CN112465851B - Parameter detection method based on surface profile curve of weld joint on surface of pressure vessel - Google Patents
Parameter detection method based on surface profile curve of weld joint on surface of pressure vessel Download PDFInfo
- Publication number
- CN112465851B CN112465851B CN202011031080.9A CN202011031080A CN112465851B CN 112465851 B CN112465851 B CN 112465851B CN 202011031080 A CN202011031080 A CN 202011031080A CN 112465851 B CN112465851 B CN 112465851B
- Authority
- CN
- China
- Prior art keywords
- parameter
- weld
- feature
- image
- welding seam
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title abstract description 21
- 238000003466 welding Methods 0.000 claims abstract description 64
- 238000013135 deep learning Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 238000000605 extraction Methods 0.000 claims description 24
- 238000000034 method Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 15
- 238000012937 correction Methods 0.000 claims description 10
- 230000007547 defect Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 239000010410 layer Substances 0.000 claims description 5
- 238000010586 diagram Methods 0.000 claims description 4
- 239000002356 single layer Substances 0.000 claims description 4
- 238000005493 welding type Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 claims description 3
- 239000010953 base metal Substances 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 239000000463 material Substances 0.000 claims description 2
- 239000002184 metal Substances 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 210000001503 joint Anatomy 0.000 claims 4
- 238000011176 pooling Methods 0.000 claims 1
- 238000007689 inspection Methods 0.000 abstract description 4
- 230000000737 periodic effect Effects 0.000 abstract 1
- 230000000007 visual effect Effects 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000001066 destructive effect Effects 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000011089 mechanical engineering Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/181—Segmentation; Edge detection involving edge growing; involving edge linking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30152—Solder
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E30/00—Energy generation of nuclear origin
- Y02E30/30—Nuclear fission reactors
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The utility model discloses a parameter detection method based on a pressure vessel surface weld surface profile curve, which comprises the following steps: generating a welding seam surface profile curve image by using a welding seam surface profile point set of the pressure vessel; constructing a deep learning network structure taking a welding seam surface profile curve image as input and taking network output as coordinate position information; training the deep learning network by taking a welding seam surface profile curve image of a manually marked image parameter characteristic point as a training set; inputting the weld profile curve image to be detected into a trained deep learning network structure, and outputting parameter feature point coordinate positions by the network; and calculating parameter characteristic point coordinate information as a calculation index according to weld parameter definition, and calculating parameter values. The utility model realizes the automatic detection of the welding line surface parameters of the pressure vessel based on machine vision, and plays an important fundamental role in the periodic inspection of the pressure vessel and other works.
Description
Technical Field
The utility model relates to the technical field of pattern recognition, in particular to a parameter detection method based on a pressure vessel surface weld surface profile curve.
Background
The size of the welding seam of the pressure container can reflect the stress concentration degree, and the surface parameter detection is an important guarantee for the safe operation of the pressure container.
Common pressure vessel weld surface parameter detection methods are ruler-based weld surface parameter detection methods and vision-based weld surface parameter detection methods. A welding seam surface parameter detection method based on a ruler is as in patent CN208860222U, CN201921452040.4, a ruler with a specific shape is designed to be in contact with the surface of a pressure container, welding seam parameter information is read through the scale of the ruler, the method is simple and direct to operate, but the contact measurement method is complex in flow, destructive to the welding seam surface and low in measurement efficiency. The visual weld surface parameter detection method is as in patent CN109239081A, and the weld parameters are calculated in the form of image characteristic points by acquiring weld surface images, performing image operation such as image threshold segmentation and noise filtering treatment, and the like.
The specific patent reference and the related documents are as follows:
1) "a weld inspection ruler", patent number CN208860222U. The utility model provides a weld joint inspection ruler, which belongs to the technical field of inspection equipment and comprises a main ruler, wherein one surface of the main ruler is connected with a height ruler and an undercut depth ruler in a sliding manner. The utility model can measure the surplus height, width and undercut depth of the weld joint surfaces with different sizes within a certain specification range. But the utility model cannot measure the parameter misalignment amount.
