CN109991306A - The Classification and Identification and positioning of metal works welding defect based on fluorescentmagnetic particle(powder) - Google Patents
The Classification and Identification and positioning of metal works welding defect based on fluorescentmagnetic particle(powder) Download PDFInfo
- Publication number
- CN109991306A CN109991306A CN201711469746.7A CN201711469746A CN109991306A CN 109991306 A CN109991306 A CN 109991306A CN 201711469746 A CN201711469746 A CN 201711469746A CN 109991306 A CN109991306 A CN 109991306A
- Authority
- CN
- China
- Prior art keywords
- defect
- welding
- powder
- identification
- classification
- 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
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/72—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables
- G01N27/82—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws
- G01N27/83—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws by investigating stray magnetic fields
- G01N27/84—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws by investigating stray magnetic fields by applying magnetic powder or magnetic ink
Landscapes
- Chemical & Material Sciences (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Or Analyzing Materials By The Use Of Magnetic Means (AREA)
Abstract
Magnetic powder inspection is widely used in the detection of bearing weld crack, but whether defect is still artificial progress visual inspection.The present invention is directed to the longitudinal measure defect (such as crackle, stomata, slag inclusion) that metal works generate in the welding process, metal works face magnetic trace image is shot using industrial camera, the feature of doubtful defect area is extracted with digital image processing techniques, these features are then based on, the automatic identification and defective locations positioning of cylinder crackle are completed using classifier (MLP) technology of neural network.The experimental results showed that not only true, false defect has higher discrimination, but also improves detection efficiency this method to welding defect, metal cutting and metal heat treatmet etc., there is certain application study meaning.
Description
Technical field
The present invention relates to intelligent measurement fields, are a kind of dysnusia identifying systems based on fluorescentmagnetic particle(powder), this can be made soft
Part system completes the Classification and Identification detection and defect location of metal works welding defect in the environment of factory, belongs to machine recognition
Technical field.
Background technique
Welding is the multivariable of a highly linear, close coupling, while the complexity again with a large amount of random uncertain factors
Process, so detection difficulty is larger, with the quick development of China's manufacturing industry, welding technical ability uses more and more extensive, welding skill
It can be horizontal also higher and higher.New welding technical ability method continues to bring out, and professional welding technique is also towards digitlization, intelligence side
To continuous development, traditional human eye judgement is no longer satisfied the needs of social development, and there is also many defects.
Due to the continuous development of detection technique and intellectualized technology, people start to explore the side of some automatic detections
The characteristics of method, the stochastic uncertainty that some occur by welding defect, carries out welding defect with statistical analysis technique
Research, establishes the welding defect online recognition method based on statistical method, and also someone passes through radiographic inspection and ultrasonic wave
Detection technique proposes the welding defect online test method based on radiographic inspection and ultrasonic wave, and there are also some to propose based on small
Wave theory etc., although these methods promote the development of detection technique, there is also respective some disadvantages: testing cost
Height, complex procedures, detection efficiency is low, automation difficult to realize etc..
With the development of nerual network technique, people, which start to explore, is used in nerual network technique in welding detection, base
It is a highly complex nonlinear dynamic system in neural network, other than the community with General Nonlinear Systems, goes back
Have the advantages that it is adaptive, self-organization, manual intervention is few, precision is high.Fuzzy logic has the ability for imitating human thinking,
Fuzzy logic on the basis of this combines generated fuzzy neural network with neural network and lacks in Control Welding Process and welding
It falls into identification aspect and achieves many progress, the country, which has, welds GTAW process using self-organizing feature map (SCM) neural network
Defect is detected automatically, is joined by the observation to time interval parameter probability density distribution feature each in welding process according to electricity
The distribution of number histogram identifies whether the defect of corresponding position while welding occurs, and in the case where no information provides, is closed by selection
Suitable output node number realizes automatic classification, and stabilization signal is completely separable next, and also useful wavelet transformation analysis method is extracted different
Acoustic wave energy in frequency range passes through Establishment of Neural Model feature vector to splashing as characterization splashing size characteristic vector
The mapping model of amount, so as to CO2Spatter amount gives a forecast, the results showed that it reflects, it can be accurate using welding arc acoustics signal
Prediction spatter, be the new way for realizing welding quality On-line Control, realize real to on-line automaticization of welding process
When control, using nerual network technique can many experiments can obtain satisfactory as a result, external studying so that it goes without doing
It is handled when the detection of welding defect by the picture of X-ray neural network based, defect characteristic is extracted, further according to weld seam
The characteristics of defect sample, establishes " one-to-one " the cluster structure of SVM and then to specimen discerning, or utilizes decision tree and fuzzy matrix
It is identified, can also obtain good identification and classification results, but X-ray itself has harmfulness, use scope to human body
Also there is limitation, be difficult to be widely used in industrialization detection.The defect welded on metal works is proposed herein a kind of new
The intelligent recognition and positioning system based on fluorescentmagnetic particle(powder) of type, detection method is simple, and non-hazardous property, detection effect is good, has wide
General applicability.
