CN110044554A - A kind of online test method of the metal pressure container leakage based on acoustic emission signal - Google Patents
A kind of online test method of the metal pressure container leakage based on acoustic emission signal Download PDFInfo
- Publication number
- CN110044554A CN110044554A CN201910304138.3A CN201910304138A CN110044554A CN 110044554 A CN110044554 A CN 110044554A CN 201910304138 A CN201910304138 A CN 201910304138A CN 110044554 A CN110044554 A CN 110044554A
- Authority
- CN
- China
- Prior art keywords
- network
- svm
- training
- sdae
- acoustic emission
- 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
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M3/00—Investigating fluid-tightness of structures
- G01M3/02—Investigating fluid-tightness of structures by using fluid or vacuum
- G01M3/04—Investigating fluid-tightness of structures by using fluid or vacuum by detecting the presence of fluid at the leakage point
- G01M3/24—Investigating fluid-tightness of structures by using fluid or vacuum by detecting the presence of fluid at the leakage point using infrasonic, sonic, or ultrasonic vibrations
-
- 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
- G01N29/14—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 using acoustic emission techniques
-
- 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
- G01N2291/26—Scanned objects
- G01N2291/269—Various geometry objects
- G01N2291/2695—Bottles, containers
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Immunology (AREA)
- Health & Medical Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Pathology (AREA)
- Acoustics & Sound (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
Online test method is revealed based on the metal pressure container sound emission for stacking noise reduction self-encoding encoder the invention discloses a kind of.The following steps are included: acquiring all kinds of acoustic emission signals that excitation generates, line number of going forward side by side Data preprocess by acoustic emission sensor first;It determines SDAE network parameter, sets suitable cost function and optimisation strategy;Unsupervised pre-training is carried out to SDAE network, until the complete all noise reduction self-encoding encoders of training;Using the parameters such as BP algorithm trim network weight and biasing, until reaching expected accuracy rate;LS-SVM multi-categorizer model is established, the kernel function and assembly coding method of multi-categorizer is determined, the optimized parameter of classifier is found using particle swarm optimization algorithm;Using trained LS-SVM classifier as the classification layer of SDAE network, final SDAE-LS-SVM network is obtained.Online leak detection finally is carried out to metal pressure container, trained network will be inputted after the data processing of acoustic emission sensor acquisition and carries out on-line checking, obtains the leak detection result of pressure vessel.
Description
Technical field
The invention belongs to the leakage on-line checking fields of metal pressure container, are related to a kind of based on stacking noise reduction self-encoding encoder
Online test method is revealed in sound emission with the metal pressure container of LS-SVM.
Background technique
Pressure vessel is the widely used device in the industry fields such as petrochemical industry, medicine, food, belongs to industrial production
With the important infrastructure of people's daily life.Since these equipment contain high temperature, high pressure, inflammable, explosive or hypertoxic Jie mostly
Matter, pressure vessel is once occur cracking or leakage, and often concurrent catastrophic failures such as explosion, fire or poisoning, cause the people raw
The heavy losses of property are ordered, and cause Heavy environmental pollution, social influence is severe.
In order to guarantee the safety used and normalization of pressure vessel, system of the various non-destructive testing technologies in pressure vessel
It makes and is used widely in in-service periodic inspection, existing lossless detection method acoustic emission detection, ray detection, ultrasound inspection
The conventional lossless detection method such as survey, Magnetic testing and Liquid penetrant testing.But the methods of magnetic powder and infiltration, which need to stop production, to be examined, ultrasound
Detection or the energy that detects of radiographic inspection method are not suitable for long-continued detection from nondestructive detecting instrument, and sound emission
Detection belongs to a kind of dynamic testing method, can provide defect with the real-time continuous information of environmental change, and can be quick
Large-scale component, and the activity defect of detection Damage Structure safety are detected, defect multidate information is provided.
Acoustic emission signal belongs to non-linear, non-stationary detection signal, the detection side of currently used pressure vessel leakage
Method has the methods of artificial detection method, reverse transmittance nerve network (BPNN), DS fusion and support vector machines.However BPNN method is deposited
In the problem that convergence rate is excessively slow, and it is easy sunken people's local optimum;DS fusion method can not obtain usually in identification process
A large amount of sample data;Traditional SVM is the classifier for being directed to two classification problems, when the problem is large in scale, secondary rule
The solution for the problem of drawing can become extremely complex, be not easy determining problem there is also parameter and kernel function.
Since pressure vessel and equipment belongs to equipment industrial extensive and being used for a long time, once container occurs cracking and leak
Situations such as, with traditional detection method there may be that can not detect in time in production process, most probably lead to personnel and wealth
The heavy losses of production.Therefore the leakage problem of timely and accurately on-line checking pressure vessel, to preventing accident, guarantor
The safety of member and property is of great significance.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of based on the metal pressure for stacking noise reduction self-encoding encoder and LS-SVM
Online test method is revealed in the sound emission of force container.This method stacks noise reduction autocoder (SDAE) network by building to learn
The feature for practising several typical acoustic emission signals that metal pressure container generates, using least square method supporting vector machine (LS-SVM) structure
At multi-categorizer more classification are carried out to leakage and noise, the parameter of the classifier is adjusted using subgroup optimization (PSO) algorithm
To optimal value, transmitting signal of the trained neural network to pressure vessel is finally subjected to on-line checking identification, to reach pressure
The purpose of the leakage on-line checking of force container.