2) "digital display weld joint detection ruler", patent number CN201921452040.4. The utility model provides a digital display welding seam detection ruler, belongs to the field of measuring instruments, and provides a digital display welding seam detection ruler capable of measuring the width, the height, the recess and the undercut depth of a welding seam. The device avoids errors of manual reading, can reduce the number of detection tools, but the measurement principle is consistent with that of the welding seam ruler, and the pressure vessel to be detected can be scratched.
3) "method for detecting weld quality parameters based on structured light and visual imaging", patent number CN109239081a. The utility model relates to a welding seam quality parameter detection method based on structured light and visual imaging, which comprises the following steps: under the condition that the laser is turned off, the industrial camera shoots welding seam images, under the condition that the laser is turned on, the industrial camera shoots welding seam images containing laser lines, the two welding seam images are preprocessed respectively, and finally, a curve extremum method is adopted to determine parameter characteristic points, so that the applicability of the curve extremum method is low.
4) Tian Yingzhong published in 2019 by university of Shanghai electro-mechanical engineering and Automation of the university, paper "visual sensor-based weld width detection method", which captures weld images in a designated area by means of an industrial camera. According to different gray values of the welding seam and the workpiece, setting the ROI, sequentially carrying out algorithms such as image graying, median filtering, self-adaptive threshold segmentation and the like to binarize the welding seam image, and then obtaining the width of the welding seam through a sequence searching method. The method is only suitable for measuring the width of the parameter, and cannot measure other parameters of the welding line.
Disclosure of Invention
In order to solve the technical problems, the utility model aims to provide a parameter detection method based on a surface profile curve of a welding line on the surface of a pressure container.
The aim of the utility model is achieved by the following technical scheme:
a parameter detection method based on a pressure vessel surface weld surface profile curve comprises the following steps:
a, generating a welding seam surface profile curve image by using a welding seam surface profile point set of a pressure container;
b, constructing a deep learning network structure taking a welding seam surface profile curve image as input and taking network output as coordinate position information;
c, training the deep learning network by taking a welding seam surface profile curve image of the manually marked image parameter feature points as a training set;
inputting the weld profile curve image to be detected into a trained deep learning network structure, and outputting parameter feature point coordinate positions by the network;
and E, calculating parameter characteristic point coordinate information as a calculation index according to weld parameter definition, and calculating parameter values.
One or more embodiments of the present utility model may have the following advantages over the prior art:
compared with the traditional measuring scheme based on an industrial camera, the method adopts a high-precision laser profile sensor, utilizes the high performance of deep learning, can be applied to various welding seams with obvious surface differences, and improves the applicability of the measuring device.
Drawings
FIG. 1 is a flow chart of a method for parameter detection based on a pressure vessel surface weld surface profile curve.
Detailed Description
In order to make the objects, technical solutions and advantages of the present utility model more apparent, the present utility model will be described in further detail with reference to the following examples and the accompanying drawings.
As shown in fig. 1, the flow of the parameter detection method based on the surface profile curve of the welding seam on the surface of the pressure vessel comprises a welding seam image acquisition stage; a deep network design stage; training a training set network; a pressure vessel weld actual measurement stage; a parameter calculation stage; the method specifically comprises the following steps:
step 10, generating a welding seam surface profile curve image by using a welding seam surface profile point set of a pressure container;
step 20, constructing a deep learning network architecture taking a weld surface profile curve image as input and taking network output as coordinate position information;
step 30, training a deep learning network by taking a welding seam surface profile curve image of a manually marked image parameter characteristic point as a training set;
step 40, inputting the weld profile curve image to be detected into a trained deep learning network structure, and outputting parameter feature point coordinate positions by the network;
step 50, calculating parameter feature point coordinate information as a calculation index according to the weld parameter definition, and calculating parameter values.
The step 10 specifically includes:
the surface profile point set of the welding line surface of the pressure vessel surface is P= { P 1 (x 1 ,y 1 ),…,P n (x n ,y n ) Construction image I resolution is set to a (x n -x 1 )，a(y n -y 1 ) Wherein a has a value in the range of 1.1 to 1.5 and the image satisfies I (I, j, 1) =i (I, j, 2) =i (I, j, 3) =255, wherein (I, j) ∈ { (x) 1 ,y 1 ),…,(x n ,y n )}，j＝y 1 ,…,y n The pixel value at the other pixel locations is 0.
