CN111724369A - U-shaped welding seam detection method - Google Patents
U-shaped welding seam detection method Download PDFInfo
- Publication number
- CN111724369A CN111724369A CN202010560213.5A CN202010560213A CN111724369A CN 111724369 A CN111724369 A CN 111724369A CN 202010560213 A CN202010560213 A CN 202010560213A CN 111724369 A CN111724369 A CN 111724369A
- Authority
- CN
- China
- Prior art keywords
- point
- welding seam
- value
- calculating
- data
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention discloses a U-shaped welding seam detection method, which takes laser sensor data at a certain moment on a robot track route as prior information to detect a welding seam, and comprises the following steps: denoising the data, namely denoising the sensor data; determining a reference line, so as to facilitate subsequent feature extraction; detecting an angular point, finding out a weld inflection point and calculating the width of a weld; calculating the approximate area, and solving the area of the welding seam by using a differential method; calculating the height of the welding line by taking the datum line as a reference; performing inclination identification to obtain a middle axis fitting straight line of the welding line and calculating an angle between the middle axis fitting straight line and a datum line; and (5) detecting the profile, namely detecting the defects of the weld profile by using a deep learning method. The welding seam detection method avoids the defects of manual detection, can reduce the error identification rate of welding seam detection, improves the accuracy and efficiency of welding seam analysis, and enables the welding seam analysis to be more comprehensive and intelligent through a series of algorithm detection.
Description
Technical Field
The invention relates to the technical field of automobile welding, in particular to a method for identifying a butt weld of a metal sheet structure by a robot, and particularly relates to a weld surface detection method.
Background
Welding has been developed as an important processing method in manufacturing industry, and is widely applied to the fields of aviation, aerospace, metallurgy, petroleum, automobile manufacturing, national defense and the like. In a welded product, the quality of a welding seam directly influences the service life of the product. Therefore, the size of the welding seam must be strictly controlled according to the design requirements in the production process, and the generation of various defects must be strictly controlled.
The measurement of the weld surface dimensions and the assessment of surface weld defects are usually carried out by visual inspection by manual inspection. When the size of the welding seam is measured by a visual inspection method, tools such as a magnifying glass, a ruler, an undercut measurer and the like are usually adopted for measurement; when the defects are evaluated, the evaluating personnel is required to have stronger professional knowledge and abundant working experience. Meanwhile, in the visual inspection detection process, workers are easily affected by factors such as large workload, poor working environment and knowledge cognition difference, so that the accuracy of results is reduced. In addition, when the detection is performed manually, the detection personnel easily feel fatigue after long-time detection, and the reliability of the detection result is reduced. Therefore, it is difficult to ensure the normative, objectivity and scientificity of the result by measuring the result with a visual inspection method.
Disclosure of Invention
In order to solve the problems, the invention provides a method for detecting a U-shaped welding seam according to the characteristics of the robot welding industry. And in the moving process of the robot, the laser sensor transmits data in real time. And (4) taking the section formed by all distance data at a certain moment on the welding track route as prior information to detect the welding seam. The technical scheme of the invention is as follows:
and carrying out median filtering on the distance data of the laser sensor. Selecting a data point to be processed, and arranging all data points in a neighborhood window with the length of an odd number in a sequence from small to large by taking the data point as a middle position; and using the sorted median value as the value of the data point to be processed.
Step 2, determining a datum line:
in the cross section of the weld, the reference line is the weldment groove plane. And determining a datum line so as to facilitate subsequent feature extraction and weld analysis.
Visualizing all data points of the laser sensor at a certain moment to obtain a form of a curve in a welding seam section, and setting an equation of a straight line and a undetermined coefficient;
in order that the fitted approximate curve can reflect the change trend of the given data point as much as possible, the sum of the squares of the errors of the actual value and the fitted straight line is selected as a measurement standard, namely the minimum value of the sum of the squares of the errors under the least square method is required;
establishing an equation set according to a least square method and solving the equation set to obtain a coefficient to be determined of a fitted straight line;
and obtaining a fitted linear equation, further calculating an included angle between the datum line and the horizontal line, and if the included angle does not meet the standard, determining that the welding line is unqualified.