In order to achieve the above objectives, technical solution of the present invention provides a kind of based on stacking noise reduction self-encoding encoder and LS-SVM
Online test method is revealed in the sound emission of metal pressure container, the described method comprises the following steps:
1) leakage that is generated by excitation by the acquisition of the acoustic emission sensor that is mounted on pressure vessel, noise and normal
Acoustic emission signal, then the data of acquisition are pre-processed.
2) parameters such as SDAE network depth, quantity, learning rate and the number of iterations of each layer neuron are determined, according to task need
Ask the suitable cost function of setting and optimisation strategy.
3) the unsupervised layer-by-layer pre-training of greediness is carried out to SDAE network, until all noise reduction autocodings are completed in training
Device.Using parameters such as the weight of BP algorithm fine tuning whole network and biasings, until reaching expected accuracy rate requirement.
4) LS-SVM multi-categorizer model is established, the kernel function and assembly coding method of multi-categorizer is determined, utilizes grain
Subgroup optimization algorithm finds the kernel functional parameter of classifier and the optimum combination value of penalty.Finally by trained LS-SVM
Classification layer of the classifier as SDAE network, obtains final SDAE-LS-SVM network.
5) after the data for acquiring acoustic emission sensor in real time are pre-processed, trained network is revealed for input
On-line checking, output pressure container is with the presence or absence of leakage.
What the present invention reached has the beneficial effect that the invention proposes one kind based on deep neural network progress stress metal appearance
Online test method is revealed in the sound emission of device.The method overcome artificial detection method speed is slow and conventional method feature mentions
The shortcomings that relying on artificial experience is taken, due to the introducing of denoising strategy, this method can obtain more robust from noisy data
Layered characteristic expression, the robustness and classification capacity of Enhanced feature.Due to using LS-SVM as classifier, this method is not only
It is stronger to small sample generalization ability, and training speed is faster, is converted into SVM solution quadratic programming problem to solve linear equation
Group greatly simplifies the complexity of calculating, and the introducing of particle swarm optimization algorithm can be effectively to the kernel functional parameter of LS-SVM
It is optimized with penalty, to improve the accuracy of pressure vessel leakage identification classification results.
Detailed description of the invention
In order to keep the purpose of the present invention, technical scheme and beneficial effects clearer, the present invention provides following attached drawing and carries out
Illustrate:
Fig. 1 is the implementing procedure block diagram of the leakage online test method of metal pressure container of the present invention.
Fig. 2 is the structural schematic diagram of self-encoding encoder.
Fig. 3 is unsupervised pre-training schematic diagram
Fig. 4 is the algorithm flow schematic diagram that particle swarm optimization algorithm optimizes LS-SVM parameter
Fig. 5 is SDAE-LS-SVM network structure
Specific embodiment
In order to illustrate more clearly of the object, technical solutions and advantages of the present invention, with reference to the accompanying drawing to side of the invention
Method is described, and Fig. 1 is the flow diagram of the leakage online measuring technique scheme of metal pressure container of the invention, specific implementation
Steps are as follows:
1) leakage that is generated by excitation by the acquisition of the acoustic emission sensor that is mounted on pressure vessel, noise and normal
Acoustic emission signal, then carry out data prediction.
The acoustic emission sensor for installing unified specification in each key position of pressure vessel first, by the side for applying excitation
Acoustic emission signal when formula simulation leakage and noise occur, to obtain the sound emission data of needs, the type of data and excitation side
Formula is as shown in table 1 below.The acoustic emission source that pressure vessel scene often detects is divided into normal, leakage signal by the present invention, interference is believed
Numbers three kinds, normal data directly acquires the acoustic emission signal in normal situation, and the sound emission data of leakage can pass through opening
The valve of pressure vessel generates, acoustical signal that the lead that broken with the HB pencil-lead of 0.5mm is cracked with analogue pressure vessel, friction letter
Number just by being obtained with different materials friction pressure container, water droplet signal is simulated by way of dripping toward pressure vessel, electromagnetism
Noise is then realized using operation heavy-duty motor aside, as composite interference then by each noise-like signal combined crosswise one
Get up to acquire acoustic emission signal.
All kinds of acoustic emission source energisation mode tables of 1 pressure vessel of table
It after obtaining sound emission data, needs to extract the characteristic parameter in sound emission data, counting, event meter will be hit
Number, amplitude, energy counting, duration, rise time, RMS voltage, this eight parameters of average signal level are sent out as sound
Penetrate the characteristic parameter of signal.Characteristic parameter also needs that (as shown in formula 1) is normalized before inputting network, for
The sample of composite interference, renormalization after combining the two.