The step 20 specifically includes:
the network input is the preprocessed weld image, and the network output is the pixel position of the parameter characteristic point. Firstly, preprocessing a weld image, outputting parameter undercut classification information by a network feature classification module, and determining the number c of laser image feature points; secondly, processing and outputting a c-dimensional feature map by a deconvolution layer of a branch of the feature extraction moduleThe feature point positions are coarsely extracted:
processing and outputting the 2c feature map in the second branch deconvolution layer of the feature processing moduleRough extraction of feature point correction information:
finally, outputting the characteristic diagram by two branchesCombining the information to obtain the positions of the parameter characteristic points in the input image, namely:
the step 30 specifically includes:
the front end CNN of the feature extraction module is in a ResNet structure, the ResNet output multidimensional feature image enters two branch networks respectively, wherein a branch I in the feature extraction module is used as a feature point position coarse extraction task, the ResNet convolution network output deep feature image is deconvoluted by a convolution kernel with single-layer step length of 2, scale of 3x3 and dimension of feature point number c, and the feature image output dimension is reduced to c, namelyWherein each dimension characteristic map corresponds to rough position information of each characteristic point of the input weld joint center line image.
In theoretical regression object A i Design ofIf the single point position of the characteristic point in the image is directly used, A i The positions except the middle characteristic points are negative samples, and the positive and negative samples are seriously unbalanced, so that a characteristic point distance threshold T is introduced, the pixel distance between the positive and negative samples is within T, the positive and negative samples are positive samples, the problem of unbalance of the positive and negative samples is effectively solved, and a theoretical output characteristic diagram A can be obtained i The method comprises the following steps:
the range of the feature point distance threshold T and the background can be regarded as two classification tasks, so that Focalloss of the feature classification task can be used as a loss function, wherein the adjustment parameters alpha and gamma are consistent with the Focalloss of the feature classification task, and then the loss function L is branched M The method comprises the following steps:
feature extraction module branch two is a feature point position correction task, a ResNet convolution network output deep feature map is subjected to single-layer step length of 2, scale of 3 multiplied by 3, dimension of 2c convolution kernel deconvolution, and output of 2c dimension feature map B j Scale and branch one output a i Concordance, at A i Based on the approximate position of the feature points, a feature map B is output on the two-branch theory j And B is connected with j+1 Adding a characteristic point position correction value A i At the true value position, B j And B is connected with j+1 The corresponding element position value is the pixel difference from the theoretical feature point to X, Y axis of the image coordinate system at the element position, so that the two-branch theoretical output B j And B is connected with j+1 The definition is as follows:
wherein ω is the scale of the input image scale and the scale of the feature extraction ResNet network output feature map. The feature point correction task is a numerical regression task, so that a loss function is established by using Huberloss:
L N ＝∑K
wherein delta isThe feature extraction module loss function is obtained by combining the feature extraction branch feature point rough extraction loss function Focalloss and the feature point position correction task loss function Huberloss, wherein the feature extraction module loss function is L N +L M 。
The welding seam parameters of the pressure container are the parameter residual height h of the longitudinal welding seam and the circumferential welding seam re Parameter width l w Parameter undercut depth h uc Parameter misalignment amount h mis With fillet concavity or convexity h c Size h of welding leg fs Wherein the parameter remains high h re Width l w As basic parameters, any longitudinal and circumferential weld joint exists, and the parameter undercut depth h uc Parameter misalignment amount h mis As a defect parameter, a defect-free weld is free of the parameter. According to different types and characteristics of longitudinal weld joints, circumferential weld joints and fillet weld joints of the pressure vessel and different defect parameter undercut depth h uc Parameter misalignment amount h mis There is a case where the type m of the pressure vessel to be inspected is determined.
Selecting parameter residual height h by image welding seam parameter characteristic points re Characteristic point P re Selecting the highest point of the convex part of the welding line curve; width of weld on both sides w Feature pointsSelecting a junction between a parent metal part and a welding line part on a welding line curve; parameter undercut h uc The feature points are selected as the lowest point positions of the concave nearby the width feature points; two-side parameter misalignment of weldQuantity h mis Characteristic points->When no undercut exists, the undercut is overlapped with the width characteristic points on the two sides, and when an undercut defect exists, the undercut concave curve and the base material curve are selected as junction points; fillet weld concavity or convexity h c Characteristic point P c Selecting the position of the convex or concave extreme point of the weld curve; fillet weld two-sided leg size h fs Characteristic points->The junction between the base metal part and the weld part on the fillet weld curve is selected.