Step 3, angular point detection:
the corner points are important characteristic points as two boundary contact points of the welding line and the plane and the inflection points of the welding line. In order to make the corner detection more stable, the algorithm needs to satisfy the conditions of unchanged scale, unchanged rotation, noise influence resistance and the like. Based on the requirements, a concept of applying a gray difference value of adjacent pixel points is provided, and a gray change value is calculated in the image by utilizing a moving window.
Performing binarization operation on the welding seam section to obtain a section image;
constructing a mathematical model, calculating a gray level difference value of a moving window, and simplifying and solving gray level change generated by translation of an image window;
filtering each pixel of the sectional image by using a horizontal difference operator and a vertical difference operator to obtain partial derivatives in the x direction and the y direction so as to obtain a partial derivative matrix;
performing Gaussian smoothing filtering on the partial derivative matrix to eliminate some isolated points and bulges to obtain a new partial derivative matrix;
and defining a corner response function according to the partial derivative matrix, and calculating the corner response function corresponding to each pixel by using the new partial derivative matrix. Judging whether the pixel is an angular point or not according to the angular point response function value;
and carrying out local maximum suppression on the corner response function, and selecting the maximum value of the corner response function. In the corner response function, points satisfying two conditions of being larger than a certain threshold and a local maximum in a certain field at the same time are regarded as corners;
and calculating the distance between the two corner points according to the obtained corner points, wherein the distance is the width of the welding seam. And if the width value does not meet the standard, the welding seam is considered to be unqualified.
Step 4, approximate area calculation:
the area of the weld in the cross section of the weld is the area of the graph enclosed by the curve between the two angular points and the reference line. The area can be approximately calculated by a infinitesimal method, the narrow trapezoid is approximately replaced by the corresponding narrow trapezoid of the curved edge trapezoid, and the sum of the areas of the narrow trapezoids is used as the approximate value of the curved edge trapezoid.
Determining an integral interval according to the calculation results of the reference line and the angular point, dividing the integral interval into a plurality of small intervals, and approximately solving the area of each small interval by using a small trapezoid;
calculating the distance between two adjacent data points as the height of each small trapezoid, and respectively calculating the distance between two adjacent data points and the reference line as the upper bottom and the lower bottom of each small trapezoid;
according to a trapezoid area calculation formula, the area of each small trapezoid is obtained, and the areas of all the small trapezoids are summed, wherein the value is an approximate value of the cross section area of the welding line;
and if the area value does not meet the standard, the welding seam is considered to be unqualified.
Step 5, calculating the height of the welding seam:
selecting curve segments between the angular points according to the calculation results of the angular points and the reference line; calculating the distance between each data point and the datum line; selecting the farthest distance as the height of the welding seam; and if the height value does not meet the standard, the welding seam is considered to be unqualified.
Step 6, inclination recognition:
selecting a data segment between two angular points, and converting each data point into voxel grid data;
judging whether the converted voxel grids meet eight connectivity or not for any two adjacent data points, if not, calculating the slope between the two points, and carrying out grid filling operation according to the slope;
splicing the connected curves and the reference lines to form a closed graph, selecting one point in the closed graph, and carrying out flood filling of the four-way communication;
the following operations are performed a number of times until the entire voxel map is no longer changed: for each row of voxels, scanning from left to right, if the current voxel is filled, judging whether the current voxel is an isolated point, a middle point, a straight line segment point and a boundary point which causes the increase of a connected component after removal through a 3 x 3 matrix taking the voxel as the center. If the current node does not belong to any of the above classifications, the voxel is considered removed.
And constructing error differences at each point in the skeleton obtained in the last step, and combining all error terms to obtain a least square problem. For the least square problem, selecting a corresponding learning rate, and carrying out multiple calculation operations until the change of a solving parameter is less than a threshold value or the iteration times exceeds a set upper limit;
and (4) calculating the angle between the obtained middle axis fitting straight line and the datum line, and if the angle does not meet the requirement, determining that the inclination occurs and the welding line is unqualified.