Since the whole fine tuning stage of network needs part tape label data, all kinds of sound emission letters in part are randomly selected
Number production label data.
2) parameters such as SDAE network depth, quantity, learning rate and the number of iterations of each layer neuron are determined, according to task need
Ask the suitable cost function of setting and optimisation strategy.
The structure of SDAE contains input layer, 3 DAE layers of noise reduction self-encoding encoders, classification layer and output layer, wherein 3
DAE layers are to stack the hidden layer of each trained DAE.Classification layer choosing selects the multi-categorizer of LS-SVM composition, the party
Method is not only stronger to small sample generalization ability, and training speed is faster, preferably SDAE can be combined to carry out the more of acoustic emission source
Classification.
Ordinary circumstance AE is made of three layers of feed forward type neural network, and target is to reappear input signal as far as possible, and Fig. 2 is allusion quotation
Type self-encoding encoder structure chart.The training process of AE mainly includes coding stage and decoding stage.Given input sample x ∈ [0,1]N, N
For the number (also referred to as input dimension) of input layer, detailed process is as follows for the coding and decoding of AE:
Coding stage: pass through coding function fθInput x is performed the encoding operation to obtain hidden layer expression, also referred to as coded vector
h∈RM(M is hidden layer neuron number), mathematic(al) representation are as follows:
H=fθ(x)=s (Wx+b) (2)
In formula, s () is nonlinear activation function sigmoid function, i.e., s (z)=1/ (1+exp (- z));W∈RM×NAnd b
∈RMRespectively weight matrix and bias vector.
Decoding stage: pass through decoding functions gθ′H, which is decoded to obtain the reconstruct output of x, to be indicated to hidden layerIts mathematical table
Up to formula are as follows:
In formula, s () is similarly nonlinear activation function sigmoid function, W ' ∈ RM×NWith b ' ∈ RMRespectively hidden layer
Weight matrix and bias vector between output layer, general W '=WT.The training process of AE be by minimize x withWeight
Structure error seeks parameter θ={ W, b, b ' }, and loss function calculates as follows, and wherein K is that training sample is total:
The variant of self-encoding encoder has sparse self-encoding encoder SAE and noise reduction self-encoding encoder DAE, since the latter DAE can be from sample
Middle study to more robustness feature, to reduce DAE to the sensibility of random perturbation small in input data, to distortion or
It is the sample that the characteristic noise containing certain statistical is added in input sample, institute that the input data of missing, which has stronger generalization ability,
According to selection noise reduction self-encoding encoder DAE.
It first carries out plus makes an uproar processing to being originally inputted before DAE coding training, then noisy sample is coded and decoded
Study is originally inputted from the reconstruct of noisy sample.To original input sample x according to qDDistribution is with probability ν (also referred to as noise level)
Random noise is added, it is made to become noisy sampleIn actual test and measurement, to consider noise to institute
The influence of method non-destructive tests effect is proposed, considers 5%, 10%, influence of 20% noise to mentioned method recognition result selects
Wherein effect it is best as final.
The hidden layer of multiple DAE, which is successively accumulated, becomes SDAE, and the output of preceding layer DAE is the defeated of its later layer DAE
Enter, its network number of plies and node in hidden layer determines the final feature vector that network extracts, and will affect the sound of pressure vessel
The accuracy of emission source category identification.It needs in the case where other network parameters are constant, it is true by changing the alignment of the network number of plies
The influence of rate is to select the optimal number of plies, and the default choice number of plies is 3 layers here.
Initial data x and reconstruct dataQuantity n it is consistent with input signal length, it is assumed that the hidden layer neuron of SAE1
Quantity is m, the weight matrix that W is n × m in corresponding (1) and (2) formula, and W' is the matrix of m × n, and SAE2 and SAE3 is also with such
It pushes away, the learning rate μ of every layer of SAE is set as 0.5, and the quantity of the number of iterations can be selected by way of experiment in actual use
Optimal quantity.
3) the unsupervised layer-by-layer pre-training of greediness is carried out to SDAE network, until all noise reduction autocoders are completed in training
(DAE).Using parameters such as the weight of BP algorithm fine tuning whole network and biasings, until reaching expected accuracy rate requirement.
The training of SDAE includes two steps: unsupervised pre-training is finely tuned with there is supervision.First with no label data pair
The each layer of layer-by-layer pre-training of DAE goes learning characteristic, the layer-by-layer Training strategy of unsupervised Greedy of use, each time individually training
One layer, and using training result as the input of higher level, then arrive top use instead supervised learning it is top-down to model into
Row fine tuning study.Unsupervised pre-training schematic diagram is as shown in figure 3, specific training process is as follows:
(1) it is obtained using the DAE1 that no label data collection x training is located at the bottom by the learning process of coding and decoding
To first character representation layer h1, since the limitation and adding of model capability is made an uproar operation, DAE1 is enable to learn to data itself
Structure, to obtain the feature for having more expression ability than inputting;
(2) by h1As the input of DAE2 model, continues training study and obtain the 2nd character representation layer h2;
(3) until training the 3rd DAE model of completion, the 3rd character representation layer h is obtainedn。
3 character representation layers are finally obtained, each character representation layer represents the different characteristic table of original input data
Show, also corresponds to the characteristic information of different stage.