Summary of the invention
A kind of view-based access control model system and image processing techniques are provided it is an object of the invention to solve above-mentioned technical problem
And a kind of Intelligent welding defects detection and welding defect localization method of machine learning.
The technical scheme is that workpiece slot pickup is placed at fixed magnetization contact using manipulator, movable magnetic
When changing contact and extending to contact workpiece top position, magnetization circuit starts to carry out transverse and longitudinal magnetization to workpiece, completes living after magnetization
Dynamic magnetization contact retracts original place.Manipulator will magnetize workpiece send to mesh mounting table after return to workpiece slot crawl it is next to magnetic
The welding workpiece of change.Meanwhile magnetic flaw detection ink flusher can spray oiliness magnetic flaw detection ink to the magnetization workpiece on mounting table, make its welding
Magnetic flaw detection ink is come into full contact with, netted mounting table has magnetic flaw detection ink circulation and stress function.Then, manipulator will spray the workpiece of magnetic flaw detection ink
Automatic band is to detection position, and under annular black light light irradiation, three industrial cameras around workpiece start to take pictures to workpiece
Magnetic trace image is obtained, takes pictures and completes later image data and be sent to computer through interchanger, magnetizing system constantly repeats just now
Work.
The picture for reaching computer is pre-processed: pretreated purpose is to improve magnetic trace picture quality, to inhibit
Irrelevant information and Enhanced feature information, the quality of processing result directly influence subsequent feature extraction.Here by reality
Verifying, the G channel image after separation is especially good in the interference of anti-purple light, so isolating the color image in the channel G first here
As next processing picture；In magnetic trace Image Acquisition link, due to acquisition ambient black, camera optoelectronic noise, camera lens by
To the factors such as isolated magnetic powder particle of magnetic flaw detection ink pollution, workpiece welding absorption, cause magnetic trace image there are salt-pepper noise, gray scales not
Situations such as and contrast is unobvious.And adaptive median filter has very strong noise removal capability for salt-pepper noise, is usually used in
The protection of marginal information, but since the defect of some forms is very subtle, be of great significance in carrying out classification realization, and want
Some useful information can be removed while salt-pepper noise may be removed using median filtering, so cooperating mean filter here
Be used together, can not only denoise but also can retain useful information in this way, also for can extract specific location and be split must
Want step；Wiring processes the image into carry out Morphological scale-space, threshold values processing etc. again and is easy to neural network differentiation and handles
Image.
Next MLP neural network progress defect Classification and Identification and defect position extracting is utilized: to having completed to pre-process
Picture afterwards carries out the work of two aspects, on the one hand passes through training to MLP neural network using training sample, then pass through survey
The center and minimum rectangle coordinate position, MLP neural network for trying sample extraction sample defect are anti-for what is learnt
Be standard supervision learning algorithm in the field of pattern-recognition to propagation algorithm, and have verified that calculate neurology and
In parallel and distributed process field, MLP has proved to be a kind of general function approximation method, can be used to fitting complexity
Function, can be by by multiple perceptron groups although single perceptron can not solve XOR problem to solve classification problem
It closes, realizes the segmentation of complex space.Here use based on neural network algorithm to gray-scale image segmentation method, to grey image into
Row dividing processing, can extract the feature of each defect of detection image after MLP neural metwork training, formed feature to
Amount, then by respective algorithms extract each feature of correspondence for the center of defect and the minimum rectangle coordinate of defect,
To realize the positioning of welding defect.On the other hand the feature of the welding defect extracted by the first step is as MLP neural network
Input layer by training to difference lack classify.Metal welding defect is divided into six major class, including crackle, stomata, folder
Miscellaneous, incomplete fusion, lack of penetration and other shape defects etc., distributional pattern difference different according to defect region, each major class lack
It include again different subclassifications in falling into, as dreg defect is divided into strip slag inclusion and spherical slag inclusion according to the difference of form.Computer
Quantitative calculating is good in detection, is bad at fuzzy Judgment.By the measurement and calculating of computer, except a small amount of defect can use determining numerical value
Definition is outer, and most of defect and being not suitable for calculation machine of using tricks measures and judges, so herein in conjunction with situation common in weld seam,
The main metal welding seam defect for discussing that following five major class are common: crackle, stomata, slag inclusion, incomplete fusion and lack of penetration, further according to passing through
The obtained information of first step MLP neural network establishes the circumscribed rectangular model of defect area, can obtain the geometry of defect
Feature includes defect mass center, defect area, defect perimeter, the long axis of circumscribed rectangle, short axle, area.Recycle its weld defect
Characteristic parameter modeled, calculate the geometrical characteristic parameter and parameters for shape characteristic of defect area, final mass center of choosing is to welding
Stitch center line distance, length-width ratio, short axle and defect area ratio, rectangular degree, circularity, Hai Wude diameter this six kinds of shape features pair
Different types of defect and noise are analyzed and are recognized, then the identification of defect kind is completed by MLP neural network.