Step 7, contour detection:
selecting multi-frame data from a plurality of welding seam samples, screening out image data with defects of pores, slag inclusion, curling and the like, and segmenting and labeling the image data to obtain a training set;
and establishing a U-net network, and training by using the obtained training set to obtain a U-net network model. The U-net network structure is as follows: the U-net systolic path, i.e. the conventional convolutional network, performs two 3 × 3 convolutions first, followed by one max pooling operation, and down-samples the entire graph. Repeating the operation for more than four times; and secondly, performing two times of 3 × 3 convolution, performing up-sampling by using an up-convolution operation, and cutting the previous bottom characteristic diagram. Repeating the operation for more than four times; thirdly, performing 3 × 3 convolution twice, and performing 1 × 1 convolution once again; fourthly, calculating by using a softmax function to obtain final output;
inputting the data to be detected into the trained network model for prediction, and outputting a prediction result;
and if the predicted result shows that the defect area exists, dividing the area and judging that the welding seam is unqualified.
The invention provides a complete set of U-shaped welding seam detection method, which focuses on the analysis of data characteristics according to the fact that distance data detected by a sensor are prior information. Noise information interference can be removed through noise processing, characteristics such as angular points, center axes, height and width are further extracted, and finally welding seam contour detection is carried out through a convolutional neural network, so that welding seam surface analysis is completed. In the process, the detection system can identify complex fault types more easily through image processing, machine learning and deep learning, and compared with a traditional manual mode, the detection system has better precision and efficiency. Meanwhile, the section data processing mode in the invention can be suitable for various devices, and can obtain better effect after converting data in different formats into point cloud data, thereby having good expandability.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
Fig. 2 is a cross-sectional view of a weld at a point in time.
Fig. 3 is a sample of a void defect.
FIG. 4 is a slag inclusion defect sample.
Detailed Description
In the moving process of the robot, the laser sensor is fixed on a clamp at the tail end of the robot, and the sensor collects and transmits back data so as to analyze the quality of a welding seam. According to the characteristics of the robot welding industry, the invention provides a U-shaped welding seam detection method, which is described by taking the following section data at a certain moment as an example, as shown in fig. 1, and the example comprises the following steps:
since noise such as isolated points and outliers generated by the laser sensor affects the accuracy of the final ideal model, it is necessary to perform targeted removal or adjustment according to the distribution. The noise generated by the scanning measurement of the laser sensor has diversity: the active non-contact coverage type measurement is interfered by the environment, and useless data and outlier noise of the surface of a non-measurement target object can be generated; sharp features on the surface of an object and the like cause sharp change of laser incidence angle and possibly cause measurement to generate burr noise; the laser ranging approach without a cooperative target necessarily generates tiny random noise.
And carrying out median filtering on the distance data of the laser sensor. When filtering a sequence xj (— infinity < j < ∞), a window with an odd number M is defined, where M is 2N +1 and N is a positive integer. For a certain data point x (i) to be processed, the data samples in the neighboring window are x (i-N), …, x (i), …, x (i + N), where x (i) is the data sample value located in the center of the window. The M sample values are arranged in descending order, where the value, the sample value at i, is defined as the output value of the median filter. Performing the median filtering operation on all data points;
through experiments, the median filtering when M is 65 has better noise removal results on the data.
Step 2, determining a reference line:
as shown in fig. 2, a reference line L is determined for subsequent feature extraction for weld analysis.
Visualizing all distance data points of the laser sensor at a certain moment, analyzing the form of a curve, and determining that the form of a fitted curve is a straight line, wherein the equation is y ═ ax + b;
for the curve fitting function φ (x), it is not required to strictly pass through all data points (x)i，yi) That is to say that the fitting function φ (x) is at xiThe deviations (also called residuals) are not all exactly equal to zero, i.e. they are the sets of contradictory equations:
in order that the fitted approximate curve can reflect the change trend of the given data point as much as possible, the sum of the squares of the errors of the actual value and the fitted straight line is selected as a measurement standard, namely the minimum value of the sum of the squares of the errors under the least square method is required:
establishing a normal equation set according to a least square method and solving the normal equation set to obtain undetermined parameters a and b of a datum line equation:
and after a reference line equation L is obtained, calculating an included angle between the reference line and the horizontal line. The slope of the equation for the baseline L is known as a, which is related to the horizontal line L0The included angle theta is arctan a.