Supervision fine tuning is carried out after layer-by-layer pre-training, successively heap comes by all character representation layers, and input has
Label data, by thering is the softmax of supervision to return layer to the weight of character representation layer and bigoted being finely adjusted.It is passed using reversed
It broadcasts algorithm to finely tune network, the update of weight is carried out using gradient descent algorithm, the process of algorithm is as follows:
I. to each node i of output layer lnl, residual error formula is
Ii. for hidden layer l=nl-1,nl- 2 ..., 2, residual expression is
I in formula (10), j respectively represent j-th of node of hidden layer l and j-th of node of hidden layer l+1,1 < j≤sl,
ρjRepresent the average activation value of j-th of node.
Iii. loss function takes partial derivative to W, b
In formula, C (W, b;X, y) it is the mean square error function output and input.
Iv. parameter update is carried out according to the following formula, until reaching expected accuracy rate requirement.
In formula, parameter η is learning rate when updating.The deep layer that the SDAE network finely tuned can extract input data is special
Sign, then using this feature as the input of LS-SVM classifier.
4) LS-SVM multi-categorizer model is established, the kernel function and assembly coding method of multi-categorizer is determined, utilizes grain
Subgroup optimization algorithm finds the kernel functional parameter of classifier and the optimum combination value of penalty.Finally by trained LS-SVM
Classification layer of the classifier as SDAE network, obtains final SDAE-LS-SVM network.
LS-SVM classifier is the deformation of SVM, and Quadratic Programming Solution problem reduction is Solving Linear by it, is avoided
The use of insensitive loss function, greatly simplifies the complexity of calculating, furthermore it also has extensive to Small Sample Database
The features such as ability is strong and training speed is fast.If training dataset are as follows:
S={ (xi,yi)|xi∈Rl,yi∈ { -1 ,+1 }, i=1,2 ..., l } (10)
Wherein xiFor the training sample input of l dimension, yiFor training sample output, l is sample number.The objective optimization of LS-SVM
Function are as follows:
Wherein:For nuclear space mapping function, ω is weight vector, eiFor error variance, b is bigoted amount, and μ and γ are can
Adjust parameter.
Lagrange equation is constructed, problem is converted are as follows:
Wherein: αiFor Lagrange multiplier, partial differential is asked to ω, b, e, α respectively
It can obtain α and b by solving above formula and be so used for the LS-SVM model of Function Estimation and can be expressed as
Wherein: K (x, xi) it is kernel function, Radial basis kernel function is used herein:
LS-SVM realizes that more classification pass through the method for Minimum Output Coding (MOC), and this method calculation amount is small and easy to accomplish.
When classifying to K0 class problem, this method all distributes each classification one exclusive codingWherein ykn∈ { -1,1 } respectively represents positive class and negative class, the output bit of MOC
Number is Nm=[log2K0]。
LS-SVM model Kernel Function parameter and penalty are affected to the performance of classifier, using PSO algorithm pair
The two parameters are automatically adjusted, and find the optimal value of parameter.
There to be label data collection { (x firsti,yi),…,(xN,yN) the trained SDAE network of input, it then will be each
Group input data xiThe output z obtained by SDAEiWith the output y of known sampleiConstitute the data set of LS-SVM parameter optimization
{(zi,yi),…,(zN,yN)}.Measurement standard during PSO optimization LS-SVM just has averagely exhausted degree prediction error (MAE)
It measures:
In formula: yiFor the output of known sample, yi' the prediction for being LS-SVM exports, i.e. model output value.To improve algorithm
Generalization Capability, in parameter optimization use cross validation.
Fig. 4 is the algorithm flow schematic diagram that particle swarm optimization algorithm optimizes LS-SVM parameter, the specific steps are as follows:
(1) PSO algorithm parameter, including population size N, Studying factors c1 and c2, maximum number of iterations, inertia power are initialized
The maximum value ω of weightmaxWith minimum value ωmin, initialization inertia weight, particle initial velocity vid and position xid.
(2) by the regularization parameter γ of LS-SVM and kernel function width square σ2As the two-dimensional coordinate of each particle, root
It according to training data training LS-SVM, and carries out staying a cross validation, calculates the fitness f of particle.
(3) to each particle, by fitness f (xi) be compared with itself optimal value, it updates its own and is preferably adapted to be worth;
The value that is preferably adapted to of each particle is compared with global best fitness, the overall situation of Population Regeneration is preferably adapted to be worth.
(4) particle rapidity vid and position xid is updated by formula (7)~(9).
(5) inertia weight ω is updated by formula (10), and updates the average value f for calculating fitnessavgWith minimum value fmin。
(6) judge whether to meet termination condition, optimal classification precision is exported if meeting, if being unsatisfactory for jumping to step
b).It generally sets termination condition and is less than given accuracy to reach maximum number of iterations or fitness.
