatus acoustic fault recognition positioning method | |
CN100552668C (en) | Leakage detecting and locating method based on pressure and sound wave information fusion | |
CN109299705A (en) | Rotary machinery fault diagnosis method based on one-dimensional depth residual error convolutional neural networks | |
CN101587155B (en) | Oil soaked transformer fault diagnosis method | |
CN104712542B (en) | A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults | |
CN106611090B (en) | A kind of road side air pollutant concentration Forecasting Methodology based on reconstruct deep learning | |
CN110514924A (en) | The Winding in Power Transformer Fault Locating Method of depth convolutional neural networks fusion visual recognition | |
CN109838696A (en) | Pipeline fault diagnostic method based on convolutional neural networks | |
CN106323636A (en) | Adaptive extraction and diagnosis method for degree features of mechanical fault through stack-type sparse automatic coding depth neural network | |
CN104751229A (en) | Bearing fault diagnosis method capable of recovering missing data of back propagation neural network estimation values | |
CN104052612B (en) | A kind of Fault Identification of telecommunication service and the method and system of positioning | |
CN109858352A (en) | A kind of method for diagnosing faults based on compressed sensing and the multiple dimensioned network of improvement | |
CN108092623B (en) | A kind of photovoltaic array multisensor fault detecting and positioning method | |
CN106092578A (en) | A kind of machine tool mainshaft bearing confined state online test method based on wavelet packet and support vector machine | |
CN104021238A (en) | Lead-acid power battery system fault diagnosis method | |
CN109214356A (en) | A kind of fan transmission system intelligent fault diagnosis method based on DCNN model | |
Cheng et al. | A BRB-based effective fault diagnosis model for high-speed trains running gear systems | |
CN109886433A (en) | The method of intelligent recognition city gas pipeline defect | |
CN203083829U (en) | Automobile fault diagnosis system based on tail gas discharge detection adopting simple driving mode | |
CN108106844A (en) | A kind of Method for Bearing Fault Diagnosis of the automatic coding machine of self-tuning parameter adjustment | |
CN107272644A (en) | The DBN network fault diagnosis methods of latent oil reciprocating oil pumping unit | |
CN106197886A (en) | Hydraulic cylinder interior leakage leakage fault diagnosis appraisal procedure |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180406 |