And obtaining a fitted linear equation, further calculating an included angle between the datum line and the horizontal line, and if the included angle does not meet the standard, determining that the welding line is unqualified.
Step 3, angular point detection:
as shown in fig. 2, the corner point B, C is an important feature point as two boundary contact points of the weld and the plane, and as an inflection point of the curve. In order to make the corner detection more stable, the algorithm needs to satisfy the conditions of unchanged scale, unchanged rotation, noise influence resistance and the like. Based on the requirements, a concept of applying a gray difference value of adjacent pixel points is provided, and a gray change value is calculated in the image by utilizing a moving window.
Performing binarization operation on the welding seam section to obtain a section image;
constructing a mathematical model, calculating the gray difference of a moving window, translating the image window [ u, v ] to generate gray change E (u, v) as follows:
E(u,v)＝∑w(x,y)[I(x+u,y+v)-I(x,y)]2
to reduce the amount of computation, a simplified formula is performed using a taylor series:
by I (x + u, y + v) ═ I (x, y) + Ixu+Iyv+O(u2,v2)
To obtain E (u, v) ∑ w (x, y) Ixu+Iyv+O(u2,v2)]
For a local slight shift amount [ u, v ], the following expression can be approximated
E(u,v)≈[u,v]M
Where M is a 2 × 2 matrix, which can be obtained from the reciprocal of the image:
filtering each pixel of the sectional image by using horizontal and vertical difference operators to obtain Ix、IyFurther obtaining a partial derivative matrix M;
performing Gaussian smoothing filtering on four elements of the partial derivative matrix M so as to eliminate some unnecessary isolated points and bulges and obtain a new matrix M;
m is a covariance matrix of the gradient, a corner response function R is defined in practical application in order to apply better programming, and whether a pixel is a corner or not is judged by judging the size of R;
and (4) carrying out local maximum suppression on the corner response function R, and simultaneously selecting a maximum value of the corner response function R. In the corner response function R, points which simultaneously satisfy that R (i, j) is greater than a certain threshold and R (i, j) is a local maximum in a certain neighborhood are considered as corners;
from the resulting corner point B, C, the distance d between two corner points is calculated as:
this distance is the width of the weld. And if the width value does not meet the standard, the welding seam is considered to be unqualified.
Step 4, approximate area calculation:
as shown in fig. 2, the cross-sectional area of the weld is the area of the graph enclosed by the curve between the two corner points B, C and the reference line L. The area can be approximately calculated by a infinitesimal method, the narrow trapezoid is approximately replaced by the corresponding narrow trapezoid of the curved edge trapezoid, and the sum of the areas of the narrow trapezoids is used as the approximate value of the curved edge trapezoid.
Determining an integral interval according to the calculation results of the reference line and the angular point, dividing the integral interval into a plurality of small intervals, and approximately solving the area of each small interval by using a small trapezoid;
the distance between two adjacent data points is calculated as the height of each small trapezoid, i.e. point (x)i,yi) And point (x)i+1,yi+1) The distance between:
calculating two adjacent data points (x) respectivelyi,yi)、(xi+1,yi+1) The distance from the reference line is taken as the upper bottom and the lower bottom of each small trapezoid, and the function value of the function y corresponding to each branch point is set as f (x)0,y1,…，ynFor each integration interval, the upper base d of the trapezoid is small1Is a point (x)i,yi) Distance from reference line L, lower base d2Is a point (x)i+1,yi+1) Distance from the reference line L:
according to the trapezoidal area calculation formula, the area of each small trapezoid is obtained
And summing the areas of all the small trapezoids, wherein the value is an approximate value of the cross section area of the welding seam;
and if the area value does not meet the standard, the welding seam is considered to be unqualified.
Step 5, calculating the height of the welding seam:
as shown in fig. 2, according to the calculation result of the corner point B, C and the reference line, a curve segment between the corner points B, C is selected, and the distance between each data point in the segment and the reference line L (y ═ ax + b) is calculated:
and selecting the farthest distance as the height h of the welding seam, and if the height value does not meet the standard, determining that the welding seam is unqualified.