(7) the most optimized parameter γ and σ are selected2, establish LS-SVM identification model.
Using trained LS-SVM classifier as the classification layer of SDAE network, final SDAE-LS-SVM network is obtained,
Specific network structure is as shown in Figure 5.It will be input to network by pretreated sound emission data, and then export stress metal
The leakage of the acoustic emission source of container and disturbed condition are divided into a few class states of table 2.
2 pressure vessel acoustic emission signal source category label coding table of table
5) after the data for acquiring acoustic emission sensor in real time are pre-processed, trained network is revealed for input
On-line checking, the leak case of output pressure container.
Claims (6)
1. a kind of on-line checking side of the sound emission leakage based on the metal pressure container for stacking noise reduction self-encoding encoder and LS-SVM
Method, which comprises the steps of:
1) leakage, noise and the normal sound generated by the acoustic emission sensor acquisition being mounted on pressure vessel by excitation
Emit signal, then carries out data prediction;
2) parameters such as SDAE network depth, quantity, learning rate and the number of iterations of each layer neuron are determined, are set according to mission requirements
Fixed suitable cost function and optimisation strategy;
3) the unsupervised layer-by-layer pre-training of greediness is carried out to SDAE network, until all noise reduction autocoders are completed in training.It adopts
With parameters such as the weight of BP algorithm fine tuning whole network and biasings, required until reaching expected accuracy rate;
4) LS-SVM multi-categorizer model is established, the kernel function and assembly coding method of multi-categorizer is determined, utilizes population
Optimization algorithm finds the kernel functional parameter of classifier and the optimum combination value of penalty.Finally trained LS-SVM is classified
Classification layer of the device as SDAE network, obtains final SDAE-LS-SVM network;
5) after the data for acquiring acoustic emission sensor in real time are pre-processed, trained network reveal online for input
Detection, the leak case of output pressure container.
2. a kind of according to claim 1, the online inspection of sound emission leakage of metal pressure container based on deep neural network
Survey method, which is characterized in that the step 1) leakage, the acquisition of noise and normal acoustic emission signal, it is contemplated that be difficult to directly
The information really revealed is obtained, therefore considers to obtain corresponding acoustic emission signal data by various energisation modes.By impactometer
Number, event count, amplitude, energy counting, duration, rise time, RMS voltage, average signal level this eight parameters
Characteristic parameter as acoustic emission signal.The characteristic parameter for all kinds of sound emission data that will acquire is used as no label data and is used for
SDAE network pre-training is randomly selected all kinds of acoustic emission signal production label datas in part and is finely tuned for network, before furthermore training
It also needs that data are normalized, formula is as follows:
3. according to claim 1, a kind of sound emission leakage of the metal pressure container based on deep neural network is online
Detection method, which is characterized in that the step 2) AE is made of three layers of feed forward type neural network, and training process mainly includes coding
Stage and decoding stage are first carried out plus are made an uproar processing to being originally inputted before DAE coding training, to original input sample x according to
qDRandom noise is added with probability ν (also referred to as noise level) in distribution, it is made to become noisy sampleRespectively plus
Make an uproar 5%, 10%, 20% noise is trained, select wherein effect it is best as final ν.
Initial data x and reconstruct data are allQuantity n it is consistent with input signal length, the hidden layer neuron quantity of SAE1 is
M, the weight matrix that W is n × m in corresponding (2) and (3) formula, and W' is the matrix of m × n, SAE2 and SAE3 also and so on, often
The learning rate μ of layer SAE is set as 0.5, and the quantity of the number of iterations can be selected optimal by way of experiment in actual use
Quantity.
4. according to claim 1, a kind of sound emission leakage of the metal pressure container based on deep neural network is online
Detection method, which is characterized in that step 3) the SDAE network carries out the unsupervised layer-by-layer pre-training of greediness, is finely tuned using BP algorithm
The parameters such as the weight of whole network and biasing.3 character representation layers, each character representation layer is finally obtained in unsupervised pre-training
The different characteristic that represent original input data indicates.It has supervision trim process parameter more new formula as follows:
In formula, parameter η is learning rate when updating.The SDAE network finely tuned can extract the further feature of input data, so
Afterwards using this feature as the input of LS-SVM classifier.