By position of sound production to metal welding workpiece, magnetization, image data transmission, image procossing, utilize MLP nerve net
Network is to defect position extracting and defect kind differentiation, display processing result etc., the process to move in circles, thus the reality of detection system
When detect.
Specific embodiment
Technical solution of the present invention is described in more detail with reference to the accompanying drawings and detailed description.
, the structure chart that uses is as shown in Figure 1, mainly divide the spontaneous magnetization device of grabbing workpiece, magnetic flaw detection ink sprinkling herein
With four recycling, magnetic trace image online acquisition and software processing and identification etc. parts.
, grabbing workpiece spontaneous magnetization device
Manipulator is placed in from fixed magnetization contact from workpiece slot pickup, and activity magnetization contact is extended to contact workpiece top position
When, magnetization circuit starts to carry out transverse and longitudinal magnetization to workpiece, and activity magnetization contact retracts original place after completing magnetization.Manipulator is by magnetic
Chemical industry part returns to workpiece slot and grabs next welding workpiece to be magnetized after sending to mesh mounting table.
, magnetic flaw detection ink sprinkling and recycling
Magnetic flaw detection ink flusher can spray oiliness magnetic flaw detection ink to the magnetization welding workpiece on mounting table, connect its welding workpiece sufficiently
Magnetic flaw detection ink is touched, netted mounting table has magnetic flaw detection ink circulation and stress function.
, magnetic trace image online acquisition
Manipulator send the welding workpiece for spraying magnetic flaw detection ink to the center of circle of round platform, and three industrial cameras around welding workpiece are opened
Beginning takes pictures Welder's weldment to obtain magnetic trace image, and image data is sent to computer through interchanger.Here why
Three industrial cameras are taken, and the shooting for being on the one hand to welding workpiece be carried out within the scope of 360 degree obtains image, thus disappears
In addition to a certain defect because being on the other hand then to realize work for convenience in a certain position without the mistake being detected
Industry automation.
, software processing and identification: be mainly introduced in two sub-sections here, first part is main lacks
Sunken type and feature, second part are that MLP neural network carries out Classification and Identification and position is extracted.
, defect type and feature
Metal welding defect is divided into six major class, including crackle, stomata, is mingled with, incomplete fusion, lack of penetration and other shape defects
Deng, it is different according to defect region difference, distributional pattern, it again include different subclassifications, such as slag inclusion in each major class defect
Defect is divided into strip slag inclusion and spherical slag inclusion according to the difference of form.COMPUTER DETECTION is good at quantitative calculating, is bad to obscure and sentence
It is disconnected.By the measurement and calculating of computer, except a small amount of welding defect can be in addition to determining numerical value definition, most of welding defect is not
It is suitable for being measured and being judged with computer, so main following five major class of discussion are normal herein in conjunction with situation common in weld seam
The weld defect seen: crackle, stomata, slag inclusion, incomplete fusion and lack of penetration and its its feature are as follows:
, crackle
Due to the effect of welding stress and other factors, the combination in workpiece weldering chain region between metallic atom is destroyed, and is being connect
Form new interface at mouthful and generate gap, i.e. crackle, crackle by the difference of forming region and shape can be divided into melt run crackle,
Heat action zone crack, crater crack, plate portion crackle, transversal crack, longitudinal crack and laminar crack etc..Crack defect is in welding bead
And occur around heat-affected zone more.Crackle is shown generically as irregular black thin on picture, from centre to both ends
Black weakens, and one or both ends are tip, are parallel to each other not interlaced；
, stomata
Physical reactions present in welding process and chemical reaction can generate certain gas, these stomatas make the fusion area of weldment
And heat affected area generates hole.Situations such as welding unreasonable design, misoperation, unclean environment can all lead to the production of stomata
It is raw.Stomata can be divided into single stomata, porosity, strip blistering and linear porosity etc. by the difference of distribution situation and shape,
Again respectively in shapes such as circle, oval and strips when middle stomata is individually distributed.Gas cell distribution situation is indefinite, some single distributions have
It is a bit in intensive shape, generally the black of rule weakens with smooth image edge from centre to edge black on the image,
Boundary is obvious.