Step 6, inclination recognition:
selecting a data segment between two corner points, and dividing each data point (x)i，yi) Conversion to voxel grid data (x'i，y′i) Suppose a voxel grid represents the grid of l × l in the data points, then there is (x'i，y′i)＝(xi，yi)÷l；
For any two adjacent point cloud points, judging whether the converted voxel grids meet eight connectivity, if the two point cloud points do not meet eight connectivity, calculating a slope k between the two points, and performing grid filling operation according to the slope:
i slope is absent: filling grids from the lower grid upwards to the upper grid;
II, the absolute value of the slope is less than or equal to 1, an increment value is maintained from the left point, the filling grid is increased by 1 along the x axis every time, then the slope k is increased on the increment value, if the absolute value of the increment value is greater than 0.5, the filling grid is correspondingly moved by one grid on the y axis and filled (if the increment value is positive, the y value is increased by 1, otherwise, the y value is decreased by 1), otherwise, the filling grid is not moved on the y axis and filled, and then the increment value is increased by 1 or decreased by 1, so that the absolute value is less than 0.5;
III, the absolute value of the slope is larger than 1, an increment value is maintained from the lower point, the filling grid is increased by 1 along the y axis in each operation, then the slope k is increased on the increment value, if the absolute value of the increment value is larger than 0.5, the filling grid is correspondingly moved by one grid on the x axis and filled (if the increment value is positive, the x value is increased by 1, otherwise, the x value is decreased by 1), otherwise, the filling grid is not moved on the x axis and filled, and then the increment value is increased by 1 or decreased by 1, so that the absolute value is smaller than 0.5;
splicing the connected curve and the reference line L to form a closed graph, selecting one point in the closed graph, and carrying out flood filling of four-way communication;
the following operations are performed a number of times until the entire voxel map is no longer changed: for each row of voxels, scanning from left to right, if the current voxel is filled, judging whether the current voxel is an isolated point, a middle point, a straight line segment point and a boundary point which causes the increase of a connected component after removal through a 3 x 3 matrix taking the voxel as the center. If the current node does not belong to any of the above classifications, then the voxel is considered to be removed;
every point (x) in the skeleton obtained in the previous stepi，yi) Construction error (kx)i+b-yi)2And all error terms are combined to obtain a least square problem:
calculating the derivatives of k and b, respectively, yields:
for the least square problem, selecting a corresponding learning rate alpha, and carrying out multiple calculation operations until the change of solving parameters k and b is less than a threshold value or the iteration times exceeds a set upper limit;
and (4) calculating the angle between the obtained middle axis fitting straight line and the reference line, and if the angle does not meet the requirement, determining that the inclination occurs and the welding line is unqualified.
Step 7, contour detection:
1) selecting multi-frame data from a plurality of welding seam samples, screening out image data with defects such as pores (shown in figure 3) or slag inclusion (shown in figure 4) and the like, and segmenting and labeling the image data to obtain a training set;
2) building a U-net network, training the U-net network by using the training set obtained in the step 1), and training to obtain a U-net network model. The U-net network structure is as follows:
i U-net, a conventional convolutional network, first performs two 3 x 3 convolutions, followed by a max-pooling operation to downsample the entire graph. Repeating the operation for more than four times;
II, performing 3 × 3 convolution twice on the result in the I, then performing up-sampling by using an up-convolution operation, and cutting the bottom layer characteristic diagram before the up-sampling. Repeating the operation for more than four times;
III, performing two times of 3 × 3 convolution on the result in the II, and performing another 1 × 1 convolution;
IV, calculating the result in the III by using a softmax function to obtain final output;
3) inputting data to be detected into the U-net network model obtained in the step 2) for prediction, and outputting a prediction result;
4) if the prediction result in 3) is that a defect area exists, dividing the area and giving out a warning.
The principles and embodiments of the present invention are explained herein using specific examples, which are presented only to assist in understanding the method and its core concepts. It should be noted that, for those skilled in the art, it is possible to make various improvements and modifications to the present invention without departing from the principle of the present invention, and those improvements and modifications also fall within the scope of the claims of the present invention.