The training set manufacturing flow is as follows: firstly, collecting a set of m types of welding seam surface points of a pressure container, collecting different q welding seam surfaces of each type of welding seam, and determining 1 Xm output S of a deep learning feature extraction module according to a to-be-detected pressure container welding seam object 1 ＝[1,0,…,0] 1×m 、S 2 ＝[0,1,…,0] 1×m 、…、S m ＝[0,1,…,0] 1×m Manually marking the image parameter feature points to obtain feature point image coordinates P re (x re ,y re )、The training set comprises classified output S of each pressure container weld object and coordinate information of parameter feature points.
The step 50 specifically includes:
longitudinal weld, girth weld parameter residual height h re Calculating the index as the characteristic points of the width of two sidesAnd the residual height characteristic point P re The maximum value of the distance in the Y-axis direction of the image coordinate system is:
wherein alpha is between pixels in Y-axis directionProportional parameter of distance to actual distance, longitudinal weld and girth weld parameter width l w Calculating the index as the characteristic points of the width of two sidesIn the X-axis direction spacing of the image coordinate system, namely:
wherein beta is the proportional parameter of the pixel spacing in the X-axis direction and the actual distance, and the longitudinal weld joint and circumferential weld joint parameters undercut h uc The calculation index is the undercut feature point P uc And adjacent width feature point P w Spacing in the Y-axis direction of an image coordinate system, namely:
h uc ＝α|y w -y uc |；
longitudinal weld and girth weld parameter misalignment amount h mis Calculating the index as the characteristic points of the width of two sidesSpacing in the Y-axis direction of an image coordinate system, namely:
fillet weld concavity or convexity h c Calculating the index as characteristic points of welding legs on two sidesConnecting line direction and concavity or convexity characteristic point P c Normal spacing, namely:
fillet weld two-sided leg size h fs Calculating the index as characteristic points of welding legs on two sidesThe spacing in the wire direction is as follows:
although the embodiments of the present utility model are described above, the embodiments are only used for facilitating understanding of the present utility model, and are not intended to limit the present utility model. Any person skilled in the art can make any modification and variation in form and detail without departing from the spirit and scope of the present disclosure, but the scope of the present disclosure is still subject to the scope of the appended claims.
Claims (6)
1. A method for detecting parameters based on a surface profile curve of a weld joint on a surface of a pressure vessel, the method comprising the steps of:
step A, generating a welding seam surface profile curve image by using a welding seam surface profile point set of a pressure container;
step B, constructing a deep learning network structure taking a weld surface profile curve image as input and taking network output as coordinate position information;
step C, training the deep learning network by taking a welding seam surface profile curve image of the manually marked image parameter feature points as a training set;
step D, inputting the weld profile curve image to be detected into a trained deep learning network structure, and outputting parameter feature point coordinate positions by a network;
step E, calculating parameter characteristic point coordinate information as a calculation index according to weld parameter definition, and calculating parameter values;
the step C specifically comprises the following steps:
the front end CNN of the feature extraction module is in a ResNet structure, the ResNet output multidimensional feature map respectively enters two branch networks, wherein a branch I in the feature extraction module is a feature point position coarse extraction task, the ResNet convolution network output deep feature map is subjected to single-layer step length of 2, and the scale of 3After deconvolution of the convolution kernel with the dimension being the number c of the feature points, the output dimension of the feature map is reduced to c, namely A i Wherein each dimension feature map corresponds to rough position information of each feature point of the input weld centerline image;