, slag inclusion
Required cosolvent, metallic steel tungsten in welding process weld generated oxide, nonmetallic slag inclusion, all can
It is that slag is generated in weld seam, to form dreg defect.It removes not in time and sweeps net residual sprue, electric current selects improper, groove angle
Degree selection is improper, welding misoperation, these are all slag inclusion Producing reasons, and dreg defect is broadly divided into strip slag inclusion and spherical
Slag inclusion, strip slag inclusion is mostly parallel with weld groove on picture, and Aspect Ratio value is greater than 3, and elongated, size trend is not solid
Fixed, corner angle are clearly irregular, and color is uniform, and corner blackness weakens, and image is more visible.Spherical slag inclusion is then that Aspect Ratio value is equal to
Or the image less than 3, size are not fixed with distributed areas, single or dense distribution, distribution of color is uneven, random side
Boundary.Both the two genetic facies are same, and difference when welding operates causes form is different；
, incomplete fusion
Incomplete fusion refers between base metals and weld metal or is not completely melt between weld metal and weld metal combination
Situation.It is distinguished according to defect position, incomplete fusion includes groove incomplete fusion, root incomplete fusion and lack of inter-run fusion.Preheating is not
Foot, surface layer cleaning, which are not thorough, electric arc is directed toward the reasons such as deflection can all cause the defect of incomplete fusion, and on picture, incomplete fusion defect is aobvious
Show that Weld pipe mill is deviateed in position, occurs the interrupted distribution of bar shaped, uneven color, close to base metals one side on the same line
Color is relatively deep and is linearly distributed；
, it is lack of penetration
According to the difference for generating position, incomplete penetration defect can be divided into lack of penetration edge, incomplete inter-run penetration and incomplete root penetration, groove
That too small, welding parameter is chosen is improper, speed of welding is excessive or welding current is too small, these reasons can all cause lack of penetration lack
It falls into.
, MLP neural network carry out Classification and Identification and position extraction
Multilayer perceptron (Multi-layer perception-MLP) neural network passes through to Bayes's MLP neural network
Study, finds it when handling a large amount of real time data, and precision with higher can also ensure that the target reached, under
A simple introduction is done in face of MLP neural network algorithm and its structure:
, MLP neural network algorithm process
In MLP, the corresponding response of sample should be a phasor, the quantity one of the neuron of the dimension and output layer of phasor
It causes.And the number of plies of hidden layer and the quantity of each layer neuron are then chosen according to the actual situation.The core concept of algorithm is:
Error is obtained by through path (direction of arrow), then the error back propagation is realized the amendment of weight.The error of MLP can
To be indicated with squared error function.If the corresponding target value of some sample x is t, sample x hasnA characteristic attribute, i.e. x=
{x 1, x 2,…,xn, target value t hasJKind possible values, i.e. t=t 1, t 2,…,tJ, therefore the input layer (i.e. of the MLP
One layer) it is one sharednA neuron, output layer (i.e.LLayer, if MLP mono- is sharedLLayer) it is one sharedJA neuron.If sample x is passed through
Cross final output that through path obtains be y=y 1 L ,y 2 L ,…,yiL, outputySubscript indicateyThe neuron rope of place layer
Draw, subscript indicatesyThe layer at place, the then square error of the sample are as follows:
(14)
Why the squared error function in formula (14) will be for the ease of subsequent derivative operation, because of its not shadow divided by 2
Ring the variation tendency of error.Obviously, the target of MLP algorithm is exactly to keep E minimum.YjL in (14) is by upper one layer (i.e.
L-1 layers) it obtains after the weighted activation of output of all neurons, and the output of L-1 layers of neuron is owned by L-2 layers
It is obtained after the weighted activation of the output of neuron, it can be said that error E is the function of all weight ws, by changing weight w,
Just can reach makes the smallest purpose of error E；MLP algorithm is a kind of method of iteration, that is, we change without going through primary
Weight w makes the smallest purpose of E to reach, we only need progressive reduction E.The relationship of w and E can image be compared to mountain
Slope, the height on mountain are errors, and the dimensional space of plane is weight, and hillside is steeper (error is big), the variation (power in planar dimensions space
The variation of value) it is bigger, the variation of weight is related with error, and when weight changes, error will recalculate.Both in this way
Interaction, i.e., continuous iteration, until error is less than some value (restraining)；
, MLP neural network block diagram
Feedforward neural network is one kind and a kind of most common neural network of MLP neural network.It includes an input
Layer, output layer and several hidden layers, as shown in figure 5, the MLP includes an input layer, an output layer and one it is hidden
Containing layer, wherein the neuron of a certain layer can only be connected to next layer of neuron by a direction.
Pass through MLP neural network and extract defect following six feature: is welding defect mass center, defect area, defect perimeter, outer
Cut long axis, short axle, the area of rectangle, so that it may which the convenient position for extracting defect calculates defect further according to these defect characteristics
The geometrical characteristic parameter and parameters for shape characteristic in region, it is final to choose mass center to Weld pipe mill linear distance, length-width ratio, short axle and lack
Fall into this six kinds of shape features progress defect kind identifications of area ratio, rectangular degree, circularity, Hai Wude diameter.