Claims (9)
1. A U-shaped welding seam detection method is characterized by comprising the following steps:
a step of determining a reference line, which is to acquire data of the laser sensor and determine the reference line of the section of the welding seam;
an angular point detection step, namely acquiring an angular point of a curve in a welding seam section according to the gray level difference value of the adjacent pixel points, and calculating the width of the welding seam;
calculating an approximate area, namely obtaining the approximate area of the section of the welding seam by a infinitesimal method;
a step of calculating the height of the welding seam, which is to select curve segments between the angular points, calculate the distance between each data point and a reference line and select the farthest distance as the height of the welding seam;
an inclination identification step, namely converting the data points into voxel grid data, filling and removing the voxel grid data according to the slope to obtain a curve middle axis, and calculating an angle between the middle axis and a datum line to obtain a welding line inclination angle;
and contour detection, namely establishing a U-net network, and training the convolutional neural network by using a training set obtained by segmentation and labeling to obtain a U-net network model.
2. The U-shaped weld detection method according to claim 1, further comprising a data de-noising step comprising:
selecting a data point to be processed;
arranging all data points in a neighborhood window with odd length from small to large by taking the data point as a middle position;
and using the sorted median value as the value of the data point to be processed.
3. A U-shaped weld detection method according to claim 1, wherein the reference line determining step includes:
visualizing all data points of the laser sensor at a certain moment to obtain a form of a curve in a welding seam section, and setting an equation of a straight line and a undetermined coefficient;
establishing an equation set for the curve according to a least square method and solving the equation set to obtain a coefficient to be determined of a fitted straight line;
obtaining a fitted linear equation;
and calculating the included angle between the datum line and the horizontal line.
4. A U-shaped weld detection method according to claim 1, wherein the corner point detection step comprises:
performing binarization operation on the welding seam section to obtain a section image; constructing a mathematical model, calculating a gray level difference value of a moving window, and simplifying and solving gray level change generated by translation of an image window;
filtering each pixel of the sectional image by using a horizontal difference operator and a vertical difference operator to obtain partial derivatives in the x direction and the y direction so as to obtain a partial derivative matrix;
performing Gaussian smoothing filtering on the partial derivative matrix to eliminate some isolated points and bulges to obtain a new partial derivative matrix;
defining an angular point response function according to the partial derivative matrix, calculating an angular point response function corresponding to each pixel by using the new partial derivative matrix, and judging whether the pixel is an angular point or not according to the angular point response function value;
the method comprises the steps of inhibiting local maximum values of an angular point response function, selecting the maximum values of the angular point response function, and regarding points which meet two conditions of being larger than a certain threshold value and local maximum values in a certain field as angular points in the angular point response function;
calculating the distance between the two angular points according to the obtained angular points, wherein the distance is the width of the welding line;
in the corner response function, a point satisfying both conditions of being greater than a certain threshold and a local maximum in a certain field is regarded as a corner.
5. The U-shaped weld detection method according to claim 1, wherein the approximate area calculation step includes:
determining an integral interval according to the calculation results of the reference line and the angular point, dividing the integral interval into a plurality of small intervals, and approximately solving the area of each small interval by using a small trapezoid;
calculating the distance between two adjacent data points as the height of each small trapezoid, and respectively calculating the distance between two adjacent data points and the reference line as the upper bottom and the lower bottom of each small trapezoid;
and (4) solving the area of each small trapezoid according to a trapezoid area calculation formula, and summing the areas of all the small trapezoids, wherein the value is an approximate value of the cross section area of the welding seam.
6. The U-shaped weld detection method according to claim 1, wherein the weld height calculation step includes:
selecting curve segments between the angular points in the welding seam section according to the calculation results of the angular points and the datum lines;
calculating the distance between each data point and the datum line;
and selecting the farthest distance as the height of the welding seam.