5. according to claim 1, a kind of sound emission leakage of the metal pressure container based on deep neural network is online
Detection method, which is characterized in that the step 4) LS-SVM realizes that more classification pass through the method for Minimum Output Coding (MOC), benefit
The kernel functional parameter of classifier and the optimum combination value of penalty are found with particle swarm optimization algorithm.PSO optimizes LS-SVM mistake
Measurement standard in journey just has averagely exhausted degree prediction error (MAE) Lai Hengliang:
6. according to claim 1, a kind of sound emission leakage of the metal pressure container based on deep neural network is online
Detection method, which is characterized in that after the data that the step 5) acquires acoustic emission sensor in real time are pre-processed, input is
Trained network carries out leakage on-line checking, and the status categories of output pressure container are divided into normal, leakage, crackle, electromagnetism are made an uproar
Several classification results such as sound, water droplet, friction, composite interference.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910304138.3A CN110044554A (en) | 2019-04-16 | 2019-04-16 | A kind of online test method of the metal pressure container leakage based on acoustic emission signal |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910304138.3A CN110044554A (en) | 2019-04-16 | 2019-04-16 | A kind of online test method of the metal pressure container leakage based on acoustic emission signal |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110044554A true CN110044554A (en) | 2019-07-23 |
Family
ID=67277342
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910304138.3A Pending CN110044554A (en) | 2019-04-16 | 2019-04-16 | A kind of online test method of the metal pressure container leakage based on acoustic emission signal |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110044554A (en) |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110455512A (en) * | 2019-08-20 | 2019-11-15 | 西安电子科技大学 | The mostly integrated method for diagnosing faults of rotating machinery based on depth self-encoding encoder DAE |
CN110487911A (en) * | 2019-08-19 | 2019-11-22 | 重庆大学 | The method of pressure vessel acoustic emission signal detection based on blind source separating |
CN110501122A (en) * | 2019-08-19 | 2019-11-26 | 重庆大学 | A kind of adaptive leakage detection method of pressure vessel based on cluster |
CN110568082A (en) * | 2019-09-02 | 2019-12-13 | 北京理工大学 | cable wire breakage distinguishing method based on acoustic emission signals |
CN110726518A (en) * | 2019-10-24 | 2020-01-24 | 中国核动力研究设计院 | Positioning and monitoring system for leakage of annular sealing surface of nuclear reactor pressure vessel |
CN111256906A (en) * | 2020-02-17 | 2020-06-09 | 金陵科技学院 | Decoupling method of multidimensional force sensor based on stack sparse self-coding |
CN111272886A (en) * | 2020-03-24 | 2020-06-12 | 华东理工大学 | Pressure vessel damage positioning method based on convolutional neural network |
CN111310830A (en) * | 2020-02-17 | 2020-06-19 | 湖北工业大学 | Combine harvester blocking fault diagnosis system and method |
CN111323220A (en) * | 2020-03-02 | 2020-06-23 | 武汉大学 | Fault diagnosis method and system for gearbox of wind driven generator |
CN112014478A (en) * | 2020-08-28 | 2020-12-01 | 武汉大学 | Defect echo blind extraction self-adaptive method submerged in ultrasonic grass-shaped signal |
CN112431726A (en) * | 2020-11-22 | 2021-03-02 | 华能国际电力股份有限公司 | Method for monitoring bearing state of gearbox of wind turbine generator |
CN112464837A (en) * | 2020-12-03 | 2021-03-09 | 中国人民解放军战略支援部队信息工程大学 | Shallow sea underwater acoustic communication signal modulation identification method and system based on small data samples |
CN112634945A (en) * | 2020-12-15 | 2021-04-09 | 浙江和达科技股份有限公司 | Intelligent water leakage sound identification method based on cloud platform |
CN112926504A (en) * | 2021-03-23 | 2021-06-08 | 重庆商务职业学院 | Acoustic emission signal denoising method based on noise reduction self-encoder |
CN112986393A (en) * | 2021-02-22 | 2021-06-18 | 深圳市市政设计研究院有限公司 | Bridge inhaul cable damage detection method and system |
CN113052207A (en) * | 2021-03-10 | 2021-06-29 | 中国特种设备检测研究院 | Stainless steel environment cracking judgment method and system based on Hybrid idea |
CN113088359A (en) * | 2021-03-30 | 2021-07-09 | 重庆大学 | Triethylene glycol loss online prediction method of triethylene glycol dehydration device driven by technological parameters |
CN113466022A (en) * | 2020-03-31 | 2021-10-01 | 丰田自动车株式会社 | Pressure testing method and pressure testing device |
CN113657019A (en) * | 2021-07-06 | 2021-11-16 | 大唐互联科技(武汉)有限公司 | Heat supply pipe network early warning system |
CN113916463A (en) * | 2021-09-01 | 2022-01-11 | 西安邮电大学 | Leakage detection method for hazardous chemical transportation storage tank |
CN114417688A (en) * | 2022-01-26 | 2022-04-29 | 重庆大学 | Valve internal leakage rate detection method based on acoustic emission |