Two, as shown in Fig. 2, a kind of metal works welding defect Classification and Identification and positioning system frame based on fluorescentmagnetic particle(powder)
Figure, illustrates feature extraction, neural network in conjunction with the step of system block diagram:
, defect position extracting step:
, Image Acquisition
Manipulator is placed in from fixed magnetization contact from workpiece slot pickup, and activity magnetization contact is extended to contact workpiece top position
When, magnetization circuit starts to carry out transverse and longitudinal magnetization to workpiece, and activity magnetization contact retracts original place after completing magnetization.Manipulator is by magnetic
Chemical industry part returns to workpiece slot and grabs next workpiece to be magnetized after sending to mesh mounting table.Meanwhile magnetic flaw detection ink flusher
Oiliness magnetic flaw detection ink can be sprayed to the magnetization workpiece on mounting table, so that its welding is come into full contact with magnetic flaw detection ink, netted mounting table has magnetic
Suspension circulation recycles function.Then, by the workpiece for spraying magnetic flaw detection ink, band shines to detection position in annular black light lamp manipulator automatically
It penetrates down, three industrial cameras around workpiece start to be taken pictures to obtain magnetic trace image to workpiece；
, separation obtain G Channel Color picture
The image of workpiece is got from the specific environment of plant, in the RGB image to host computer for then transmitting magnetic trace.It carries out
The picture in the process channel analysis of experiments G is the smallest for light source interference before processing, since the advantage of fluorescentmagnetic particle(powder) is mainly magnetic
Powder imaging, if image, which will directly be carried out binary conversion treatment, to be interfered by very strong purple light, so that Magnetic testing cannot be embodied
Color image is split into R, G, channel B picture herein, and extracts the channel G by advantage, is primarily due to the color image in the channel G
It is the smallest for being disturbed；
, by extract color image be converted into binary image；
, filtering processing
Here carried out in such a way that adaptive median filter and mean filter combine, due to there are salt-pepper noises in picture must
Median filtering, which need be used, to be removed, but while removing noise can remove together some useful marginal informations
Fall, this is totally unfavorable for image segmentation, when being in workpiece edge especially for defect, so value filtering in use here
When to cooperate mean filter to be used together, it is as follows that both filter specific features:
The first, median filtering is a kind of nonlinear image smoothing method, with mean filter and other linear filter phases
Than it can filter out salt-pepper noise well, and median filtering may be defined as:
(1)
In formula:WithGrey scale pixel value is respectively exported and inputs, W is template window.Window W
It can be with line taking shape, rectangular, cross, circle, diamond shape etc.；
The second, Quick and equal filter method, form are as follows:
(2)
On a same row, when from a pixelIt is moved to next pixelWhen, the sum that adds up
Variation are as follows:
(3)
Wherein:Be the cumulative of the pixel for the left-hand column removed and；
Be the cumulative of the pixel of the right-hand column newly added and.
Mean filter can allow picture to thicken, but have no effect on and carry out image segmentation, such two kinds of filters here
The mode that wave combines is conducive to us and extracts the specific location of defect, lays the foundation for following defect position extracting；
, Morphological scale-space
Filtered image is passed through into the available opposite obvious shortcoming shape of opening operation and closed operation；
, two-value threshold segmentation
Self-adaption binaryzation segmentation, the median by obtaining picture minimal gray and maximum gray scale carry out threshold values processing, are formed
The defect shape being more clear；
, utilize that MLP neural network carries out sample training, test sample process carries out defect Segmentation and defect position extracting:
, carry out MLP neural network sample training and test sample
Neural network MLP training process is to contain multilayer node by multi-layer perception (MLP), next layer of section of every node layer and network
Point is fully connected.The node on behalf input data of input layer, the node of other layers is by by the power of node in input data and layer
Weight W and deviation b linear combination and one activation primitive of application, obtain this layer output.Multi-layer perception (MLP) by direction propagation come
Learning model, wherein we use logic loss function and L-BFGS, and K+1 layer multi-layer perceptron classifier can be write as matrix
Form is as follows:
(5)
Middle layer node uses sigmoid equation:
(6)
Output layer uses softmax equation:
(7)
Wherein N represents class number in output layer；
, neural network MLP segmentation.Picture after filtering passes through multilayer perceptron neural network (Multi-layer
Perceptron neural networks, MLP) basic model in each node be a perceptron, by by multiple senses
Know that device combines, realize the segmentation of the how middle feature space of defect, is combined by multiple perceptrons non-thread to realize by formula (4)
Property classifying face, wherein θ () indicates jump function or sign function:
(8)
, by MLP neural network to test sample.Show that classification results carry out sample training and obtain differentiation as a result, can sentence
Disconnected which sample out is defective；
, the feature that is obtained according to MLP neural network, and then coordinate position of the defect in picture can be obtained such as by calculating again
Fig. 4.