in theoretical regression object A i In design, if the single point position of the characteristic point in the image is directly used, A i The positions except the middle characteristic points are negative samples, and the positive and negative samples are seriously unbalanced, so that a characteristic point distance threshold T is introduced, the pixel distance between the positive and negative samples is within T, the positive and negative samples are positive samples, the problem of unbalance of the positive and negative samples is effectively solved, and a theoretical output characteristic diagram A can be obtained i The method comprises the following steps:
the range of the feature point distance threshold T and the background can be regarded as two classification tasks, so that Focalloss of the feature classification task can be used as a loss function, wherein the adjustment parameters alpha and gamma are consistent with the Focalloss of the feature classification task, and then the loss function L is branched M The method comprises the following steps:
feature extraction module branch two is a feature point position correction task, a ResNet convolution network output deep feature map is subjected to single-layer step length of 2, scale of 3 multiplied by 3, dimension of 2c convolution kernel deconvolution, and output of 2c dimension feature map B j Scale and branch one output a i Concordance, at A i Based on the approximate position of the feature points, a feature map B is output on the two-branch theory j And B is connected with j+1 Adding a characteristic point position correction value A i At the true value position, B j And B is connected with j+1 The corresponding element position value is the pixel difference from the theoretical feature point to X, Y axis of the image coordinate system at the element position, so that the two-branch theoretical output B j And B is connected with j+1 The definition is as follows:
wherein ω is the scale ratio of the input image scale and the feature extraction ResNet network output feature map; the feature point correction task is a numerical regression task, so that a loss function is established by using Huberloss:
wherein delta isThe feature extraction module loss function is obtained by combining the feature extraction branch feature point rough extraction loss function Focalloss and the feature point position correction task loss function Huberloss, wherein the feature extraction module loss function is L N +L M ；
And E, calculating indexes of the middle joint parameters are as follows:
longitudinal weld, girth weld parameter residual height h re Calculating the index as the characteristic points of the width of two sidesAnd the residual height characteristic point P re The maximum value of the distance in the Y-axis direction of the image coordinate system is:
wherein alpha is the proportional parameter of the pixel spacing in the Y-axis direction and the actual distance, and the parameter width l of the longitudinal welding seam and the circumferential welding seam w Calculating the index as the characteristic points of the width of two sidesIn the X-axis direction spacing of the image coordinate system, namely:
wherein beta is the proportional parameter of the pixel spacing in the X-axis direction and the actual distance, and the longitudinal weld joint and circumferential weld joint parameters undercut h uc The calculation index is the undercut feature point P uc And adjacent width feature point P w Spacing in the Y-axis direction of an image coordinate system, namely:
h uc ＝α|y w -y uc |；
longitudinal weld and girth weld parameter misalignment amount h mis Calculating the index as the characteristic points of the width of two sidesSpacing in the Y-axis direction of an image coordinate system, namely:
fillet weld concavity or convexity h c Calculating the index as characteristic points of welding legs on two sidesConnecting line direction and concavity or convexity characteristic point P c Normal spacing, namely:
fillet weld two-sided leg size h fs Calculating the index as characteristic points of welding legs on two sidesThe spacing in the wire direction is as follows:
2. the method for detecting parameters based on the surface profile curve of the weld surface of the pressure vessel according to claim 1, wherein the image flow of the surface profile curve of the weld surface generated in the step a is as follows:
the surface profile point set of the welding line surface of the pressure vessel surface is P= { P 1 (x 1 ,y 1 ),…,P n (x n ,y n ) Construction image I resolution is set to a (x n -x 1 )，a(y n -y 1 ) Wherein a has a value in the range of 1.1 to 1.5 and the image satisfies I (I, j, 1) =i (I, j, 2) =i (I, j, 3) =255, wherein (I, j) ∈ { (x) 1 ,y 1 ),…,(x n ,y n )}，j＝y 1 ,…,y n The pixel value at the other pixel locations is 0.