CN114526451A (en) * | 2022-02-21 | 2022-05-24 | 南京邮电大学 | Underground space rock mass pipeline water seepage acoustic emission fluctuation level identification method and device |
CN116448338A (en) * | 2023-04-06 | 2023-07-18 | 中国特种设备检测研究院 | Acoustic emission detection method for full-capacity LNG storage tank leakage |
WO2024037853A1 (en) * | 2022-08-16 | 2024-02-22 | Robert Bosch Gmbh | Diagnostic system and diagnostic method for determining a state of a pressurised gas tank made of fibre-reinforced plastic |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009243994A (en) * | 2008-03-31 | 2009-10-22 | Univ Nihon | Fluid-tight inspection device after piping repair |
CN104634879A (en) * | 2015-02-04 | 2015-05-20 | 北京科技大学 | Metallic material fatigue loading testing and fatigue damage nondestructive testing analytical method |
CN105469049A (en) * | 2015-11-24 | 2016-04-06 | 昆明理工大学 | Leakage sound emission signal identification method based on multi-scale morphological decomposition energy spectrum entropy and support vector machine |
CN105488520A (en) * | 2015-11-24 | 2016-04-13 | 昆明理工大学 | Multi-resolution singular-spectrum entropy and SVM based leakage acoustic emission signal identification method |
CN107884475A (en) * | 2017-10-18 | 2018-04-06 | 常州大学 | A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net |
CN109406637A (en) * | 2018-11-05 | 2019-03-01 | 中国计量大学 | A kind of experimental provision for horizontal tank corrosion leakage Acoustic Emission location |
-
2019
- 2019-04-16 CN CN201910304138.3A patent/CN110044554A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009243994A (en) * | 2008-03-31 | 2009-10-22 | Univ Nihon | Fluid-tight inspection device after piping repair |
CN104634879A (en) * | 2015-02-04 | 2015-05-20 | 北京科技大学 | Metallic material fatigue loading testing and fatigue damage nondestructive testing analytical method |
CN105469049A (en) * | 2015-11-24 | 2016-04-06 | 昆明理工大学 | Leakage sound emission signal identification method based on multi-scale morphological decomposition energy spectrum entropy and support vector machine |
CN105488520A (en) * | 2015-11-24 | 2016-04-13 | 昆明理工大学 | Multi-resolution singular-spectrum entropy and SVM based leakage acoustic emission signal identification method |
CN107884475A (en) * | 2017-10-18 | 2018-04-06 | 常州大学 | A kind of city gas pipeline method for diagnosing faults based on deep learning neutral net |
CN109406637A (en) * | 2018-11-05 | 2019-03-01 | 中国计量大学 | A kind of experimental provision for horizontal tank corrosion leakage Acoustic Emission location |
Non-Patent Citations (1)
Title |
---|
张庆磊 等: "基于SDAE-SVM的压力机轴承故障诊断方法", 《数字制造科学》 * |
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110487911A (en) * | 2019-08-19 | 2019-11-22 | 重庆大学 | The method of pressure vessel acoustic emission signal detection based on blind source separating |
CN110501122A (en) * | 2019-08-19 | 2019-11-26 | 重庆大学 | A kind of adaptive leakage detection method of pressure vessel based on cluster |
CN110455512A (en) * | 2019-08-20 | 2019-11-15 | 西安电子科技大学 | The mostly integrated method for diagnosing faults of rotating machinery based on depth self-encoding encoder DAE |
CN110455512B (en) * | 2019-08-20 | 2021-01-05 | 西安电子科技大学 | Rotary mechanical multi-integration fault diagnosis method based on depth self-encoder DAE |
CN110568082A (en) * | 2019-09-02 | 2019-12-13 | 北京理工大学 | cable wire breakage distinguishing method based on acoustic emission signals |
CN110726518A (en) * | 2019-10-24 | 2020-01-24 | 中国核动力研究设计院 | Positioning and monitoring system for leakage of annular sealing surface of nuclear reactor pressure vessel |
CN110726518B (en) * | 2019-10-24 | 2021-05-28 | 中国核动力研究设计院 | Positioning and monitoring system for leakage of annular sealing surface of nuclear reactor pressure vessel |
CN111310830A (en) * | 2020-02-17 | 2020-06-19 | 湖北工业大学 | Combine harvester blocking fault diagnosis system and method |
CN111256906A (en) * | 2020-02-17 | 2020-06-09 | 金陵科技学院 | Decoupling method of multidimensional force sensor based on stack sparse self-coding |
CN111323220B (en) * | 2020-03-02 | 2021-08-10 | 武汉大学 | Fault diagnosis method and system for gearbox of wind driven generator |
CN111323220A (en) * | 2020-03-02 | 2020-06-23 | 武汉大学 | Fault diagnosis method and system for gearbox of wind driven generator |
CN111272886A (en) * | 2020-03-24 | 2020-06-12 | 华东理工大学 | Pressure vessel damage positioning method based on convolutional neural network |
CN113466022A (en) * | 2020-03-31 | 2021-10-01 | 丰田自动车株式会社 | Pressure testing method and pressure testing device |
CN112014478A (en) * | 2020-08-28 | 2020-12-01 | 武汉大学 | Defect echo blind extraction self-adaptive method submerged in ultrasonic grass-shaped signal |
CN112431726A (en) * | 2020-11-22 | 2021-03-02 | 华能国际电力股份有限公司 | Method for monitoring bearing state of gearbox of wind turbine generator |
CN112464837A (en) * | 2020-12-03 | 2021-03-09 | 中国人民解放军战略支援部队信息工程大学 | Shallow sea underwater acoustic communication signal modulation identification method and system based on small data samples |
CN112634945A (en) * | 2020-12-15 | 2021-04-09 | 浙江和达科技股份有限公司 | Intelligent water leakage sound identification method based on cloud platform |