, defect kind identification step:
, by MLP neural network to defect position extracting, the description of associated shape, i.e. defect mass center, defect face can be obtained
Product, defect perimeter, the long axis of circumscribed rectangle, short axle, area；
, by analyzing all kinds of defective patterns, description and the shape feature for calculating defect, including in figure mass center to Weld pipe mill
Linear distance, length-width ratio, short axle and defect area ratio, rectangular degree, circularity, Hai Wude diameter, according to this six kinds of shape features point
Analyse and recognize different types of defect and noise；
, mass center to Weld pipe mill linear distance
Lack of penetration defect pattern is distributed on axis of a weld, calculates mass center to the distance of axis of a weld, can be to not welding
Saturating defect and other kinds of defect are distinguish, and indicate mass center to Weld pipe mill linear distance, formula with G [1] are as follows:
(9)
Wherein: Sx, Sy: defect center-of-mass coordinate;K: gained boundary slope after welded seam area extracts；B: welded seam area up-and-down boundary
Y-intercept average value；b1, b2: welded seam area up-and-down boundaryy Y-intercept;
, short axle and defect area ratio
The differentiation for the defects of above-mentioned circumscribed rectangle ratio of semi-minor axis length is convenient for spherical slag inclusion, stomata, and short axle and defect area ratio
Feature is distinguish the strips such as lack of penetration, crackle defect area and other types defect convenient for differentiation circle and strip region,
Short axle and defect area ratio, formula are indicated with G [2] are as follows:
(10)
, length-width ratio
Be described by flat degree of the ratio of semi-minor axis length to target area, distinguish crackle, stomata, it is lack of penetration and noise lack
It falls into, indicates ratio of semi-minor axis length, formula with G [3] are as follows:
(11)
, rectangular degree
Reflect defect area to the full level of circumscribed rectangle, therefore rectangular degree range is [0,1].When defect area in it is elongated,
Curved shape, rectangle angle value become smaller；When defect area is round, rectangular degree is；When rectangular degree is maximum value 1, defect area
Domain is full of entire circumscribed rectangle, indicates rectangular degree feature, formula with G [4] are as follows:
(12)
, circularity
For distinguishing the shape feature of crackle, stomata, spherical dreg defect, round degree, the description defect of defect area are indicated
The complexity on boundary indicates circularity feature, formula with G [5] are as follows:
(13)
, Hai Wude diameter
It is diameter of a circle identical with weld defect region area for the differentiation of weld defect and isolated noise point, with G [6]
Indicate Hai Wude diameter, formula are as follows:
(14)
The detection image of the weld defects such as crackle in database, stomata, mismatch is selected, is split processing to it as previously described simultaneously
Calculate its characteristic parameter；
, again by MLP neural network to characteristic parameter carry out defect kind judgement.
, as shown in figure 3, a kind of metal works welding defect Classification and Identification and positioning system process based on fluorescentmagnetic particle(powder)
Figure passes through image information collecting, magnetic trace image preprocessing (including separation and Extraction G channel image, binaryzation, filtering etc.), image
Defect kind identification and defect position extracting are arrived in segmentation, Morphological scale-space and feature extraction.
The foregoing is only a preferred embodiment of the present invention, the application scope of application of the invention is without being limited thereto, appoint
What those familiar with the art within the technical scope of the present disclosure, the technical solution that can be become apparent to
Simple change or equivalence replacement each fall in the application scope of application of the invention.
Claims (4)
1. a kind of judge welding defect type and positioning system based on the picture of the metal works of fluorescentmagnetic particle(powder), the system is main
Know including scene image information collection, magnetic trace image preprocessing, image segmentation, Morphological scale-space and feature extraction, crackle type
Other and these parts of defect position extracting.
2. the dysnusia Classification and Identification and positioning system according to claim 1 based on fluorescentmagnetic particle(powder), which is characterized in that
Using welding defect type and defective locations are judged based on the picture of the metal works of fluorescentmagnetic particle(powder), resisted by RGB separation
Pollute the interference of light source (such as fluorescence).
3. the dysnusia Classification and Identification and positioning system according to claim 1 based on fluorescentmagnetic particle(powder), which is characterized in that
The mode for having used median filtering and mean filter to combine during characteristic processing removes salt-pepper noise, Lai Tigao median filtering
In missing that the defect information at useful edge is disposed together, the effect for picture is very beneficial for segmentation that treated in this way
Fruit.