3. The method for detecting parameters based on the surface profile curve of the weld seam on the surface of the pressure vessel according to claim 1, wherein the calculation flow of the deep learning network structure in the step B is as follows:
the network input is a preprocessed weld image, and the network output is a parameter characteristic point pixel position; firstly, preprocessing a weld image, outputting parameter undercut classification information by a network feature classification module, and determining the number c of laser image feature points; secondly, processing and outputting a c-dimensional feature map by a deconvolution layer of a branch of the feature extraction moduleThe feature point positions are coarsely extracted:
processing and outputting the 2c feature map in the second branch deconvolution layer of the feature processing moduleRough extraction of feature point correction information:
finally, outputting the characteristic diagram by two branchesCombining the information to obtain the positions of the parameter characteristic points in the input image, namely:
4. the method for detecting parameters based on the surface profile curve of the weld joint on the surface of the pressure vessel according to claim 3, wherein the feature classification module structure is as follows: the front end of the feature classification module is in a ResNet structure, a ResNet output multidimensional feature map is firstly reduced to a common one-dimensional feature map of a classification task through a 1×1 convolution kernel with a step length of 1, and then the feature classification module output scale is converted into a 1×m classification result by a global mean value pooling layerWherein m is the required classification category, designing multiple classification task branch output->The loss function Focalloss with the theoretical output S is:
5. the method for detecting parameters based on the surface profile of a weld on a surface of a pressure vessel according to claim 1, wherein in the step C:
the welding seam parameters of the pressure container are the parameter residual height h of the longitudinal welding seam and the circumferential welding seam re Parameter width l w Parameter undercut depth h uc Parameter misalignment amount h mis With fillet concavity or convexity h c Size h of welding leg fs Wherein the parameter remains high h re Width l w As basic parameters, any longitudinal and circumferential weld joint exists, and the parameter undercut depth h uc Parameter misalignment amount h mis As a defect parameter, a defect-free weld is free of the parameter; according to different characteristics of longitudinal weld joints, circumferential weld joints, fillet weld joints and weld joint types of the pressure vessel and different undercut depths h of defect parameters uc Parameter misalignment amount h mis Determining the type m of the detected pressure container in the presence;
selecting parameter residual height h by image welding seam parameter characteristic points re Characteristic point P re Selecting the highest point of the convex part of the welding line curve; width of weld on both sides w Feature pointsSelecting a junction between a parent metal part and a welding line part on a welding line curve; parameter undercut h uc The feature points are selected as the lowest point positions of the concave nearby the width feature points; two-side parameter misalignment h of welding seam mis Feature pointsWhen no undercut exists, the undercut is overlapped with the width characteristic points on the two sides, and when an undercut defect exists, the undercut concave curve and the base material curve are selected as junction points; fillet weld concavity or convexity h c Characteristic point P c Selecting the position of the convex or concave extreme point of the weld curve; fillet weld two-sided leg size h fs Characteristic points->The junction between the base metal part and the weld part on the fillet weld curve is selected.
6. The method for detecting parameters based on the surface profile of a weld on a surface of a pressure vessel according to claim 1, wherein in the step C:
the training set manufacturing flow is as follows: collecting a set of m types of welding seam surface points of the pressure vessel, collecting different q welding seam surfaces of each type of welding seam, and determining 1 Xm output S of the deep learning feature extraction module according to the welding seam object of the pressure vessel to be detected 1 ＝[1,0,…,0] 1×m 、S 2 ＝[0,1,…,0] 1×m 、…、S m ＝[0,1,…,0] 1×m Manually marking the image parameter feature points to obtain feature point image coordinatesThe training set comprises classified output S of each pressure container weld object and coordinate information of parameter feature points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011031080.9A CN112465851B (en) | 2020-09-27 | 2020-09-27 | Parameter detection method based on surface profile curve of weld joint on surface of pressure vessel |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011031080.9A CN112465851B (en) | 2020-09-27 | 2020-09-27 | Parameter detection method based on surface profile curve of weld joint on surface of pressure vessel |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112465851A CN112465851A (en) | 2021-03-09 |
| CN112465851B true CN112465851B (en) | 2023-08-01 |
Family