CN112986393A (en) * | 2021-02-22 | 2021-06-18 | 深圳市市政设计研究院有限公司 | Bridge inhaul cable damage detection method and system |
CN112986393B (en) * | 2021-02-22 | 2023-09-29 | 深圳市市政设计研究院有限公司 | Bridge inhaul cable damage detection method and system |
CN113052207A (en) * | 2021-03-10 | 2021-06-29 | 中国特种设备检测研究院 | Stainless steel environment cracking judgment method and system based on Hybrid idea |
CN112926504A (en) * | 2021-03-23 | 2021-06-08 | 重庆商务职业学院 | Acoustic emission signal denoising method based on noise reduction self-encoder |
CN113088359A (en) * | 2021-03-30 | 2021-07-09 | 重庆大学 | Triethylene glycol loss online prediction method of triethylene glycol dehydration device driven by technological parameters |
CN113657019A (en) * | 2021-07-06 | 2021-11-16 | 大唐互联科技(武汉)有限公司 | Heat supply pipe network early warning system |
CN113916463A (en) * | 2021-09-01 | 2022-01-11 | 西安邮电大学 | Leakage detection method for hazardous chemical transportation storage tank |
CN113916463B (en) * | 2021-09-01 | 2023-09-19 | 西安邮电大学 | Dangerous chemical transportation storage tank leakage detection method |
CN114417688A (en) * | 2022-01-26 | 2022-04-29 | 重庆大学 | Valve internal leakage rate detection method based on acoustic emission |
CN114417688B (en) * | 2022-01-26 | 2024-06-21 | 重庆大学 | Acoustic emission-based valve internal leakage rate detection method |
CN114526451A (en) * | 2022-02-21 | 2022-05-24 | 南京邮电大学 | Underground space rock mass pipeline water seepage acoustic emission fluctuation level identification method and device |
WO2024037853A1 (en) * | 2022-08-16 | 2024-02-22 | Robert Bosch Gmbh | Diagnostic system and diagnostic method for determining a state of a pressurised gas tank made of fibre-reinforced plastic |
CN116448338A (en) * | 2023-04-06 | 2023-07-18 | 中国特种设备检测研究院 | Acoustic emission detection method for full-capacity LNG storage tank leakage |
CN116448338B (en) * | 2023-04-06 | 2024-01-30 | 中国特种设备检测研究院 | Acoustic emission detection method for full-capacity LNG storage tank leakage |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110044554A (en) | A kind of online test method of the metal pressure container leakage based on acoustic emission signal | |
CN111830408B (en) | Motor fault diagnosis system and method based on edge calculation and deep learning | |
Li et al. | A deep transfer nonnegativity-constraint sparse autoencoder for rolling bearing fault diagnosis with few labeled data | |
CN101587155B (en) | Oil soaked transformer fault diagnosis method | |
CN111476263A (en) | Bearing defect identification method based on SDAE and improved GWO-SVM | |
CN107316046A (en) | A kind of method for diagnosing faults that Dynamic adaptiveenhancement is compensated based on increment | |
CN104616033A (en) | Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine) | |
Li et al. | IncepTCN: A new deep temporal convolutional network combined with dictionary learning for strong cultural noise elimination of controlled-source electromagnetic data | |
CN109740254B (en) | Ship diesel engine abrasive particle type identification method based on information fusion | |
CN110096810A (en) | A kind of industrial process flexible measurement method based on layer-by-layer data extending deep learning | |
Kraljevski et al. | Machine learning for anomaly assessment in sensor networks for NDT in aerospace | |
CN110533007A (en) | A kind of vehicle-mounted strain of bridge influences the identification of line feature intelligent and extracting method | |
CN113469470A (en) | Energy consumption data and carbon emission correlation analysis method based on electric brain center | |
Tian et al. | Joint learning model for underwater acoustic target recognition | |
CN112799130A (en) | Seismic wave amplitude prediction method based on deep learning | |
CN103971133A (en) | Automatic steel plate surface defect recognition method based on case-based reasoning | |
Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
CN117370766A (en) | Satellite mission planning scheme evaluation method based on deep learning | |
Feng | Bank Green Credit Risk Assessment and Management by Mobile Computing and Machine Learning Neural Network under the Efficient Wireless Communication | |
CN109636194B (en) | Multi-source cooperative detection method and system for major change of power transmission and transformation project | |
Yao et al. | A Hybrid Approach for Fault Diagnosis of Railway Rolling Bearings Using STWD‐EMD‐GA‐LSSVM | |
CN115017984A (en) | Early warning method and system for failure risk of aircraft engine | |
CN110487911A (en) | The method of pressure vessel acoustic emission signal detection based on blind source separating | |
CN113108949A (en) | Model fusion-based sonde temperature sensor error prediction method | |
CN113607068A (en) | Method for establishing and extracting recognition model of photoacoustic measurement signal characteristics |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190723 |
|
WD01 | Invention patent application deemed withdrawn after publication |