7. The U-shaped weld detection method according to claim 1, wherein the inclination recognition step includes:
selecting a data segment between two angular points, and converting each data point into voxel grid data;
judging whether the voxel grid meets eight connectivity or not for any two adjacent data points, if not, calculating the slope between the two points, and carrying out grid filling operation according to the slope;
splicing the connected curves and the reference lines to form a closed graph, selecting one point in the closed graph, and carrying out flood filling of the four-way communication;
the following operations are performed a number of times until the entire voxel map is no longer changed: scanning each row of voxels from left to right, if the current voxel is filled, judging whether the current voxel is an isolated point, a middle point, a straight line segment point and a boundary point which causes the increase of connected components after removal through a 3 x 3 matrix taking the voxel as the center, and if the current node does not belong to any one of the above categories, considering that the voxel is removed;
constructing error difference for each point in the obtained skeleton, simultaneously establishing all error items to obtain a least square problem, selecting corresponding learning rate for the least square problem, and carrying out multiple times of calculation operation until the change of solving parameters is less than a threshold value or the iteration times exceeds a set upper limit;
and (5) fitting the obtained middle axis with a straight line, and calculating the angle between the middle axis and the reference line.
8. A U-shaped weld detection method according to claim 1, wherein the profile detection step comprises:
selecting multi-frame data from a plurality of welding seam samples, screening out images with defects of pores, slag inclusion and the like, and carrying out segmentation calibration on the images to obtain a training set;
establishing a U-net network, and training by using a training set obtained by labeling to obtain a U-net network model;
inputting the data to be detected into the trained network model for prediction, and outputting a prediction result;
and if the predicted result shows that the defect area exists, dividing the area and sending out the prompt information of unqualified welding seams.
9. The weld detection method of claim 7, wherein the performing a grid filling operation based on the slopes of two adjacent data points further comprises:
the i slope is absent: filling grids from the lower grid upwards to the upper grid;
II, the absolute value of the slope is less than or equal to 1, an increment value is maintained from the left point, the filling grid is increased by 1 along the x axis every time, then the slope k is increased on the increment value, if the absolute value of the increment value is greater than 0.5, the filling grid is correspondingly moved by one grid on the y axis and filled, if the increment value is positive, the y value is increased by 1, otherwise, the y value is decreased by 1, otherwise, the filling grid is not moved on the y axis and filled, and then the increment value is increased by 1 or decreased by 1, so that the absolute value is less than 0.5;
III, starting from a lower point, maintaining an increment value, increasing the filling grid by 1 along the y axis in each operation, then increasing the slope k on the increment value, if the absolute value of the increment value is more than 0.5 at the moment, correspondingly moving the filling grid by one grid on the x axis and filling, if the increment value is positive, adding 1 to the x value, otherwise, subtracting 1 from the x value, otherwise, not moving the filling grid on the x axis and filling, and then adding 1 or subtracting 1 from the increment value to enable the absolute value to be less than 0.5.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010551995 | 2020-06-17 | ||
| CN2020105519956 | 2020-06-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111724369A true CN111724369A (en) | 2020-09-29 |
Family
ID=72567496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010560213.5A Pending CN111724369A (en) | 2020-06-17 | 2020-06-18 | U-shaped welding seam detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111724369A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112614105A (en) * | 2020-12-23 | 2021-04-06 | 东华大学 | Depth network-based 3D point cloud welding spot defect detection method |
| CN112950633A (en) * | 2021-04-19 | 2021-06-11 | 上海电机学院 | Aluminum alloy weld surface defect detection method based on line structured light |
| CN113369761A (en) * | 2021-07-09 | 2021-09-10 | 北京石油化工学院 | Method and system for guiding robot welding seam positioning based on vision |
| CN115082436A (en) * | 2022-07-22 | 2022-09-20 | 山东易斯特工程工具有限公司 | Shield tunneling machine tool bit production defect detection method |