ID=74832857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011031080.9A Active CN112465851B (en) | 2020-09-27 | 2020-09-27 | Parameter detection method based on surface profile curve of weld joint on surface of pressure vessel |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112465851B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113393518A (en) * | 2021-07-15 | 2021-09-14 | 北京京仪光电技术研究所有限公司 | Automatic extraction method for relative numerical coordinates of picture curve |
| CN113409313B (en) * | 2021-08-18 | 2021-11-09 | 济宁联威车轮制造有限公司 | Wheel weld surface defect detection method based on computer vision |
| CN115338556A (en) * | 2022-08-03 | 2022-11-15 | 湖南科技大学 | Weld joint quality detection method for thick-wall welding workpiece and computer equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109239081A (en) * | 2018-09-18 | 2019-01-18 | 广东省特种设备检测研究院珠海检测院 | Weldquality parameter detection method based on structure light and visual imaging |
| CN109900706A (en) * | 2019-03-20 | 2019-06-18 | 易思维（杭州）科技有限公司 | A kind of weld seam and weld defect detection method based on deep learning |
| CN109977948A (en) * | 2019-03-20 | 2019-07-05 | 哈尔滨工业大学 | A kind of stirring friction welding seam defect identification method based on convolutional neural networks |
| CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
| CN110675370A (en) * | 2019-09-04 | 2020-01-10 | 武汉理工大学 | Welding simulator virtual weld defect detection method based on deep learning |
-
2020
- 2020-09-27 CN CN202011031080.9A patent/CN112465851B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109239081A (en) * | 2018-09-18 | 2019-01-18 | 广东省特种设备检测研究院珠海检测院 | Weldquality parameter detection method based on structure light and visual imaging |
| CN109900706A (en) * | 2019-03-20 | 2019-06-18 | 易思维（杭州）科技有限公司 | A kind of weld seam and weld defect detection method based on deep learning |
| CN109977948A (en) * | 2019-03-20 | 2019-07-05 | 哈尔滨工业大学 | A kind of stirring friction welding seam defect identification method based on convolutional neural networks |
| CN110675370A (en) * | 2019-09-04 | 2020-01-10 | 武汉理工大学 | Welding simulator virtual weld defect detection method based on deep learning |
| CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
Non-Patent Citations (1)
| Title |
|---|
| 基于改进深度学习模型的焊缝缺陷检测算法;谷静;谢泽群;张心雨;;宇航计测技术(第03期);第1-4页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112465851A (en) | 2021-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112465851B (en) | Parameter detection method based on surface profile curve of weld joint on surface of pressure vessel | |
| CN109087274B (en) | Electronic device defect detection method and device based on multi-dimensional fusion and semantic segmentation | |
| WO2019134252A1 (en) | Method and device for automated portrayal and accurate measurement of width of structural crack | |
| CN107341802B (en) | Corner sub-pixel positioning method based on curvature and gray scale compounding | |
| Abouelatta | 3D surface roughness measurement using a light sectioning vision system | |
| US8428335B2 (en) | Combining feature boundaries | |
| CN105913415A (en) | Image sub-pixel edge extraction method having extensive adaptability | |
| CN110766669B (en) | Pipeline measuring method based on multi-view vision | |
| KR940003829B1 (en) | Position recognizing method | |
| CN110751604A (en) | Machine vision-based steel pipe weld defect online detection method | |
| CN112200790B (en) | Cloth defect detection method, device and medium | |
| CN115937203A (en) | Visual detection method, device, equipment and medium based on template matching | |
| Wankhede et al. | A low cost surface strain measurement system using image processing for sheet metal forming applications | |
| CN113970560A (en) | Defect three-dimensional detection method based on multi-sensor fusion | |
| CN113947583A (en) | Weld joint nondestructive testing method based on deep learning two-dimensional time sequence image | |
| Gadelmawla | Development of Computer Vision Algorithms for Measurement and Inspection of Spur Gears. | |
| CN111815580A (en) | Image edge identification method and small module gear module detection method | |
| Wang et al. | A binocular vision method for precise hole recognition in satellite assembly systems | |
| Ziqiang et al. | Research of the algorithm calculating the length of bridge crack based on stereo vision | |
| CN114862892A (en) | Two-dimensional and three-dimensional pavement crack identification method based on enhanced depth edge characteristics | |
| Zou et al. | Laser-based precise measurement of tailor welded blanks: a case study | |
| Mikołajczyk et al. | Camera-Based Automatic System for Tool Measurements and Recognition | |
| CN114092396A (en) | Method and device for detecting corner collision flaw of packaging box | |
| CN112414316A (en) | Strain gauge sensitive grid size parameter measuring method | |
| Kang et al. | Automatic circle pattern extraction and camera calibration using fast adaptive binarization and plane homography |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20210309 Assignee: Guangdong Mingyang Electric Co.,Ltd. Assignor: SOUTH CHINA University OF TECHNOLOGY Contract record no.: X2023980047999 Denomination of invention: A parameter detection method based on the surface contour curve of pressure vessel surface welds Granted publication date: 20230801 License type: Common License Record date: 20231123 |