4. the dysnusia Classification and Identification and positioning system according to claim 1 based on fluorescentmagnetic particle(powder), which is characterized in that
The metal works welding defect Classification and Identification and positioning system utilization (Multi-layer based on fluorescentmagnetic particle(powder)
Perception-MLP) to treated, image is split nerve net, and Process fusion multilayer perceptron utilizes (MLP) later
The outstanding classification situation that neural network provides is carried out defect kind identification using MLP after image procossing, then passes through defect spy
Sign extracts center and minimum rectangle frame position, to obtain the position of defect and carry out defect kind judging.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711469746.7A CN109991306A (en) | 2017-12-29 | 2017-12-29 | The Classification and Identification and positioning of metal works welding defect based on fluorescentmagnetic particle(powder) |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711469746.7A CN109991306A (en) | 2017-12-29 | 2017-12-29 | The Classification and Identification and positioning of metal works welding defect based on fluorescentmagnetic particle(powder) |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109991306A true CN109991306A (en) | 2019-07-09 |
Family
ID=67108439
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711469746.7A Pending CN109991306A (en) | 2017-12-29 | 2017-12-29 | The Classification and Identification and positioning of metal works welding defect based on fluorescentmagnetic particle(powder) |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109991306A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110717872A (en) * | 2019-10-08 | 2020-01-21 | 江西洪都航空工业集团有限责任公司 | Method and system for extracting characteristic points of V-shaped welding seam image under laser-assisted positioning |
| CN110806736A (en) * | 2019-11-19 | 2020-02-18 | 北京工业大学 | Method for detecting quality information of forge pieces of die forging forming intelligent manufacturing production line |
| CN112435245A (en) * | 2020-11-27 | 2021-03-02 | 济宁鲁科检测器材有限公司 | Magnetic mark defect automatic identification method based on Internet of things |
| CN112816545A (en) * | 2020-09-30 | 2021-05-18 | 中国石油天然气股份有限公司 | Method and device for determining area of storage tank repairing plate |
| CN113390956A (en) * | 2021-06-18 | 2021-09-14 | 西安建筑科技大学 | Double-magnetic-sensor probe and magnetic flux leakage detection defect quantitative evaluation method based on probe |
| CN113740416A (en) * | 2021-08-27 | 2021-12-03 | 东风商用车有限公司 | Engine crankshaft inspection method |
| CN114166805A (en) * | 2021-11-03 | 2022-03-11 | 格力电器（合肥）有限公司 | NTC temperature sensor detection method and device, NTC temperature sensor and manufacturing method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002122571A (en) * | 2000-10-12 | 2002-04-26 | Kenzo Miya | Method and apparatus for inspection of defect |
| US6950545B1 (en) * | 1999-10-26 | 2005-09-27 | Hitachi, Ltd. | Nondestructive inspection method and apparatus |
| CN101320430A (en) * | 2008-06-25 | 2008-12-10 | 南昌大学 | Extraction algorithm for arc-removing of underwater weld joint tracing signal based on laser sensor |
| CN103984951A (en) * | 2014-04-25 | 2014-08-13 | 西南科技大学 | Automatic defect recognition method and system for magnetic particle testing |
| CN105090676A (en) * | 2015-07-12 | 2015-11-25 | 西南科技大学 | Pressure bearing pipe damage repair process |
| CN107145896A (en) * | 2017-03-14 | 2017-09-08 | 西南科技大学 | Dysnusia identifying system based on fluorescentmagnetic particle(powder) |
-
2017
- 2017-12-29 CN CN201711469746.7A patent/CN109991306A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6950545B1 (en) * | 1999-10-26 | 2005-09-27 | Hitachi, Ltd. | Nondestructive inspection method and apparatus |
| JP2002122571A (en) * | 2000-10-12 | 2002-04-26 | Kenzo Miya | Method and apparatus for inspection of defect |
| CN101320430A (en) * | 2008-06-25 | 2008-12-10 | 南昌大学 | Extraction algorithm for arc-removing of underwater weld joint tracing signal based on laser sensor |
| CN103984951A (en) * | 2014-04-25 | 2014-08-13 | 西南科技大学 | Automatic defect recognition method and system for magnetic particle testing |
| CN105090676A (en) * | 2015-07-12 | 2015-11-25 | 西南科技大学 | Pressure bearing pipe damage repair process |
| CN107145896A (en) * | 2017-03-14 | 2017-09-08 | 西南科技大学 | Dysnusia identifying system based on fluorescentmagnetic particle(powder) |
Non-Patent Citations (6)
| Title |
|---|
| A.A.CARVALHO 等: "MFL signals and artificial neural networks applied to detection and classification of pipe weld defects", 《NDT & E INTERNATIONAL》 * |
| YUAN TIAN 等: "Automatic defect detection in X-ray images using image data fusion", 《TSINGHUA SCIENCE AND TECHNOLOGY》 * |
| 林果: "基于荧光磁粉的智能无损检测技术研究及实现", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 申清明 等: "焊缝缺陷类型识别方法的研究", 《西安交通大学学报》 * |
| 陈亮: "基于形态学的磁粉检测焊缝裂纹缺陷识别技术研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