| CN116342611A (en) * | 2023-05-31 | 2023-06-27 | 无锡日联科技股份有限公司 | Method and device for detecting bonding wires of micro semiconductor |
| CN116559298A (en) * | 2023-07-12 | 2023-08-08 | 苏州精控能源科技有限公司 | Battery welding quality detection method and device |
| CN117079100A (en) * | 2023-08-14 | 2023-11-17 | 北京大学 | Weld defect recognition system based on deep learning |
| CN117132406A (en) * | 2023-10-27 | 2023-11-28 | 深圳市泰铼科技有限公司 | Financial product assessment method and system based on deep learning |
-
2020
- 2020-06-18 CN CN202010560213.5A patent/CN111724369A/en active Pending
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112614105A (en) * | 2020-12-23 | 2021-04-06 | 东华大学 | Depth network-based 3D point cloud welding spot defect detection method |
| CN112614105B (en) * | 2020-12-23 | 2022-08-23 | 东华大学 | Depth network-based 3D point cloud welding spot defect detection method |
| CN112950633A (en) * | 2021-04-19 | 2021-06-11 | 上海电机学院 | Aluminum alloy weld surface defect detection method based on line structured light |
| CN112950633B (en) * | 2021-04-19 | 2023-06-23 | 上海电机学院 | Aluminum alloy weld joint surface defect detection method based on line structured light |
| CN113369761A (en) * | 2021-07-09 | 2021-09-10 | 北京石油化工学院 | Method and system for guiding robot welding seam positioning based on vision |
| CN115082436A (en) * | 2022-07-22 | 2022-09-20 | 山东易斯特工程工具有限公司 | Shield tunneling machine tool bit production defect detection method |
| CN116342611A (en) * | 2023-05-31 | 2023-06-27 | 无锡日联科技股份有限公司 | Method and device for detecting bonding wires of micro semiconductor |
| CN116342611B (en) * | 2023-05-31 | 2023-09-08 | 无锡日联科技股份有限公司 | Method and device for detecting bonding wires of micro semiconductor |
| CN116559298A (en) * | 2023-07-12 | 2023-08-08 | 苏州精控能源科技有限公司 | Battery welding quality detection method and device |
| CN116559298B (en) * | 2023-07-12 | 2023-09-15 | 苏州精控能源科技有限公司 | Battery welding quality detection method and device |
| CN117079100A (en) * | 2023-08-14 | 2023-11-17 | 北京大学 | Weld defect recognition system based on deep learning |
| CN117079100B (en) * | 2023-08-14 | 2024-02-09 | 北京大学 | Weld defect recognition system based on deep learning |
| CN117132406A (en) * | 2023-10-27 | 2023-11-28 | 深圳市泰铼科技有限公司 | Financial product assessment method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111724369A (en) | U-shaped welding seam detection method | |
| CN110047073B (en) | X-ray weld image defect grading method and system | |
| KR20230004570A (en) | On-site inspection method based on digital data model of welding | |
| EP2137519B1 (en) | Method and apparatus for detecting defects in optical components | |
| Ye et al. | Weld bead recognition using laser vision with model-based classification | |
| CN111523611A (en) | Gluing detection method | |
| CN114943739B (en) | Aluminum pipe quality detection method | |
| Mery et al. | A review of methods for automated recognition of casting defects | |
| US8983232B2 (en) | Method for evaluating the centerline of an arbitrarily shaped object | |
| CN110503638B (en) | Spiral adhesive quality online detection method | |
| CN114626470B (en) | Aircraft skin key feature detection method based on multi-type geometric feature operator | |
| CN115719332A (en) | Welding quality detection method | |
| CN114897908B (en) | Machine vision-based method and system for analyzing defects of selective laser powder spreading sintering surface | |
| CN112950633A (en) | Aluminum alloy weld surface defect detection method based on line structured light | |
| CN113516651A (en) | Welding joint defect detection method and device based on residual error network | |
| CN114240944A (en) | Welding defect detection method based on point cloud information | |
| Yang et al. | Detection of weld groove edge based on multilayer convolution neural network | |
| CN115656182A (en) | Sheet material point cloud defect detection method based on tensor voting principal component analysis | |
| CN110414388B (en) | Hump and fusion penetration online early warning method based on depth prediction network | |
| CN117274167A (en) | Three-dimensional information-based method for extracting appearance defects of formed weld joints | |
| CN101629812B (en) | Measuring method of welding joint shaping organizational characteristics | |
| Cai et al. | Weld beads and defects automatic identification, localization, and size calculation based on a lightweight fully convolutional neural network | |
| CN117260046A (en) | Seamless welding system and method for doors and windows | |
| CN116579955A (en) | New energy battery cell weld reflection point denoising and point cloud complement method and system | |
| Yao et al. | Robust locally weighted regression for profile measurement of magnesium alloy tube in hot bending process |
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 |