| 陈林宇: "基于磁痕图像的无损探伤关键技术研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110717872A (en) * | 2019-10-08 | 2020-01-21 | 江西洪都航空工业集团有限责任公司 | Method and system for extracting characteristic points of V-shaped welding seam image under laser-assisted positioning |
| CN110717872B (en) * | 2019-10-08 | 2022-11-01 | 江西洪都航空工业集团有限责任公司 | Method and system for extracting characteristic points of V-shaped welding seam image under laser-assisted positioning |
| CN110806736A (en) * | 2019-11-19 | 2020-02-18 | 北京工业大学 | Method for detecting quality information of forge pieces of die forging forming intelligent manufacturing production line |
| CN110806736B (en) * | 2019-11-19 | 2021-10-15 | 北京工业大学 | Method for detecting quality information of forge pieces of die forging forming intelligent manufacturing production line |
| CN112816545A (en) * | 2020-09-30 | 2021-05-18 | 中国石油天然气股份有限公司 | Method and device for determining area of storage tank repairing plate |
| CN112435245A (en) * | 2020-11-27 | 2021-03-02 | 济宁鲁科检测器材有限公司 | Magnetic mark defect automatic identification method based on Internet of things |
| CN113390956A (en) * | 2021-06-18 | 2021-09-14 | 西安建筑科技大学 | Double-magnetic-sensor probe and magnetic flux leakage detection defect quantitative evaluation method based on probe |
| CN113390956B (en) * | 2021-06-18 | 2024-02-20 | 西安建筑科技大学 | Double-magnetic-sensor probe and magnetic leakage detection defect quantitative evaluation method based on same |
| CN113740416A (en) * | 2021-08-27 | 2021-12-03 | 东风商用车有限公司 | Engine crankshaft inspection method |
| CN113740416B (en) * | 2021-08-27 | 2023-05-30 | 东风商用车有限公司 | Engine crankshaft inspection method |
| CN114166805A (en) * | 2021-11-03 | 2022-03-11 | 格力电器（合肥）有限公司 | NTC temperature sensor detection method and device, NTC temperature sensor and manufacturing method |
| CN114166805B (en) * | 2021-11-03 | 2024-01-30 | 格力电器（合肥）有限公司 | NTC temperature sensor detection method and device, NTC temperature sensor and manufacturing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109991306A (en) | The Classification and Identification and positioning of metal works welding defect based on fluorescentmagnetic particle(powder) | |
| Ma et al. | A vision-based method for lap weld defects monitoring of galvanized steel sheets using convolutional neural network | |
| Yang et al. | An automatic welding defect location algorithm based on deep learning | |
| Senthil Kumar et al. | Vision inspection system for the identification and classification of defects in MIG welding joints | |
| Wang et al. | Automatic identification of different types of welding defects in radiographic images | |
| Zapata et al. | An adaptive-network-based fuzzy inference system for classification of welding defects | |
| CN113362326A (en) | Method and device for detecting welding spot defects of battery | |
| CN109886298A (en) | A kind of detection method for quality of welding line based on convolutional neural networks | |
| CN105259252A (en) | Method for automatically identifying defect type of polyethylene electrofusion joint through ultrasonic phased array inspection | |
| Liang et al. | Defect detection of rail surface with deep convolutional neural networks | |
| CN107782733A (en) | Image recognition the cannot-harm-detection device and method of cracks of metal surface | |
| Nizam et al. | Vision based identification and classification of weld defects in welding environments: A review | |
| CN105205821A (en) | Weld image segmentation method | |
| Lee et al. | Review on the recent welding research with application of CNN-based deep learning part I: Models and applications | |
| CN111738369A (en) | Weld penetration state and penetration depth real-time prediction method based on visual characteristics of molten pool | |
| Yusof et al. | Automated asphalt pavement crack detection and classification using deep convolution neural network | |
| Cai et al. | Real-time identification of molten pool and keyhole using a deep learning-based semantic segmentation approach in penetration status monitoring | |
| Nazarov et al. | Classification of defects in welds using a convolution neural network | |
| Kumar et al. | Quality level assessment for imperfections in GMAW | |
| Jin et al. | Prediction model for back-bead monitoring during gas metal arc welding using supervised deep learning | |
| CN113592813B (en) | New energy battery welding defect detection method based on deep learning semantic segmentation | |
| Hong et al. | Filter-PCA-based process monitoring and defect identification during climbing helium arc welding process using DE-SVM | |
| Gao et al. | Penetration state recognition based on stereo vision in GMAW process by deep learning | |
| Hong et al. | Real-time quality monitoring of ultrathin sheets edge welding based on microvision sensing and SOCIFS-SVM | |
| Li et al. | A fine-grained flexible graph convolution network for visual inspection of resistance spot welds using cross-domain features |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190709 |